Copyright © 2005-2015 Jean-Francois Dockes
Permission is granted to copy,
distribute and/or modify this document under the terms
of the GNU Free Documentation License, Version 1.3 or
any later version published by the Free Software
Foundation; with no Invariant Sections, no Front-Cover
Texts, and no Back-Cover Texts. A copy of the license
can be found at the following location: GNU web site.
This document introduces full text search notions and describes the installation and use of the Recoll application. This version describes Recoll 1.22.
Table of Contents
This document introduces full text search notions and describes the installation and use of the Recoll application. This version describes Recoll 1.22.
Recoll was for a long
time dedicated to Unix-like systems. It was only lately
(2015) ported to MS-Windows.
Many references in this manual, especially file locations,
are specific to Unix, and not valid on Windows. Some described features are
also not available on Windows. The manual will be
progressively updated. Until this happens, most references to
shared files can be translated by looking under the Recoll
installation directory (esp. the
Share subdirectory). The user configuration
is stored by default under
AppData/Local/Recoll inside the user
directory, along with the index itself.
If you do not like reading manuals (who does?) but wish to give Recoll a try, just install the application and start the recoll graphical user interface (GUI), which will ask permission to index your home directory by default, allowing you to search immediately after indexing completes.
Do not do this if your home directory contains a huge number of documents and you do not want to wait or are very short on disk space. In this case, you may first want to customize the configuration to restrict the indexed area (for the very impatient with a completed package install, from the recoll GUI: → , then adjust the Top directories section).
Also be aware that, on Unix/Linux, you may need to install the appropriate supporting applications for document types that need them (for example antiword for Microsoft Word files).
The Recoll installation for Windows is self-contained and includes most useful auxiliary programs. You will just need to install Python 2.7.
Recoll is a full text search application, which means that it finds your data by content rather than by external attributes (like the file name). You specify words (terms) which should or should not appear in the text you are looking for, and receive in return a list of matching documents, ordered so that the most relevant documents will appear first.
You do not need to remember in what file or email message you stored a given piece of information. You just ask for related terms, and the tool will return a list of documents where these terms are prominent, in a similar way to Internet search engines.
Full text search applications try to determine which documents are most relevant to the search terms you provide. Computer algorithms for determining relevance can be very complex, and in general are inferior to the power of the human mind to rapidly determine relevance. The quality of relevance guessing is probably the most important aspect when evaluating a search application.
In many cases, you are looking for all the forms of a
word, including plurals, different tenses for a verb, or
terms derived from the same root or stem (example:
floor, floors, floored,
flooring...). Queries are usually automatically
expanded to all such related terms (words that reduce to
the same stem). This can be prevented for searching for a
Stemming, by itself, does not accommodate for
misspellings or phonetic searches. A full text search
application may also support this form of approximation.
For example, a search for
aliterattion returning no
result may propose, depending on index contents,
alliteration alteration alterations
altercation as possible replacement terms.
The Xapian library manages an index database which describes where terms appear in your document files. It efficiently processes the complex queries which are produced by the Recoll query expansion mechanism, and is in charge of the all-important relevance computation task.
Recoll provides the mechanisms and interface to get data into and out of the index. This includes translating the many possible document formats into pure text, handling term variations (using Xapian stemmers), and spelling approximations (using the aspell speller), interpreting user queries and presenting results.
In a shorter way, Recoll does the dirty footwork, Xapian deals with the intelligent parts of the process.
The Xapian index can be big (roughly the size of the original document set), but it is not a document archive. Recoll can only display documents that still exist at the place from which they were indexed. (Actually, there is a way to reconstruct a document from the information in the index, but the result is not nice, as all formatting, punctuation and capitalization are lost).
Recoll stores all internal data in Unicode UTF-8 format, and it can index files of many types with different character sets, encodings, and languages into the same index. It can process documents embedded inside other documents (for example a pdf document stored inside a Zip archive sent as an email attachment...), down to an arbitrary depth.
Stemming is the process by which Recoll reduces words to their radicals so that searching does not depend, for example, on a word being singular or plural (floor, floors), or on a verb tense (flooring, floored). Because the mechanisms used for stemming depend on the specific grammatical rules for each language, there is a separate Xapian stemmer module for most common languages where stemming makes sense.
Recoll stores the unstemmed versions of terms in the main index and uses auxiliary databases for term expansion (one for each stemming language), which means that you can switch stemming languages between searches, or add a language without needing a full reindex.
Storing documents written in different languages in the same index is possible, and commonly done. In this situation, you can specify several stemming languages for the index.
Recoll currently makes no attempt at automatic language recognition, which means that the stemmer will sometimes be applied to terms from other languages with potentially strange results. In practise, even if this introduces possibilities of confusion, this approach has been proven quite useful, and it is much less cumbersome than separating your documents according to what language they are written in.
By default, Recoll
strips most accents and diacritics from terms, and converts
them to lower case before either storing them in the index
or searching for them. As a consequence, it is impossible
to search for a particular capitalization of a term
us), or to discriminate two terms based on
Recoll versions 1.18 and newer can optionally store the raw terms, without accent stripping or case conversion. In this configuration, default searches will behave as before, but it is possible to perform searches sensitive to case and diacritics. This is described in more detail in the section about index case and diacritics sensitivity.
Recoll has many
parameters which define exactly what to index, and how to
classify and decode the source documents. These are kept in
configuration files. A
default configuration is copied into a standard location
(usually something like
installation. The default values set by the configuration
files in this directory may be overridden by values set
inside your personal configuration, found by default in the
.recoll sub-directory of your
home directory. The default configuration will index your
home directory with default parameters and should be
sufficient for giving Recoll a try, but you may want to
adjust it later, which can be done either by editing the
text files or by using configuration menus in the
Some other parameters affecting only the recoll GUI are stored in
the standard location defined by Qt.
The indexing process is started automatically the first time you execute the recoll GUI. Indexing can also be performed by executing the recollindex command. Recoll indexing is multithreaded by default when appropriate hardware resources are available, and can perform in parallel multiple tasks among text extraction, segmentation and index updates.
Searches are usually performed inside the recoll GUI, which has many options to help you find what you are looking for. However, there are other ways to perform Recoll searches: mostly a command line interface, a Python programming interface, a KDE KIO slave module, and Ubuntu Unity Lens (for older versions) or Scope (for current versions) modules.
Indexing is the process by which the set of documents is
analyzed and the data entered into the database.
Recoll indexing is
normally incremental: documents will only be processed if
they have been modified since the last run. On the first
execution, all documents will need processing. A full index
build can be forced later by specifying an option to the
indexing command (recollindex
recollindex skips files
which caused an error during a previous pass. This is a
performance optimization, and a new behaviour in version
1.21 (failed files were always retried by previous
versions). The command line option
-k can be set to retry failed files, for
example after updating a filter.
The following sections give an overview of different aspects of the indexing processes and configuration, with links to detailed sections.
Depending on your data, temporary files may be needed
during indexing, some of them possibly quite big. You can
TMPDIR environment variables to
determine where they are created (the default is to use
TMPDIR has the nice property that it may
also be taken into account by auxiliary commands executed
Recoll indexing can be performed along two different modes:
Real time indexing: indexing takes place as soon as a file is created or changed. recollindex runs as a daemon and uses a file system alteration monitor such as inotify, Fam or Gamin to detect file changes.
The choice between the two methods is mostly a matter of preference, and they can be combined by setting up multiple indexes (ie: use periodic indexing on a big documentation directory, and real time indexing on a small home directory). Monitoring a big file system tree can consume significant system resources.
The choice of method and the parameters used can be configured from the recoll GUI: →
The recollindex process, which will checkpoint its state and exit. A later restart of indexing will mostly resume from where things stopped (the file tree walk has to be restarted from the beginning).menu also has entries to start or stop the current indexing operation. Stopping indexing is performed by killing the
When the real time indexer is running, only a stop operation is available from the menu. When no indexing is running, you have a choice of updating the index or rebuilding it (the first choice only processes changed files, the second one zeroes the index before starting so that all files are processed).
The parameters describing what is to be indexed and local preferences are defined in text files contained in a configuration directory.
All parameters have defaults, defined in system-wide files.
Without further configuration, Recoll will index all appropriate files from your home directory, with a reasonable set of defaults.
A default personal configuration directory
$HOME/.recoll/) is created
when a Recoll program is
first executed. It is possible to create other
configuration directories, and use them by setting the
variable, or giving the
option to any of the Recoll commands.
In some cases, it may be interesting to index different areas of the file system to separate databases. You can do this by using multiple configuration directories, each indexing a file system area to a specific database. Typically, this would be done to separate personal and shared indexes, or to take advantage of the organization of your data to improve search precision.
The generated indexes can be queried concurrently in a transparent manner.
For index generation, multiple configurations are totally independant from each other. When multiple indexes need to be used for a single search, some parameters should be consistent among the configurations.
Recoll knows about quite a few different document types. The parameters for document types recognition and processing are set in configuration files.
Most file types, like HTML or word processing files, only hold one document. Some file types, like email folders or zip archives, can hold many individually indexed documents, which may themselves be compound ones. Such hierarchies can go quite deep, and Recoll can process, for example, a LibreOffice document stored as an attachment to an email message inside an email folder archived in a zip file...
Recoll indexing processes plain text, HTML, OpenDocument (Open/LibreOffice), email formats, and a few others internally.
Other file types (ie: postscript, pdf, ms-word, rtf
...) need external applications for preprocessing. The
list is in the installation section.
After every indexing operation, Recoll updates a list of commands
that would be needed for indexing existing files types.
This list can be displayed by selecting the menu option
→ in the
It is stored in the
text file inside the configuration directory.
By default, Recoll will try to index any file type that it has a way to read. This is sometimes not desirable, and there are ways to either exclude some types, or on the contrary to define a positive list of types to be indexed. In the latter case, any type not in the list will be ignored.
Excluding types can be done by adding wildcard name
patterns to the
list, which can be done from the GUI Index configuration
menu. For versions 1.20 and later, you can alternatively
list in the configuration file. This can be redefined for
You can also define an exclusive list of MIME types to
be indexed (no others will be indexed), by settting the
configuration variable. Example:
indexedmimetypes = text/html application/pdf
It is possible to redefine this parameter for subdirectories. Example:
[/path/to/my/dir] indexedmimetypes = application/pdf
(When using sections like this, don't forget that they remain in effect until the end of the file or another section indicator).
indexedmimetypes, can be set
either by editing the
main configuration file (
recoll.conf), or from the GUI index
Indexing may fail for some documents, for a number of reasons: a helper program may be missing, the document may be corrupt, we may fail to uncompress a file because no file system space is available, etc.
Recoll versions prior to 1.21 always retried to index files which had previously caused an error. This guaranteed that anything that may have become indexable (for example because a helper had been installed) would be indexed. However this was bad for performance because some indexing failures may be quite costly (for example failing to uncompress a big file because of insufficient disk space).
The indexer in Recoll
versions 1.21 and later does not retry failed file by
default. Retrying will only occur if an explicit option
-k) is set on the
command line, or if a script executed when recollindex starts up
says so. The script is defined by a configuration
checkneedretryindexscript), and makes a
rather lame attempt at deciding if a helper command may
have been installed, by checking if any of the common
bin directories have
In the rare case where the index becomes corrupted
(which can signal itself by weird search results or
crashes), the index files need to be erased before
restarting a clean indexing pass. Just delete the
xapiandb directory (see
next section), or,
alternatively, start the next recollindex with the
-z option, which will reset
the database before indexing.
The default location for the index data is the
xapiandb subdirectory of the
$HOME/.recoll/xapiandb/. This can be
changed via two different methods (with different
You can specify a different configuration
directory by setting the
RECOLL_CONFDIR environment variable,
or using the
-c option to
the Recoll commands.
This method would typically be used to index
different areas of the file system to different
indexes. For example, if you were to issue the
recoll -c ~/.indexes-email
Then Recoll would
use configuration files stored in
~/.indexes-email/ and, (unless
specified otherwise in
recoll.conf) would look for the
Using multiple configuration directories and configuration options allows you to tailor multiple configurations and indexes to handle whatever subset of the available data you wish to make searchable.
For a given configuration directory, you can
specify a non-default storage location for the index
by setting the
parameter in the configuration file (see the
configuration section). This method would mainly
be of use if you wanted to keep the configuration
directory in its default location, but desired
another location for the index, typically out of disk
The size of the index is determined by the size of the set of documents, but the ratio can vary a lot. For a typical mixed set of documents, the index size will often be close to the data set size. In specific cases (a set of compressed mbox files for example), the index can become much bigger than the documents. It may also be much smaller if the documents contain a lot of images or other non-indexed data (an extreme example being a set of mp3 files where only the tags would be indexed).
Of course, images, sound and video do not increase the index size, which means that nowadays (2012), typically, even a big index will be negligible against the total amount of data on the computer.
The index data directory (
xapiandb) only contains data that can be
completely rebuilt by an index run (as long as the original
documents exist), and it can always be destroyed
Xapian versions usually support several formats for index storage. A given major Xapian version will have a current format, used to create new indexes, and will also support the format from the previous major version.
Xapian will not convert automatically an existing index from the older format to the newer one. If you want to upgrade to the new format, or if a very old index needs to be converted because its format is not supported any more, you will have to explicitly delete the old index, then run a normal indexing process.
-z option to
is not sufficient to change the format, you will have to
delete all files inside the index directory (typically
starting the indexing.
The Recoll index does not hold copies of the indexed documents. But it does hold enough data to allow for an almost complete reconstruction. If confidential data is indexed, access to the database directory should be restricted.
Recoll will create the configuration directory with a mode of 0700 (access by owner only). As the index data directory is by default a sub-directory of the configuration directory, this should result in appropriate protection.
If you use another setup, you should think of the kind
of protection you need for your index, set the directory
and files access modes appropriately, and also maybe
umask used during
Variables set inside the Recoll configuration files control which areas of the file system are indexed, and how files are processed. These variables can be set either by editing the text files or by using the dialogs in the recoll GUI.
The first time you start recoll, you will be asked
whether or not you would like it to build the index. If you
want to adjust the configuration before indexing, just
click Cancel at this point,
which will get you into the configuration interface. If you
exit at this point,
will have created a
directory containing empty configuration files, which you
can edit by hand.
The configuration is documented inside the installation chapter
of this document, or in the recoll.conf(5) man page, but
the most current information will most likely be the
comments inside the sample file. The most immediately
useful variable you may interested in is probably
topdirs, which determines what
subtrees get indexed.
The applications needed to index file types other than text, HTML or email (ie: pdf, postscript, ms-word...) are described in the external packages section.
As of Recoll 1.18 there are two incompatible types of Recoll indexes, depending on the treatment of character case and diacritics. The next section describes the two types in more detail.
indexes can be created by using several configuration
directories which are usually set to index different
areas of the file system. A specific index can be
selected for updating or searching, using the
variable or the
-c option to
When working with the recoll index
configuration GUI, the configuration directory for which
parameters are modified is the one which was selected by
RECOLL_CONFDIR or the
-c parameter, and there is no
way to switch configurations within the GUI.
Additional configuration directory (beyond
~/.recoll) must be created
by hand (mkdir or such), the GUI
will not do it. This is to avoid mistakenly creating
additional directories when an argument is mistyped.
A typical usage scenario for the multiple index feature would be for a system administrator to set up a central index for shared data, that you choose to search or not in addition to your personal data. Of course, there are other possibilities. There are many cases where you know the subset of files that should be searched, and where narrowing the search can improve the results. You can achieve approximately the same effect with the directory filter in advanced search, but multiple indexes will have much better performance and may be worth the trouble.
A recollindex program instance can only update one specific index.
The main index (defined by
-c) is always active. If this is
undesirable, you can set up your base configuration to
index an empty directory.
The different search interfaces (GUI, command line, ...) have different methods to define the set of indexes to be used, see the appropriate section.
If a set of multiple indexes are to be used together for searches, some configuration parameters must be consistent among the set. These are parameters which need to be the same when indexing and searching. As the parameters come from the main configuration when searching, they need to be compatible with what was set when creating the other indexes (which came from their respective configuration directories).
Most importantly, all indexes to be queried concurrently must have the same option concerning character case and diacritics stripping, but there are other constraints. Most of the relevant parameters are described in the linked section.
As of Recoll version
1.18 you have a choice of building an index with terms
stripped of character case and diacritics, or one with
raw terms. For a source term of
Résumé, the former will
resume, the latter
Each type of index allows performing searches insensitive to case and diacritics: with a raw index, the user entry will be expanded to match all case and diacritics variations present in the index. With a stripped index, the search term will be stripped before searching.
A raw index allows for another possibility which a
stripped index cannot offer: using case and diacritics to
discriminate between terms, returning different results
when searching for
résumé. Read the section
about search case and diacritics sensitivity for more
The type of index to be created is controlled by the
configuration variable which can only be changed by
editing the configuration file. Any change implies an
index reset (not automated by Recoll), and all indexes in a search
must be set in the same way (again, not checked by
not set, Recoll 1.18
creates a stripped index by default, for compatibility
with previous versions.
As a cost for added capability, a raw index will be slightly bigger than a stripped one (around 10%). Also, searches will be more complex, so probably slightly slower, and the feature is still young, so that a certain amount of weirdness cannot be excluded.
One of the most adverse consequence of using a raw index is that some phrase and proximity searches may become impossible: because each term needs to be expanded, and all combinations searched for, the multiplicative expansion may become unmanageable.
The Recoll indexing process recollindex can use multiple threads to speed up indexing on multiprocessor systems. The work done to index files is divided in several stages and some of the stages can be executed by multiple threads. The stages are:
You can also read a longer document about the transformation of Recoll indexing to multithreading.
The threads configuration is controlled by two configuration file parameters.
This variable defines the job input queues configuration. There are three possible queues for stages 2, 3 and 4, and this parameter should give the queue depth for each stage (three integer values). If a value of -1 is used for a given stage, no queue is used, and the thread will go on performing the next stage. In practise, deep queues have not been shown to increase performance. A value of 0 for the first queue tells Recoll to perform autoconfiguration (no need for anything else in this case, thrTCounts is not used) - this is the default configuration.
This defines the number of threads used for each stage. If a value of -1 is used for one of the queue depths, the corresponding thread count is ignored. It makes no sense to use a value other than 1 for the last stage because updating the Xapian index is necessarily single-threaded (and protected by a mutex).
The following example would use three queues (of depth 2), and 4 threads for converting source documents, 2 for processing their text, and one to update the index. This was tested to be the best configuration on the test system (quadri-processor with multiple disks).
thrQSizes = 2 2 2 thrTCounts = 4 2 1
The following example would use a single queue, and the complete processing for each document would be performed by a single thread (several documents will still be processed in parallel in most cases). The threads will use mutual exclusion when entering the index update stage. In practise the performance would be close to the precedent case in general, but worse in certain cases (e.g. a Zip archive would be performed purely sequentially), so the previous approach is preferred. YMMV... The 2 last values for thrTCounts are ignored.
thrQSizes = 2 -1 -1 thrTCounts = 6 1 1
The following example would disable multithreading. Indexing will be performed by a single thread.
thrQSizes = -1 -1 -1
Most parameters for a given index configuration can be
set from a recoll GUI running on
this configuration (either as default, or by setting
RECOLL_CONFDIR or the
The interface is started from the Global parameters, Local parameters, Web history (which is explained in the next section) and Search parameters.→ menu entry. It is divided in four tabs,
The Global parameters tab allows setting global variables, like the lists of top directories, skipped paths, or stemming languages.
The Local parameters tab allows setting variables that can be redefined for subdirectories. This second tab has an initially empty list of customisation directories, to which you can add. The variables are then set for the currently selected directory (or at the top level if the empty line is selected).
The Search parameters section defines parameters which are used at query time, but are global to an index and affect all search tools, not only the GUI.
The meaning for most entries in the interface is
self-evident and documented by a
ToolTip popup on the text label. For
more detail, you will need to refer to the configuration
section of this guide.
The configuration tool normally respects the comments and most of the formatting inside the configuration file, so that it is quite possible to use it on hand-edited files, which you might nevertheless want to backup first...
With the help of a Firefox extension, Recoll can index the Internet pages that you visit. The extension was initially designed for the Beagle indexer, but it has recently be renamed and better adapted to Recoll.
The extension works by copying visited WEB pages to an indexing queue directory, which Recoll then processes, indexing the data, storing it into a local cache, then removing the file from the queue.
This feature can be enabled in the GUI Index configuration panel, or by editing
the configuration file (set
processwebqueue to 1).
A current pointer to the extension can be found, along with up-to-date instructions, on the Recoll wiki.
A copy of the indexed WEB pages is retained by Recoll in a local cache (from which previews can be fetched). The cache size can be adjusted from the Index configuration / Web history panel. Once the maximum size is reached, old pages are purged - both from the cache and the index - to make room for new ones, so you need to explicitly archive in some other place the pages that you want to keep indefinitely.
User extended attributes are named pieces of information that most modern file systems can attach to any file.
Recoll versions 1.19 and later process extended attributes as document fields by default. For older versions, this has to be activated at build time.
A freedesktop standard defines a few special attributes, which are handled as such by Recoll:
If set, this overrides any other determination of the file MIME type.
By default, other attributes are handled as Recoll fields. On Linux, the
user prefix is removed from
the name. This can be configured more precisely inside the
fields configuration file.
During indexing, it is possible to import metadata for each file by executing commands. For example, this could extract user tag data for the file and store it in a field for indexing.
See the section about
metadatacmds field in
the main configuration chapter for a description of the
As an example, if you would want Recoll to use tags managed by tmsu, you would add the following to the configuration file:
[/some/area/of/the/fs] metadatacmds = ; tags = tmsu tags %f
Depending on the tmsu
version, you may need/want to add options like
You may want to restrict this processing to a subset of
the directory tree, because it may slow down indexing a bit
Note the initial semi-colon after the equal sign.
In the example above, the output of tmsu is used to set a
tags. The field
name is arbitrary and could be
myfield just the same, but
tags is an alias for the standard
keywords field, and the tmsu output will just
augment its contents. This will avoid the need to extend
Once re-indexing is performed (you'll need to force the
file reindexing, Recoll
will not detect the need by itself), you will be able to
search from the query language, through any of its aliases:
compact field search syntax is supported for recoll 1.20
and later. For older versions, you would need to repeat the
tags: specifier for each term,
You should be aware that tags changes will not be
detected by the indexer if the file itself did not change.
One possible workaround would be to update the file
ctime when you modify the
tags, which would be consistent with how extended
attributes function. A pair of chmod commands could
accomplish this, or a
. Alternatively, just couple the tag update with a
recollindex -e -i
Indexing is always performed by the recollindex program,
which can be started either from the command line or from
the menu in the
program. When started from the GUI, the indexing will run
on the same configuration recoll was started on.
When started from the command line, recollindex will use
RECOLL_CONFDIR variable or
confdir option to specify
a non-default configuration directory.
If the recoll program finds no index when it starts, it will automatically start indexing (except if canceled).
The recollindex indexing process can be interrupted by sending an interrupt (Ctrl-C, SIGINT) or terminate (SIGTERM) signal. Some time may elapse before the process exits, because it needs to properly flush and close the index. This can also be done from the recoll GUI → menu entry.
After such an interruption, the index will be somewhat inconsistent because some operations which are normally performed at the end of the indexing pass will have been skipped (for example, the stemming and spelling databases will be inexistant or out of date). You just need to restart indexing at a later time to restore consistency. The indexing will restart at the interruption point (the full file tree will be traversed, but files that were indexed up to the interruption and for which the index is still up to date will not need to be reindexed).
recollindex has a number of other options which are described in its man page. Only a few will be described here.
-z will reset the
index when starting. This is almost the same as
destroying the index files (the nuance is that the
Xapian format version
will not be changed).
-Z will force the
update of all documents without resetting the index
first. This will not have the "clean start" aspect of
-z, but the advantage is that
the index will remain available for querying while it is
rebuilt, which can be a significant advantage if it is
very big (some installations need days for a full index
-k will force
retrying files which previously failed to be indexed, for
example because of a missing helper program.
Of special interest also, maybe, are the
-i allows indexing
an explicit list of files (given as command line
parameters or read on
tells recollindex to ignore
file selection parameters from the configuration.
Together, these options allow building a custom file
selection process for some area of the file system, by
adding the top directory to the
skippedPaths list and using an
appropriate file selection method to build the file list
to be fed to recollindex
-if. Trivial example:
find . -name indexable.txt -print | recollindex -if
-i will not descend into
subdirectories specified as parameters, but just add them
as index entries. It is up to the external file selection
method to build the complete file list.
The most common way to set up indexing is to have a
cron task execute it every night. For example the
would do it every day at 3:30AM (supposing recollindex is in your
30 3 * * * recollindex > /some/tmp/dir/recolltrace 2>&1
Or, using anacron:
1 15 su mylogin -c "recollindex recollindex > /tmp/rcltraceme 2>&1"
As of version 1.17 the Recoll GUI has dialogs to manage
crontab entries for
recollindex. You can
reach them from the → menu. They only
work with the good old cron, and do not give
access to all features of cron scheduling.
The usual command to edit your
crontab is crontab
-e (which will usually start the
to edit the file). You may have more sophisticated tools
available on your system.
Please be aware that there may be differences between your usual interactive command line environment and the one seen by crontab commands. Especially the PATH variable may be of concern. Please check the crontab manual pages about possible issues.
Real time monitoring/indexing is performed by starting
-m command. With this option, recollindex will detach
from the terminal and become a daemon, permanently
monitoring file changes and updating the index.
Under KDE, Gnome and some other desktop
environments, the daemon can automatically started when you
log in, by creating a desktop file inside the
~/.config/autostart directory. This can
be done for you by the Recoll GUI. Use the
With older X11 setups, starting the daemon is normally performed as part of the user session script.
rclmon.sh script can
be used to easily start and stop the daemon. It can be
found in the
For example, my out of fashion xdm-based session has a
.xsession script with the following lines
at the end:
recollconf=$HOME/.recoll-home recolldata=/usr/local/share/recoll RECOLL_CONFDIR=$recollconf $recolldata/examples/rclmon.sh start fvwm
The indexing daemon gets started, then the window manager, for which the session waits.
By default the indexing daemon will monitor the state of
the X11 session, and exit when it finishes, it is not
necessary to kill it explicitly. (The X11 server monitoring can be disabled
-x to recollindex).
If you use the daemon completely out of an X11 session, you need to add option
-x to disable X11 session monitoring (else the
daemon will not start).
By default, the messages from the indexing daemon will
be setn to the same file as those from the interactive
may want to change this by setting the
daemloglevel configuration parameters.
Also the log file will only be truncated when the daemon
starts. If the daemon runs permanently, the log file may
grow quite big, depending on the log level.
When building Recoll,
the real time indexing support can be customised during
package configuration with the
--with[out]-inotify options. The default is
currently to include inotify monitoring on systems that
support it, and, as of Recoll 1.17, gamin support on FreeBSD.
While it is convenient that data is indexed in real time, repeated indexing can generate a significant load on the system when files such as email folders change. Also, monitoring large file trees by itself significantly taxes system resources. You probably do not want to enable it if your system is short on resources. Periodic indexing is adequate in most cases.
On Linux systems, monitoring a big tree may need
increasing the resources available to inotify, which are
normally defined in
### inotify # # cat /proc/sys/fs/inotify/max_queued_events - 16384 # cat /proc/sys/fs/inotify/max_user_instances - 128 # cat /proc/sys/fs/inotify/max_user_watches - 16384 # # -- Change to: # fs.inotify.max_queued_events=32768 fs.inotify.max_user_instances=256 fs.inotify.max_user_watches=32768
Especially, you will need to trim your tree or adjust
if indexing exits with a message about errno
ENOSPC (28) from
When using the real time monitor, it may happen that some files need to be indexed, but change so often that they impose an excessive load for the system.
Recoll provides a
configuration option to specify the minimum time before
which a file, specified by a wildcard pattern, cannot be
reindexed. See the
mondelaypatterns parameter in the
The recoll program provides the main user interface for searching. It is based on the Qt library.
recoll has two search modes:
Simple search (the default, on the main screen) has a single entry field where you can enter multiple words.
Advanced search (a panel accessed through the Tools menu or the toolbox bar icon) has multiple entry fields, which you may use to build a logical condition, with additional filtering on file type, location in the file system, modification date, and size.
In most cases, you can enter the terms as you think them, even if they contain embedded punctuation or other non-textual characters. For example, Recoll can handle things like email addresses, or arbitrary cut and paste from another text window, punctation and all.
The main case where you should enter text differently from how it is printed is for east-asian languages (Chinese, Japanese, Korean). Words composed of single or multiple characters should be entered separated by white space in this case (they would typically be printed without white space).
Some searches can be quite complex, and you may want to re-use them later, perhaps with some tweaking. Recoll versions 1.21 and later can save and restore searches, using XML files. See Saving and restoring queries.
Start the recoll program.
Possibly choose a search mode: Any term, All terms, File name or Query language.
Enter search term(s) in the text field at the top of the window.
Click the Search button or hit the Enter key to start the search.
The initial default search mode is Query language. Without special directives, this will look for documents containing all of the search terms (the ones with more terms will get better scores), just like the All terms mode which will ignore such directives. Any term will search for documents where at least one of the terms appear.
The Query Language features are described in a separate section.
All search modes allow wildcards inside terms
may want to have a look at the section about
wildcards for more information about this.
File name will specifically look for file names. The point of having a separate file name search is that wild card expansion can be performed more efficiently on a small subset of the index (allowing wild cards on the left of terms without excessive penality). Things to know:
White space in the entry should match white space in the file name, and is not treated specially.
The search is insensitive to character case and accents, independantly of the type of index.
An entry without any wild card character and not
capitalized will be prepended and appended with '*'
If you have a big index (many files), excessively generic fragments may result in inefficient searches.
You can search for exact phrases (adjacent words in a
given order) by enclosing the input inside double quotes.
When using a stripped index, character case has no
influence on search, except that you can disable stem
expansion for any term by capitalizing it. Ie: a search
floor will also normally
floored, etc., but a search
Floor will only look for
floor, in any character
case. Stemming can also be disabled globally in the
preferences. When using a raw index, the
rules are a bit more complicated.
Recoll remembers the last few searches that you performed. You can use the simple search text entry widget (a combobox) to recall them (click on the thing at the right of the text field). Please note, however, that only the search texts are remembered, not the mode (all/any/file name).
Typing Esc Space while entering a word in the simple search entry will open a window with possible completions for the word. The completions are extracted from the database.
Double-clicking on a word in the result list or a preview window will insert it into the simple search entry field.
You can cut and paste any text into an All terms or Any
term search field, punctuation, newlines and all -
except for wildcard characters (single
? characters are ok). Recoll will process it and produce a
meaningful search. This is what most differentiates this
mode from the Query
Language mode, where you have to care about the
After starting a search, a list of results will instantly be displayed in the main list window.
By default, the document list is presented in order of relevance (how well the system estimates that the document matches the query). You can sort the result by ascending or descending date by using the vertical arrows in the toolbar.
Clicking on the
link for an entry will open an internal preview window
for the document. Further
Preview clicks for the same search will
open tabs in the existing preview window. You can use
to force the creation of another preview window, which
may be useful to view the documents side by side. (You
can also browse successive results in a single preview
window by typing Shift+ArrowUp/Down in the
will start an external viewer for the document. By
default, Recoll lets the
desktop choose the appropriate application for most
document types (there is a short list of exceptions, see
further). If you prefer to completely customize the
choice of applications, you can uncheck the Use desktop preferences option in the
GUI preferences dialog, and click the Choose editor applications button to
adjust the predefined Recoll choices. The tool accepts
multiple selections of MIME types (e.g. to set up the
editor for the dozens of office file types).
Even when Use desktop preferences is checked, there is a small list of exceptions, for MIME types where the Recoll choice should override the desktop one. These are applications which are well integrated with Recoll, especially evince for viewing PDF and Postscript files because of its support for opening the document at a specific page and passing a search string as an argument. Of course, you can edit the list (in the GUI preferences) if you would prefer to lose the functionality and use the standard desktop tool.
You may also change the choice of applications by
mimeview configuration file if you
find this more convenient.
Each result entry also has a right-click menu with an Open With entry. This lets you choose an application from the list of those which registered with the desktop for the document MIME type.
Open edit links may not be
present for all entries, meaning that Recoll has no configured way to
preview a given file type (which was indexed by name
only), or no configured external editor for the file
type. This can sometimes be adjusted simply by tweaking
mimeview configuration files (the
latter can be modified with the user preferences
The format of the result list entries is entirely configurable by using the preference dialog to edit an HTML fragment.
You can click on the
details link at the top of the results page to see
the query actually performed, after stem expansion and
Double-clicking on any word inside the result list or a preview window will insert it into the simple search text.
The result list is divided into pages (the size of which you can change in the preferences). Use the arrow buttons in the toolbar or the links at the bottom of the page to browse the results.
When a search yields no result, and if the aspell dictionary is configured, Recoll will try to check for misspellings among the query terms, and will propose lists of replacements. Clicking on one of the suggestions will replace the word and restart the search. You can hold any of the modifier keys (Ctrl, Shift, etc.) while clicking if you would rather stay on the suggestion screen because several terms need replacement.
Apart from the preview and edit links, you can display a pop-up menu by right-clicking over a paragraph in the result list. This menu has the following entries:
Copy File Name
Save to File
Preview Parent document
Open Parent document
Open Snippets Window
The Preview and Open entries do the same thing as the corresponding links.
Open With lets you
open the document with one of the applications claiming
to be able to handle its MIME type (the information
comes from the
Run Script allows starting an arbitrary command on the result file. It will only appear for results which are top-level files. See further for a more detailed description.
The Copy File Name and Copy Url copy the relevant data to the clipboard, for later pasting.
Save to File allows saving the contents of a result document to a chosen file. This entry will only appear if the document does not correspond to an existing file, but is a subdocument inside such a file (ie: an email attachment). It is especially useful to extract attachments with no associated editor.
The Open/Preview Parent document entries allow working with the higher level document (e.g. the email message an attachment comes from). Recoll is sometimes not totally accurate as to what it can or can't do in this area. For example the Parent entry will also appear for an email which is part of an mbox folder file, but you can't actually visualize the mbox (there will be an error dialog if you try).
If the document is a top-level file, Open Parent will start the default file manager on the enclosing filesystem directory.
The Find similar entry will select a number of relevant term from the current document and enter them into the simple search field. You can then start a simple search, with a good chance of finding documents related to the current result. I can't remember a single instance where this function was actually useful to me...
The Open Snippets Window entry will only appear for documents which support page breaks (typically PDF, Postscript, DVI). The snippets window lists extracts from the document, taken around search terms occurrences, along with the corresponding page number, as links which can be used to start the native viewer on the appropriate page. If the viewer supports it, its search function will also be primed with one of the search terms.
In Recoll 1.15 and newer, the results can be displayed in spreadsheet-like fashion. You can switch to this presentation by clicking the table-like icon in the toolbar (this is a toggle, click again to restore the list).
Clicking on the column headers will allow sorting by the values in the column. You can click again to invert the order, and use the header right-click menu to reset sorting to the default relevance order (you can also use the sort-by-date arrows to do this).
Both the list and the table display the same underlying results. The sort order set from the table is still active if you switch back to the list mode. You can click twice on a date sort arrow to reset it from there.
The header right-click menu allows adding or deleting columns. The columns can be resized, and their order can be changed (by dragging). All the changes are recorded when you quit recoll
Hovering over a table row will update the detail area at the bottom of the window with the corresponding values. You can click the row to freeze the display. The bottom area is equivalent to a result list paragraph, with links for starting a preview or a native application, and an equivalent right-click menu. Typing Esc (the Escape key) will unfreeze the display.
Apart from the Open and Open With operations, which allow starting an application on a result document (or a temporary copy), based on its MIME type, it is also possible to run arbitrary commands on results which are top-level files, using the Run Script entry in the results pop-up menu.
The commands which will appear in the Run Script submenu must be defined by
.desktop files inside the
scripts subdirectory of the
current configuration directory.
Here follows an example of a
.desktop file, which could be named for
exact file name inside the directory is irrelevant):
[Desktop Entry] Type=Application Name=MyFirstScript Exec=/home/me/bin/tryscript %F MimeType=*/*
defines the label which will appear inside the
Run Script menu. The
Exec attribute defines the
program to be run, which does not need to actually be a
script, of course. The
MimeType attribute is not used, but
needs to exist.
The commands defined this way can also be used from links inside the result paragraph.
As an example, it might make sense to write a script which would move the document to the trash and purge it from the Recoll index.
The default format for the result list entries and the detail area of the result table display an icon for each result document. The icon is either a generic one determined from the MIME type, or a thumbnail of the document appearance. Thumbnails are only displayed if found in the standard freedesktop location, where they would typically have been created by a file manager.
Recoll has no capability to create thumbnails. A relatively simple trick is to use the Open parent document/folder entry in the result list popup menu. This should open a file manager window on the containing directory, which should in turn create the thumbnails (depending on your settings). Restarting the search should then display the thumbnails.
There are also some pointers about thumbnail generation on the Recoll wiki.
The preview window opens when you first click a
Preview link inside the
Subsequent preview requests for a given search open new tabs in the existing window (except if you hold the Shift key while clicking which will open a new window for side by side viewing).
Starting another search and requesting a preview will create a new preview window. The old one stays open until you close it.
You can close a preview tab by typing Ctrl-W (Ctrl + W) in the window. Closing the last tab for a window will also close the window.
Of course you can also close a preview window by using the window manager button in the top of the frame.
You can display successive or previous documents from the result list inside a preview tab by typing Shift+Down or Shift+Up (Down and Up are the arrow keys).
A right-click menu in the text area allows switching between displaying the main text or the contents of fields associated to the document (ie: author, abtract, etc.). This is especially useful in cases where the term match did not occur in the main text but in one of the fields. In the case of images, you can switch between three displays: the image itself, the image metadata as extracted by exiftool and the fields, which is the metadata stored in the index.
You can print the current preview window contents by typing Ctrl-P (Ctrl + P) in the window text.
The preview window has an internal search capability, mostly controlled by the panel at the bottom of the window, which works in two modes: as a classical editor incremental search, where we look for the text entered in the entry zone, or as a way to walk the matches between the document and the Recoll query that found it.
The preview tabs have an internal incremental search function. You initiate the search either by typing a / (slash) or CTL-F inside the text area or by clicking into the Search for: text field and entering the search string. You can then use the Next and Previous buttons to find the next/previous occurrence. You can also type F3 inside the text area to get to the next occurrence.
If you have a search string entered and you use Ctrl-Up/Ctrl-Down to browse the results, the search is initiated for each successive document. If the string is found, the cursor will be positioned at the first occurrence of the search string.
If the entry area is empty when you click the Next or Previous buttons, the editor will be scrolled to show the next match to any search term (the next highlighted zone). If you select a search group from the dropdown list and click Next or Previous, the match list for this group will be walked. This is not the same as a text search, because the occurences will include non-exact matches (as caused by stemming or wildcards). The search will revert to the text mode as soon as you edit the entry area.
Selecting the→ menu entry will open a window with radio- and check-buttons which can be used to activate query language fragments for filtering the current query. This can be useful if you have frequent reusable selectors, for example, filtering on alternate directories, or searching just one category of files, not covered by the standard category selectors.
The contents of the window are entirely customizable,
and defined by the contents of the
fragbuts.xml file inside the
configuration directory. The sample file distributed with
Recoll (which you should
be able to find under
contains an example which filters the results from the
Here follows an example:
<?xml version="1.0" encoding="UTF-8"?> <fragbuts version="1.0"> <radiobuttons> <fragbut> <label>Include Web Results</label> <frag></frag> </fragbut> <fragbut> <label>Exclude Web Results</label> <frag>-rclbes:BGL</frag> </fragbut> <fragbut> <label>Only Web Results</label> <frag>rclbes:BGL</frag> </fragbut> </radiobuttons> <buttons> <fragbut> <label>Year 2010</label> <frag>date:2010-01-01/2010-12-31</frag> </fragbut> <fragbut> <label>My Great Directory Only</label> <frag>dir:/my/great/directory</frag> </fragbut> </buttons> </fragbuts>
buttons section defines a
line of checkbuttons or radiobuttons inside the window.
Any number of buttons can be selected, but the
radiobuttons in a line are exclusive.
defines the label for a button, and the Query Language
fragment which will be added (as an AND filter) before
performing the query if the button is active.
This feature is new in Recoll 1.20, and will probably be refined depending on user feedback.
The advanced search dialog helps you build more complex queries without memorizing the search language constructs. It can be opened through the Tools menu or through the main toolbar.
Recoll keeps a history of searches. See Advanced search history.
The dialog has two tabs:
The first tab lets you specify terms to search for, and permits specifying multiple clauses which are combined to build the search.
The second tab lets filter the results according to file size, date of modification, MIME type, or location.
Click on the Start Search button in the advanced search dialog, or type Enter in any text field to start the search. The button in the main window always performs a simple search.
Click on the
details link at the top of the result page to see
the query expansion.
This part of the dialog lets you constructc a query by combining multiple clauses of different types. Each entry field is configurable for the following modes:
None of the terms.
Phrase (exact terms in order within an adjustable window).
Proximity (terms in any order within an adjustable window).
Additional entry fields can be created by clicking the Add clause button.
When searching, the non-empty clauses will be combined either with an AND or an OR conjunction, depending on the choice made on the left (All clauses or Any clause).
Entries of all types except "Phrase" and "Near" accept a mix of single words and phrases enclosed in double quotes. Stemming and wildcard expansion will be performed as for simple search.
Phrases and Proximity searches. These
two clauses work in similar ways, with the difference
that proximity searches do not impose an order on the
words. In both cases, an adjustable number (slack) of
non-matched words may be accepted between the searched
ones (use the counter on the left to adjust this
count). For phrases, the default count is zero (exact
match). For proximity it is ten (meaning that two
search terms, would be matched if found within a window
of twelve words). Examples: a phrase search for
quick fox with a slack of
0 will match
quick fox but
quick brown fox. With
a slack of 1 it will match the latter, but not
fox quick. A proximity
quick fox with
the default slack will match the latter, and also
a fox is a cunning and quick
This part of the dialog has several sections which allow filtering the results of a search according to a number of criteria
The first section allows filtering by dates of last modification. You can specify both a minimum and a maximum date. The initial values are set according to the oldest and newest documents found in the index.
The next section allows filtering the results
by file size. There are two entries for minimum
and maximum size. Enter decimal numbers. You can
use suffix multipliers:
t/T for 1E3, 1E6, 1E9, 1E12
The next section allows filtering the results by their MIME types, or MIME categories (ie: media/text/message/etc.).
You can transfer the types between two boxes, to define which will be included or excluded by the search.
The state of the file type selection can be saved as the default (the file type filter will not be activated at program start-up, but the lists will be in the restored state).
The bottom section allows restricting the search results to a sub-tree of the indexed area. You can use the Invert checkbox to search for files not in the sub-tree instead. If you use directory filtering often and on big subsets of the file system, you may think of setting up multiple indexes instead, as the performance may be better.
You can use relative/partial paths for
filtering. Ie, entering
dirA/dirB would match either
The advanced search tool memorizes the last 100 searches performed. You can walk the saved searches by using the up and down arrow keys while the keyboard focus belongs to the advanced search dialog.
The complex search history can be erased, along with the one for simple search, by selecting the→ menu entry.
Recoll automatically manages the expansion of search terms to their derivatives (ie: plural/singular, verb inflections). But there are other cases where the exact search term is not known. For example, you may not remember the exact spelling, or only know the beginning of the name.
The search will only propose replacement terms with spelling variations when no matching document were found. In some cases, both proper spellings and mispellings are present in the index, and it may be interesting to look for them explicitely.
The term explorer tool (started from the toolbar icon or from the Term explorer entry of the Tools menu) can be used to search the full index terms list. It has three modes of operations:
In this mode of operation, you can enter a
search string with shell-like wildcards (*, ?, ).
would display all index terms beginning with
(More about wildcards here).
This mode will accept a regular expression as
expression is implicitely anchored at the
press will match
expression. You can
.*press to match
the latter, but be aware that this will cause a
full index term list scan, which can be quite
This mode will perform the usual stem expansion normally done as part user input processing. As such it is probably mostly useful to demonstrate the process.
In this mode, you enter the term as you think it is spelled, and Recoll will do its best to find index terms that sound like your entry. This mode uses the Aspell spelling application, which must be installed on your system for things to work (if your documents contain non-ascii characters, Recoll needs an aspell version newer than 0.60 for UTF-8 support). The language which is used to build the dictionary out of the index terms (which is done at the end of an indexing pass) is the one defined by your NLS environment. Weird things will probably happen if languages are mixed up.
Note that in cases where Recoll does not know the beginning
of the string to search for (ie a wildcard expression
the expansion can take quite a long time because the full
index term list will have to be processed. The expansion
is currently limited at 10000 results for wildcards and
regular expressions. It is possible to change the limit
in the configuration file.
Double-clicking on a term in the result list will insert it into the simple search entry field. You can also cut/paste between the result list and any entry field (the end of lines will be taken care of).
See the section describing the use of multiple indexes for generalities. Only the aspects concerning the recoll GUI are described here.
A recoll program instance is always associated with a specific index, which is the one to be updated when requested from the menu, but it can use any number of Recoll indexes for searching. The external indexes can be selected through the external indexes tab in the preferences dialog.
Index selection is performed in two phases. A set of all usable indexes must first be defined, and then the subset of indexes to be used for searching. These parameters are retained across program executions (there are kept separately for each Recoll configuration). The set of all indexes is usually quite stable, while the active ones might typically be adjusted quite frequently.
The main index (defined by
RECOLL_CONFDIR) is always active. If this
is undesirable, you can set up your base configuration to
index an empty directory.
When adding a new index to the set, you can select either a Recoll configuration directory, or directly a Xapian index directory. In the first case, the Xapian index directory will be obtained from the selected configuration.
As building the set of all indexes can be a little
tedious when done through the user interface, you can use
environment variable to provide an initial set. This
might typically be set up by a system administrator so
that every user does not have to do it. The variable
should define a colon-separated list of index
Another environment variable,
RECOLL_ACTIVE_EXTRA_DBS allows adding to
the active list of indexes. This variable was suggested
and implemented by a Recoll user. It is mostly useful if
you use scripts to mount external volumes with
Recoll indexes. By using
can add and activate the index for the mounted volume
when starting recoll.
available for Recoll
versions 1.17.2 and later. A change was made in the same
update so that recoll will
automatically deactivate unreachable indexes when
Documents that you actually view (with the internal preview or an external tool) are entered into the document history, which is remembered.
You can display the history list by using the Tools/Doc History menu entry.
You can erase the document history by using the Erase document history entry in the menu.
The documents in a result list are normally sorted in order of relevance. It is possible to specify a different sort order, either by using the vertical arrows in the GUI toolbox to sort by date, or switching to the result table display and clicking on any header. The sort order chosen inside the result table remains active if you switch back to the result list, until you click one of the vertical arrows, until both are unchecked (you are back to sort by relevance).
Sort parameters are remembered between program invocations, but result sorting is normally always inactive when the program starts. It is possible to keep the sorting activation state between program invocations by checking the Remember sort activation state option in the preferences.
It is also possible to hide duplicate entries inside the result list (documents with the exact same contents as the displayed one). The test of identity is based on an MD5 hash of the document container, not only of the text contents (so that ie, a text document with an image added will not be a duplicate of the text only). Duplicates hiding is controlled by an entry in the GUI configuration dialog, and is off by default.
As of release 1.19, when a result document does have
undisplayed duplicates, a
Dups link will be shown with the result
list entry. Clicking the link will display the paths
(URLs + ipaths) for the duplicate entries.
Term completion. Typing Esc Space in the simple search entry field while entering a word will either complete the current word if its beginning matches a unique term in the index, or open a window to propose a list of completions.
Picking up new terms from result or preview text. Double-clicking on a word in the result list or in a preview window will copy it to the simple search entry field.
Wildcards. Wildcards can be used inside search terms in all forms of searches. More about wildcards.
Automatic suffixes. Words like
ods can be automatically turned into
clauses. This can be enabled in the Search preferences panel in the
Disabling stem expansion. Entering a
capitalized word in any search field will prevent stem
expansion (no search for
gardening if you enter
Garden instead of
garden). This is the only case where
character case should make a difference for a
Recoll search. You can
also disable stem expansion or change the stemming
language in the preferences.
Finding related documents. Selecting the Find similar documents entry in the result list paragraph right-click menu will select a set of "interesting" terms from the current result, and insert them into the simple search entry field. You can then possibly edit the list and start a search to find documents which may be apparented to the current result.
File names. File names are added as terms during indexing, and you can specify them as ordinary terms in normal search fields (Recoll used to index all directories in the file path as terms. This has been abandoned as it did not seem really useful). Alternatively, you can use the specific file name search which will only look for file names, and may be faster than the generic search especially when using wildcards.
Phrases and Proximity searches. A phrase
can be looked for by enclosing it in double quotes.
will look only for occurrences of
user immediately followed by
manual. You can use the
This phrase field of the
advanced search dialog to the same effect. Phrases can
be entered along simple terms in all simple or advanced
search entry fields (except This
AutoPhrases. This option can be set in
the preferences dialog. If it is set, a phrase will be
automatically built and added to simple searches when
This will not change radically the results, but will
give a relevance boost to the results where the search
terms appear as a phrase. Ie: searching for
virtual reality will still
find all documents where either
reality or both appear, but those
reality should appear sooner in the list.
Phrase searches can strongly slow down a query if
most of the terms in the phrase are common. This is why
autophrase option is
off by default for Recoll versions before 1.17. As of
is on by default, but very common terms will be removed
from the constructed phrase. The removal threshold can
be adjusted from the search preferences.
Phrases and abbreviations. As of
Recoll version 1.17,
dotted abbreviations like
I.B.M. are also automatically indexed
as a word without the dots:
IBM. Searching for the word inside a
company") will only match the dotted
abrreviation if you increase the phrase slack (using
the advanced search panel control, or the
o query language modifier). Literal
occurences of the word will be matched normally.
Using fields. You can use the query language and
field specifications to only search certain parts of
documents. This can be especially helpful with email,
for example only searching emails from a specific
Ajusting the result table columns. When displaying results in table mode, you can use a right click on the table headers to activate a pop-up menu which will let you adjust what columns are displayed. You can drag the column headers to adjust their order. You can click them to sort by the field displayed in the column. You can also save the result list in CSV format.
Changing the GUI geometry. It is possible to configure the GUI in wide form factor by dragging the toolbars to one of the sides (their location is remembered between sessions), and moving the category filters to a menu (can be set in the→ → panel).
Query explanation. You can get an exact description of what the query looked for, including stem expansion, and Boolean operators used, by clicking on the result list header.
Advanced search history. As of Recoll 1.18, you can display any of the last 100 complex searches performed by using the up and down arrow keys while the advanced search panel is active.
Browsing the result list inside a preview window. Entering Shift-Down or Shift-Up (Shift + an arrow key) in a preview window will display the next or the previous document from the result list. Any secondary search currently active will be executed on the new document.
Scrolling the result list from the keyboard. You can use PageUp and PageDown to scroll the result list, Shift+Home to go back to the first page. These work even while the focus is in the search entry.
Result table: moving the focus to the table. You can use Ctrl-r to move the focus from the search entry to the table, and then use the arrow keys to change the current row. Ctrl-Shift-s returns to the search.
Result table: open / preview. With the focus in the result table, you can use Ctrl-o to open the document from the current row, Ctrl-Shift-o to open the document and close recoll, Ctrl-d to preview the document.
Editing a new search while the focus is not in the search entry. You can use the Ctrl-Shift-S shortcut to return the cursor to the search entry (and select the current search text), while the focus is anywhere in the main window.
Forced opening of a preview window. You
can use Shift+Click on a
to force the creation of a preview window instead of a
new tab in the existing one.
Closing previews. Entering Ctrl-W in a tab will close it (and, for the last tab, close the preview window). Entering Esc will close the preview window and all its tabs.
Printing previews. Entering Ctrl-P in a preview window will print the currently displayed text.
Quitting. Entering Ctrl-Q almost anywhere will close the application.
Both simple and advanced query dialogs save recent history, but the amount is limited: old queries will eventually be forgotten. Also, important queries may be difficult to find among others. This is why both types of queries can also be explicitely saved to files, from the GUI menus:→
The default location for saved queries is a subdirectory of the current configuration directory, but saved queries are ordinary files and can be written or moved anywhere.
Some of the saved query parameters are part of the
or the active external indexes), and may differ when the
query is loaded from the time it was saved. In this case,
Recoll will warn of the
differences, but will not change the user
You can customize some aspects of the search interface by using theentry in the menu.
There are several tabs in the dialog, dealing with the interface itself, the parameters used for searching and returning results, and what indexes are searched.
Highlight color for query
terms: Terms from the user query are
highlighted in the result list samples and the
preview window. The color can be chosen here. Any
Qt color string should work (ie
#ff0000). The default is
Style sheet: The
name of a Qt style
sheet text file which is applied to the whole
Recoll application on startup. The default value is
empty, but there is a skeleton style sheet
directory. Using a style sheet, you can change most
parameters: colors, fonts, etc. See the sample file
for a few simple examples.
You should be aware that parameters (e.g.: the background color) set inside the Recoll GUI style sheet will override global system preferences, with possible strange side effects: for example if you set the foreground to a light color and the background to a dark one in the desktop preferences, but only the background is set inside the Recoll style sheet, and it is light too, then text will appear light-on-light inside the Recoll GUI.
Maximum text size highlighted for preview Inserting highlights on search term inside the text before inserting it in the preview window involves quite a lot of processing, and can be disabled over the given text size to speed up loading.
Prefer HTML to plain text for preview if set, Recoll will display HTML as such inside the preview window. If this causes problems with the Qt HTML display, you can uncheck it to display the plain text version instead.
Plain text to HTML line style: when displaying plain text inside the preview window, Recoll tries to preserve some of the original text line breaks and indentation. It can either use PRE HTML tags, which will well preserve the indentation but will force horizontal scrolling for long lines, or use BR tags to break at the original line breaks, which will let the editor introduce other line breaks according to the window width, but will lose some of the original indentation. The third option has been available in recent releases and is probably now the best one: use PRE tags with line wrapping.
Choose editor applicationsr: this opens a dialog which allows you to select the application to be used to open each MIME type. The default is nornally to use the xdg-open utility, but you can override it.
Exceptions: even wen xdg-open is used by default for opening documents, you can set exceptions for MIME types that will still be opened according to Recoll preferences. This is useful for passing parameters like page numbers or search strings to applications that support them (e.g. evince). This cannot be done with xdg-open which only supports passing one parameter.
Document filter choice style: this will let you choose if the document categories are displayed as a list or a set of buttons, or a menu.
Start with simple search
mode: this lets you choose the value of the
simple search type on program startup. Either a
fixed value (e.g.
Language, or the value in use when the
program last exited.
Auto-start simple search on white space entry: if this is checked, a search will be executed each time you enter a space in the simple search input field. This lets you look at the result list as you enter new terms. This is off by default, you may like it or not...
Start with advanced search dialog open : If you use this dialog frequently, checking the entries will get it to open when recoll starts.
Remember sort activation state if set, Recoll will remember the sort tool stat between invocations. It normally starts with sorting disabled.
Number of results in a result page
Result list font: There is quite a lot of information shown in the result list, and you may want to customize the font and/or font size. The rest of the fonts used by Recoll are determined by your generic Qt config (try the qtconfig command).
Edit result list paragraph format string: allows you to change the presentation of each result list entry. See the result list customisation section.
Edit result page HTML header insert: allows you to define text inserted at the end of the result page HTML header. More detail in the result list customisation section.
Date format: allows specifying the format used for displaying dates inside the result list. This should be specified as an strftime() string (man strftime).
Abstract snippet separator: for synthetic abstracts built from index data, which are usually made of several snippets from different parts of the document, this defines the snippet separator, an ellipsis by default.
Hide duplicate results: decides if result list entries are shown for identical documents found in different places.
Stemming language: stemming obviously depends on the document's language. This listbox will let you chose among the stemming databases which were built during indexing (this is set in the main configuration file), or later added with recollindex -s (See the recollindex manual). Stemming languages which are dynamically added will be deleted at the next indexing pass unless they are also added in the configuration file.
Automatically add phrase
to simple searches: a phrase will be
automatically built and added to simple searches
when looking for
terms. This will give a relevance boost to
the results where the search terms appear as a
phrase (consecutive and in order).
Autophrase term frequency threshold percentage: very frequent terms should not be included in automatic phrase searches for performance reasons. The parameter defines the cutoff percentage (percentage of the documents where the term appears).
Replace abstracts from documents: this decides if we should synthesize and display an abstract in place of an explicit abstract found within the document itself.
Dynamically build abstracts: this decides if Recoll tries to build document abstracts (lists of snippets) when displaying the result list. Abstracts are constructed by taking context from the document information, around the search terms.
Synthetic abstract size: adjust to taste...
Synthetic abstract context words: how many words should be displayed around each term occurrence.
Query language magic file
name suffixes: a list of words which
automatically get turned into
ext:xxx file name suffix clauses
when starting a query language query (ie:
doc xls xlsx...). This
will save some typing for people who use file types
a lot when querying.
indexes: This panel will let you browse for
additional indexes that you may want to search. External
indexes are designated by their database directory (ie:
Once entered, the indexes will appear in the External indexes list, and you can chose which ones you want to use at any moment by checking or unchecking their entries.
Your main database (the one the current configuration indexes to), is always implicitly active. If this is not desirable, you can set up your configuration so that it indexes, for example, an empty directory. An alternative indexer may also need to implement a way of purging the index from stale data,
The result list presentation can be exhaustively customized by adjusting two elements:
The paragraph format
HTML code inside the header section. For versions 1.21 and later, this is also used for the snippets window
The paragraph format and the header fragment can be edited from the Result list tab of the GUI configuration.
The header fragment is used both for the result list
and the snippets window. The snippets list is a table
and has a
attribute. Each paragraph in the result list is a
table, with class
but this can be changed by editing the paragraph
There are a few examples on the page about customising the result list on the Recoll web site.
This is an arbitrary HTML string where the
substitutions will be performed:
%I. Icon image name. This is
normally determined from the MIME type. The
associations are defined inside the
file. If a thumbnail for the file is found
at the standard Freedesktop location, this will
be displayed instead.
%K. Keywords (if any)
%L. Precooked Preview, Edit, and possibly Snippets links
%M. MIME type
%N. result Number inside the result page
%P. Parent folder Url. In the case of an embedded document, this is the parent folder for the top level container file.
%R. Relevance percentage
%S. Size information
%T. Title or Filename if not set.
%t. Title or Filename if not set.
The format of the Preview, Edit, and Snippets
docnum (%N) expands
to the document number inside the result page).
A link target defined as
"F%N" will open the document
corresponding to the
parent folder expansion, usually creating a file
manager window on the folder where the container file
A link target defined as
will run the corresponding script on the result file
(if the document is embedded, the script will be
started on the top-level parent). See the
section about defining scripts.
In addition to the predefined values above, all
%(fieldname) will be replaced by the
value of the field named
fieldname for this document. Only
stored fields can be accessed in this way, the value
of indexed but not stored fields is not known at this
point in the search process (see field
configuration). There are currently very few
fields stored by default, apart from the values above
filename), so this
feature will need some custom local configuration to
be useful. An example candidate would be the
recipient field which is
generated by the message input handlers.
The default value for the paragraph format string is:
"<table class=\"respar\">\n" "<tr>\n" "<td><a href='%U'><img src='%I' width='64'></a></td>\n" "<td>%L <i>%S</i> <b>%T</b><br>\n" "<span style='white-space:nowrap'><i>%M</i> %D</span> <i>%U</i> %i<br>\n" "%A %K</td>\n" "</tr></table>\n"
You may, for example, try the following for a more web-like experience:
<u><b><a href="P%N">%T</a></b></u><br> %A<font color=#008000>%U - %S</font> - %L
Note that the P%N link in the above paragraph makes the title a preview link. Or the clean looking:
<img src="%I" align="left">%L <font color="#900000">%R</font> <b>%T&</b><br>%S <font color="#808080"><i>%U</i></font> <table bgcolor="#e0e0e0"> <tr><td><div>%A</div></td></tr> </table>%K
These samples, and some others are on the web site, with pictures to show how they look.
It is also possible to define the value of the snippet separator inside the abstract section.
The Recoll KIO slave allows performing a Recoll search by entering an appropriate URL in a KDE open dialog, or with an HTML-based interface displayed in Konqueror.
The HTML-based interface is similar to the Qt-based interface, but slightly less powerful for now. Its advantage is that you can perform your search while staying fully within the KDE framework: drag and drop from the result list works normally and you have your normal choice of applications for opening files.
The alternative interface uses a directory view of search results. Due to limitations in the current KIO slave interface, it is currently not obviously useful (to me).
The interface is described in more detail inside a
help file which you can access by entering
recoll:/ inside the konqueror URL line
(this works only if the recoll KIO slave has been
The instructions for building this module are located
in the source tree. See:
kde/kio/recoll/00README.txt. Some Linux
distributions do package the kio-recoll module, so check
before diving into the build process, maybe it's already
out there ready for one-click installation.
As a sample application, the Recoll KIO slave could allow preparing a set of HTML documents (for example a manual) so that they become their own search interface inside konqueror.
This can be done by either explicitly inserting
href="recoll://..."> links around some document
areas, or automatically by adding a very small
the documents, like the following example, which would
initiate a search by double-clicking any term:
There are several ways to obtain search results as a text stream, without a graphical interface:
By passing option
to the recoll program.
By using the recollq program.
By writing a custom Python program, using the Recoll Python API.
The first two methods work in the same way and
accept/need the same arguments (except for the additional
-t to recoll). The query to be
executed is specified as command line arguments.
not built by default. You can use the
Makefile in the
query directory to build it. This is a
very simple program, and if you can program a little c++,
you may find it useful to taylor its output format to your
needs. Not that recollq is only really useful on systems
where the Qt libraries (or even the X11 ones) are not
available. Otherwise, just use
-t, which takes the exact same parameters and
options which are described for recollq
has a man page (not installed by default, look in the
doc/man directory). The Usage
string is as follows:
recollq: usage: -P: Show the date span for all the documents present in the index [-o|-a|-f] [-q] <query string> Runs a recoll query and displays result lines. Default: will interpret the argument(s) as a xesam query string query may be like: implicit AND, Exclusion, field spec: t1 -t2 title:t3 OR has priority: t1 OR t2 t3 OR t4 means (t1 OR t2) AND (t3 OR t4) Phrase: "t1 t2" (needs additional quoting on cmd line) -o Emulate the GUI simple search in ANY TERM mode -a Emulate the GUI simple search in ALL TERMS mode -f Emulate the GUI simple search in filename mode -q is just ignored (compatibility with the recoll GUI command line) Common options: -c <configdir> : specify config directory, overriding $RECOLL_CONFDIR -d also dump file contents -n [first-]<cnt> define the result slice. The default value for [first] is 0. Without the option, the default max count is 2000. Use n=0 for no limit -b : basic. Just output urls, no mime types or titles -Q : no result lines, just the processed query and result count -m : dump the whole document meta array for each result -A : output the document abstracts -S fld : sort by field <fld> -s stemlang : set stemming language to use (must exist in index...) Use -s "" to turn off stem expansion -D : sort descending -i <dbdir> : additional index, several can be given -e use url encoding (%xx) for urls -F <field name list> : output exactly these fields for each result. The field values are encoded in base64, output in one line and separated by one space character. This is the recommended format for use by other programs. Use a normal query with option -m to see the field names.
recollq 'ilur -nautique mime:text/html' Recoll query: ((((ilur:(wqf=11) OR ilurs) AND_NOT (nautique:(wqf=11) OR nautiques OR nautiqu OR nautiquement)) FILTER Ttext/html)) 4 results text/html [file:///Users/uncrypted-dockes/projets/bateaux/ilur/comptes.html] [comptes.html] 18593 bytes text/html [file:///Users/uncrypted-dockes/projets/nautique/webnautique/articles/ilur1/index.html] [Constructio... text/html [file:///Users/uncrypted-dockes/projets/pagepers/index.html] [psxtcl/writemime/recoll]... text/html [file:///Users/uncrypted-dockes/projets/bateaux/ilur/factEtCie/recu-chasse-maree....
Term synonyms: there are a number of ways to use term synonyms for searching text:
At index creation time, they can be used to alter the indexed terms, either increasing or decreasing their number, by expanding the original terms to all synonyms, or by reducing all synonym terms to a canonical one.
At query time, they can be used to match texts containing terms which are synonyms of the ones specified by the user, either by expanding the query for all synonyms, or by reducing the user entry to canonical terms (the latter only works if the corresponding processing has been performed while creating the index).
Recoll only uses
synonyms at query time. A user query term which part of a
synonym group will be optionally expanded into an
OR query for all terms in the
Synonym groups are defined inside ordinary text files. Each line in the file defines a group.
hi hello "good morning" # not sure about "au revoir" though. Is this english ? bye goodbye "see you" \ "au revoir"
As usual, lines beginning with a
# are comments, empty lines are ignored,
and lines can be continued by ending them with a
Multi-word synonyms are supported, but be aware that these will generate phrase queries, which may degrade performance and will disable stemming expansion for the phrase terms.
The synonyms file can be specified in the Search parameters tab of the GUI configuration Preferences menu entry, or as an option for command-line searches.
Once the file is defined, the use of synonyms can be enabled or disabled directly from the Preferences menu.
The synonyms are searched for matches with user terms after the latter are stem-expanded, but the contents of the synonyms file itself is not subjected to stem expansion. This means that a match will not be found if the form present in the synonyms file is not present anywhere in the document set.
The synonyms function is probably not going to help you find your letters to Mr. Smith. It is best used for domain-specific searches. For example, it was initially suggested by a user performing searches among historical documents: the synonyms file would contains nicknames and aliases for each of the persons of interest.
In some cases, the document paths stored inside the index do not match the actual ones, so that document previews and accesses will fail. This can occur in a number of circumstances:
When using multiple indexes it is a relatively
common occurrence that some will actually reside on a
remote volume, for exemple mounted via NFS. In this
case, the paths used to access the documents on the
local machine are not necessarily the same than the
ones used while indexing on the remote machine. For
have been used as a
topdirs elements while indexing, but
the directory might be mounted as
/net/server/home/me on the local
The case may also occur with removable disks. It is perfectly possible to configure an index to live with the documents on the removable disk, but it may happen that the disk is not mounted at the same place so that the documents paths from the index are invalid.
As a last exemple, one could imagine that a big directory has been moved, but that it is currently inconvenient to run the indexer.
Recoll has a facility for rewriting access paths when extracting the data from the index. The translations can be defined for the main index and for any additional query index.
The path translation facility will be useful whenever the documents paths seen by the indexer are not the same as the ones which should be used at query time.
In the above NFS example, Recoll could be instructed to rewrite
file:///home/me URL from
the index to
accesses from the client.
The translations are defined in the
ptrans configuration file, which can
be edited by hand or from the GUI external indexes
configuration dialog: → , then click the
Paths translations button on
the right below the index list.
Due to a current bug, the GUI must be restarted after
(even when they were changed from the GUI).
The query language processor is activated in the GUI simple search entry when the search mode selector is set to Query Language. It can also be used with the KIO slave or the command line search. It broadly has the same capabilities as the complex search interface in the GUI.
The language was based on the now defunct Xesam user search language specification.
If the results of a query language search puzzle you and
you doubt what has been actually searched for, you can use
Show Query link at the
top of the result list to check the exact query which was
finally executed by Xapian.
Here follows a sample request that we are going to explain:
author:"john doe" Beatles OR Lennon Live OR Unplugged -potatoes
This would search for all documents with
John Doe appearing as a
phrase in the author field (exactly what this is would
depend on the document type, ie: the
From: header, for an email message), and
lennon and either
unplugged but not
potatoes (in any
part of the document).
An element is composed of an optional field
specification, and a value, separated by a colon (the field
separator is the last colon in the element). Examples:
The colon, if present, means "contains". Xesam defines other relations, which are mostly unsupported for now (except in special cases, described further down).
All elements in the search entry are normally combined
with an implicit AND. It is possible to specify that
elements be OR'ed instead, as in
OR must be entered literally (capitals),
and it has priority over the AND associations:
word1 AND (
word3) not (
Recoll versions 1.21 and later, allow using parentheses to group elements, which will sometimes make things clearer, and may allow expressing combinations which would have been difficult otherwise.
An element preceded by a
specifies a term that should not appear.
As usual, words inside quotes define a phrase (the order
of words is significant), so that
title:"prejudice pride" is
not the same as
title:pride, and is unlikely to find a
Words inside phrases and capitalized words are not stem-expanded. Wildcards may be used anywhere inside a term. Specifying a wild-card on the left of a term can produce a very slow search (or even an incorrect one if the expansion is truncated because of excessive size). Also see More about wildcards.
To save you some typing, recent Recoll versions (1.20 and later) interpret a comma-separated list of terms as an AND list inside the field. Use slash characters ('/') for an OR list. No white space is allowed. So
will search for documents with
lennon inside the
author field (in any order), and
would search for
Modifiers can be set on a double-quote value, for
example to specify a proximity search (unordered). See
the modifier section. No space
must separate the final double-quote and the modifiers
Recoll currently manages the following default fields:
caption are synonyms which specify
data to be searched for in the document title or
from for searching the
to for searching the
keyword for searching
the document-specified keywords (few documents
actually have any).
filename for the
document's file name. This is not necessarily set for
all documents: internal documents contained inside a
compound one (for example an EPUB section) do not
inherit the container file name any more, this was
replaced by an explicit field (see next).
Sub-documents can still have a specific
filename, if it is implied by the
document format, for example the attachment file name
for an email attachment.
This is set for all documents, both top-level and
contained sub-documents, and is always the name of
the filesystem directory entry which contains the
data. The terms from this field can only be matched
by an explicit field specification (as opposed to
which are also indexed as general document content).
This avoids getting matches for all the sub-documents
when searching for the container file name.
ext specifies the
file name extension (Ex:
Recoll 1.20 and later
have a way to specify aliases for the field names, which
will save typing, for example by aliasing
cfn. See the section about the
The document input handlers used while indexing have the possibility to create other fields with arbitrary names, and aliases may be defined in the configuration, so that the exact field search possibilities may be different for you if someone took care of the customisation.
The field syntax also supports a few field-like, but special, criteria:
dir for filtering the
results on file location (Ex:
-dir also works to find results not
in the specified directory (release >= 1.15.8).
Tilde expansion will be performed as usual (except
for a bug in versions 1.19 to 1.19.11p1). Wildcards
will be expanded, but please have a
look at an important limitation of wildcards in
Relative paths also make sense, for example,
can be specified, both positive and negative. For
example the following makes sense:
dir:recoll dir:src -dir:utils -dir:common
This would select results which have both
src in the path (in any order), and
which have not either
You can also use
A special aspect of
dir clauses is that the values in
the index are not transcoded to UTF-8, and never
lower-cased or unaccented, but stored as binary. This
means that you need to enter the values in the exact
lower or upper case, and that searches for names with
diacritics may sometimes be impossible because of
character set conversion issues. Non-ASCII UNIX file
paths are an unending source of trouble and are best
You need to use double-quotes around the path value if it contains space characters.
size for filtering
the results on file size. Example:
size<10000. You can use
= as operators. You can specify a
range like the following:
size>100 size<1000. The usual
k/K, m/M, g/G, t/T can
be used as (decimal) multipliers. Ex:
size>1k to search for files
bigger than 1000 bytes.
date for searching or
filtering on dates. The syntax for the argument is
based on the ISO8601 standard for dates and time
intervals. Only dates are supported, no times. The
general syntax is 2 elements separated by a
/ character. Each
element can be a date or a period of time. Periods
are specified as
n numbers are the
respective numbers of years, months or days, any of
which may be missing. Dates are specified as
DD. The days and
months parts may be missing. If the
/ is present but an element is
missing, the missing element is interpreted as the
lowest or highest date in the index. Examples:
basic syntax for an interval of dates.
2001-03-01/P1Y2M the same
specified with a period.
2001/ from the
beginning of 2001 to the latest date in the
2001 the whole
year of 2001
P2D/ means 2
days ago up to now if there are no documents
with dates in the future.
documents from 2003 or older.
Periods can also be specified with small letters (ie: p2y).
format for specifying the MIME type.
These clauses are processed besides the normal
Boolean logic of the search. Multiple values will be
OR'ed (instead of the normal AND). You can specify
types to be excluded, with the usual
-, and use wildcards. Example:
-mime:text/plain Specifying an explicit
boolean operator before a
mime specification is not supported
and will produce strange results.
rclcat for specifying the category
(as in text/media/presentation/etc.). The
classification of MIME types in categories is defined
in the Recoll
mimeconf), and can be modified or
extended. The default category names are those which
permit filtering results in the main GUI screen.
Categories are OR'ed like MIME types above, and can
be negated with
date criteria always affect the whole
query (they are applied as a final filter), even if set
with other terms inside a parenthese.
mime (or the equivalent
rclcat) is the only field with an
OR default. You do need to
ext terms for example.
Some characters are recognized as search modifiers
when found immediately after the closing double quote of
a phrase, as in
term"modifierchars. The actual "phrase" can be a
single term of course. Supported modifiers:
l can be used to
turn off stemming (mostly makes sense with
p because stemming is
off by default for phrases).
s can be used to
turn off synonym expansion, if a synonyms file is
in place (only for Recoll 1.22 and later).
o can be used to
specify a "slack" for phrase and proximity
searches: the number of additional terms that may
be found between the specified ones. If
o is followed by an
integer number, this is the slack, else the default
p can be used to
turn the default phrase search into a proximity one
C will turn on case
sensitivity (if the index supports it).
D will turn on
diacritics sensitivity (if the index supports
A weight can be specified for a query element by
specifying a decimal value at the start of the
For Recoll versions 1.18 and later, and when working with a raw index (not the default), searches can be sensitive to character case and diacritics. How this happens is controlled by configuration variables and what search data is entered.
The general default is that searches entered without
upper-case or accented characters are insensitive to case
and diacritics. An entry of
resume will match any of
Two configuration variables can automate switching on sensitivity (they were documented but actually did nothing until Recoll 1.22):
If this is set, search sensitivity to diacritics
will be turned on as soon as an accented character
exists in a search term. When the variable is set to
resume will start
a diacritics-unsensitive search, but
résumé will be matched
exactly. The default value is false.
If this is set, search sensitivity to character
case will be turned on as soon as an upper-case
character exists in a search term except for the first one.
When the variable is set to true,
Us will start a
diacritics-unsensitive search, but
US will be matched exactly. The
default value is true (contrary to
As in the past, capitalizing the first letter of a word will turn off its stem expansion and have no effect on case-sensitivity.
You can also explicitely activate case and diacritics
sensitivity by using modifiers with the query language.
C will make the term
make it diacritics-sensitive. Examples:
will search for the term
Us will not be a
will search for the term
resume exactly (
résumé will not be a
When either case or diacritics sensitivity is activated, stem expansion is turned off. Having both does not make much sense.
Some special characters are interpreted by Recoll in search strings to expand or specialize the search. Wildcards expand a root term in controlled ways. Anchor characters can restrict a search to succeed only if the match is found at or near the beginning of the document or one of its fields.
All words entered in Recoll search fields will be processed for wildcard expansion before the request is finally executed.
The wildcard characters are:
* which matches 0
or more characters.
? which matches a
 which allow
defining sets of characters to be matched (ex:
] matches a single character which
may be 'a' or 'b' or 'c',
] matches any number.
You should be aware of a few things when using wildcards.
Using a wildcard character at the beginning of a
word can make for a slow search because
Recoll will have
to scan the whole index term list to find the
matches. However, this is much less a problem for
field searches, and queries like
can sometimes be very useful.
For Recoll version 18 only, when working with a raw index (preserving character case and diacritics), the literal part of a wildcard expression will be matched exactly for case and diacritics. This is not true any more for versions 19 and later.
* at the
end of a word can produce more matches than you
would think, and strange search results. You can
use the term
explorer tool to check what completions exist
for a given term. You can also see exactly what
search was performed by clicking on the link at the
top of the result list. In general, for natural
language terms, stem expansion will produce better
results than an ending
* (stem expansion is turned off
when any wildcard character appears in the
Due to the way that Recoll processes wildcards inside
dir path filtering
clauses, they will have a multiplicative effect on the
query size. A clause containg wildcards in several
paths elements, like, for example,
will almost certainly fail if your indexed tree is of
any realistic size.
Depending on the case, you may be able to work
around the issue by specifying the paths elements more
narrowly, with a constant prefix, or by using 2
instead of multiple wildcards, as in
docdir. The latter
query is not equivalent to the initial one because it
does not specify a number of directory levels, but
that's the best we can do (and it may be actually more
useful in some cases).
Two characters are used to specify that a search hit
should occur at the beginning or at the end of the text.
^ at the beginning of a term
or phrase constrains the search to happen at the start,
$ at the end force it to
happen at the end.
As this function is implemented as a phrase search it is possible to specify a maximum distance at which the hit should occur, either through the controls of the advanced search panel, or using the query language, for example, as in:
which would force
someterm to be found within 10 terms of
the start of the text. This can be combined with a field
search as in
This feature can also be used with an actual phrase
search, but in this case, the distance applies to the
whole phrase and anchor, so that, for example,
bla bla my unexpected term
at the beginning of the text would be a match for
Anchored searches can be very useful for searches inside somewhat structured documents like scientific articles, in case explicit metadata has not been supplied (a most frequent case), for example for looking for matches inside the abstract or the list of authors (which occur at the top of the document).
Being independant of the desktop type has its drawbacks: Recoll desktop integration is minimal. However there are a few tools available:
Here follow a few other things that may help.
It is surprisingly convenient to be able to show or hide the Recoll GUI with a single keystroke. Recoll comes with a small Python script, based on the libwnck window manager interface library, which will allow you to do just this. The detailed instructions are on this wiki page.
This is probably obsolete now. Anyway:
The Recoll source tree contains the source code to the recoll_applet, a small application derived from the find_applet. This can be used to add a small Recoll launcher to the KDE panel.
The applet is not automatically built with the main
Recoll programs, nor is
it included with the main source distribution (because
the KDE build boilerplate makes it relatively big). You
can download its source from the recoll.org download
page. Use the omnipotent
install incantation to build and
You can then add the applet to the panel by right-clicking the panel and choosing the Add applet entry.
The recoll_applet has a small text window where you can type a Recoll query (in query language form), and an icon which can be used to restrict the search to certain types of files. It is quite primitive, and launches a new recoll GUI instance every time (even if it is already running). You may find it useful anyway.
Recoll has an Application Programming Interface, usable both for indexing and searching, currently accessible from the Python language.
Another less radical way to extend the application is to write input handlers for new types of documents.
The processing of metadata attributes for documents
fields) is highly
The small programs or pieces of code which handle the
processing of the different document types for
Recoll used to be called
filters, which is still
reflected in the name of the directory which holds them
and many configuration variables. They were named this
way because one of their primary functions is to filter
out the formatting directives and keep the text content.
However these modules may have other behaviours, and the
input handler is now
progressively substituted in the documentation.
filter is still used in many
Recoll input handlers cooperate to translate from the multitude of input document formats, simple ones as opendocument, acrobat), or compound ones such as Zip or Email, into the final Recoll indexing input format, which is plain text. Most input handlers are executable programs or scripts. A few handlers are coded in C++ and live inside recollindex. This latter kind will not be described here.
There are currently (since version 1.13) two kinds of external executable input handlers:
run once and exit. They can be bare programs like
scripts using other programs. They are very simple to
write, because they just need to print the converted
document to the standard output. Their output can be
plain text or HTML. HTML is usually preferred because
it can store metadata fields and it allows preserving
some of the formatting for the GUI preview.
handlers can process multiple files (sparing the
process startup time which can be very significant),
or multiple documents per file (e.g.: for
zip or chm files). They communicate
with the indexer through a simple protocol, but are
nevertheless a bit more complicated than the older
kind. Most of new handlers are written in
Python, using a
common module to handle the protocol. There is an
exception, rclimg which is
written in Perl. The subdocuments output by these
handlers can be directly indexable (text or HTML), or
they can be other simple or compound documents that
will need to be processed by another handler.
In both cases, handlers deal with regular file system files, and can process either a single document, or a linear list of documents in each file. Recoll is responsible for performing up to date checks, deal with more complex embedding and other upper level issues.
A simple handler returning a document in
text/plain format, can transfer no
metadata to the indexer. Generic metadata, like document
size or modification date, will be gathered and stored by
Handlers that produce
text/html format can return an arbitrary
amount of metadata inside HTML
meta tags. These will be processed
according to the directives found in the
fields configuration file.
The handlers that can handle multiple documents per file
return a single piece of data to identify each document
inside the file. This piece of data, called an
ipath element will be sent back by
Recoll to extract the
document at query time, for previewing, or for creating a
temporary file to be opened by a viewer.
The following section describes the simple handlers, and
the next one gives a few explanations about the
execm ones. You could
conceivably write a simple handler with only the elements
in the manual. This will not be the case for the other
ones, for which you will have to look at the code.
Recoll simple handlers are usually shell-scripts, but this is in no way necessary. Extracting the text from the native format is the difficult part. Outputting the format expected by Recoll is trivial. Happily enough, most document formats have translators or text extractors which can be called from the handler. In some cases the output of the translating program is completely appropriate, and no intermediate shell-script is needed.
Input handlers are called with a single argument which is the source file name. They should output the result to stdout.
When writing a handler, you should decide if it will output plain text or HTML. Plain text is simpler, but you will not be able to add metadata or vary the output character encoding (this will be defined in a configuration file). Additionally, some formatting may be easier to preserve when previewing HTML. Actually the deciding factor is metadata: Recoll has a way to extract metadata from the HTML header and use it for field searches..
no) tells the handler if the
operation is for indexing or previewing. Some handlers
use this to output a slightly different format, for
example stripping uninteresting repeated keywords (ie:
Subject: for email) when
indexing. This is not essential.
You should look at one of the simple handlers, for example rclps for a starting point.
Don't forget to make your handler executable before testing !
If you can program and want to write an
execm handler, it should not be too
difficult to make sense of one of the existing modules.
There is a sample one with many comments, not actually
used by Recoll, which
would index a text file as one document per line. Look
rcltxtlines.py in the
src/filters directory in
the Recoll BitBucket repository (the sample not in
the distributed release at the moment).
You can also have a look at the slightly more complex
which uses Zip file paths as identifiers (
execm handlers sometimes
need to make a choice for the nature of the
ipath elements that they use in
communication with the indexer. Here are a few
Use ASCII or UTF-8 (if the identifier is an integer print it, for example, like printf %d would do).
If at all possible, the data should make some kind of sense when printed to a log file to help with debugging.
Recoll uses a
:) as a
separator to store a complex path internally (for
deeper embedding). Colons inside the
ipath elements output by a handler
will be escaped, but would be a bad choice as a
handler-specific separator (mostly, again, for
In any case, the main goal is that it should be easy
for the handler to extract the target document, given the
file name and the
execm handlers will also
produce a document with a null
ipath element. Depending on the type of
document, this may have some associated data (e.g. the
body of an email message), or none (typical for an
archive file). If it is empty, this document will be
useful anyway for some operations, as the parent of the
actual data documents.
There are two elements that link a file to the handler which should process it: the association of file to MIME type and the association of a MIME type with a handler.
The association of files to MIME types is mostly based
on name suffixes. The types are defined inside the
mimemap file. Example:
.doc = application/msword
If no suffix association is found for the file name, Recoll will try to execute a system command (typically file -i or xdg-mime) to determine a MIME type.
The second element is the association of MIME types to
handlers in the
mimeconf file. A sample will
probably be better than a long explanation:
[index] application/msword = exec antiword -t -i 1 -m UTF-8;\ mimetype = text/plain ; charset=utf-8 application/ogg = exec rclogg text/rtf = exec unrtf --nopict --html; charset=iso-8859-1; mimetype=text/html application/x-chm = execm rclchm
The fragment specifies that:
files are processed by executing the antiword program,
text/plain encoded in
files are processed by the rclogg script,
with default output type (
text/html, with encoding specified
in the header, or
utf-8 by default).
processed by unrtf, which
iso-8859-1 encoding is
specified because it is not the
utf-8 default, and not output by
in the HTML header section.
is processed by a persistant handler. This is
determined by the
The output HTML could be very minimal like the following example:
<html> <head> <meta http-equiv="Content-Type" content="text/html;charset=UTF-8"> </head> <body> Some text content </body> </html>
You should take care to escape some characters inside
the text by transforming them into appropriate entities.
At the very minimum, "
should be transformed into "
<" should be transformed into
<". This is not
always properly done by translating programs which output
HTML, and of course never by those which output plain
When encapsulating plain text in an HTML body, the
display of a preview may be improved by enclosing the
The character set needs to be specified in the header. It does not need to be UTF-8 (Recoll will take care of translating it), but it must be accurate for good results.
Recoll will process
meta tags inside the header
as possible document fields candidates. Documents fields
can be processed by the indexer in different ways, for
searching or displaying inside query results. This is
described in a following
By default, the indexer will process the standard
header fields if they are present:
meta/keywords are both indexed and
stored for query-time display.
A predefined non-standard
meta tag will also be processed by
Recoll without further
configuration: if a
is present and has the right format, it will be used as
the document date (for display and sorting), in
preference to the file modification date. The date format
should be as follows:
<meta name="date" content="YYYY-mm-dd HH:MM:SS"> or <meta name="date" content="YYYY-mm-ddTHH:MM:SS">
<meta name="date" content="2013-02-24 17:50:00">
Input handlers also have the possibility to "invent" field names. This should also be output as meta tags:
<meta name="somefield" content="Some textual data" />
You can embed HTML markup inside the content of custom
fields, for improving the display inside result lists. In
this case, add a (wildly non-standard)
markup attribute to tell Recoll that the value is HTML and
should not be escaped for display.
<meta name="somefield" markup="html" content="Some <i>textual</i> data" />
As written above, the processing of fields is described in a further section.
Fields are named pieces of
information in or about documents, like
The field values for documents can appear in several
ways during indexing: either output by input handlers as
meta fields in the HTML header
section, or extracted from file extended attributes, or
added as attributes of the
object when using the API, or again synthetized internally
The Recoll query language allows searching for text in a specific field.
Recoll defines a number
of default fields. Additional ones can be output by
handlers, and described in the
fields configuration file.
Fields can be:
indexed, meaning that
their terms are separately stored in inverted lists
(with a specific prefix), and that a field-specific
search is possible.
stored, meaning that
their value is recorded in the index data record for
the document, and can be returned and displayed with
A field can be either or both indexed and stored. This
and other aspects of fields handling is defined inside the
The sequence of events for field processing is as follows:
During indexing, recollindex scans
meta fields in HTML
documents (most document types are transformed into
HTML at some point). It compares the name for each
element to the configuration defining what should be
done with fields (the
If the name for the
meta element matches one for a field
that should be indexed, the contents are processed
and the terms are entered into the index with the
prefix defined in the
If the name for the
meta element matches one for a field
that should be stored, the content of the element is
stored with the document data record, from which it
can be extracted and displayed at query time.
At query time, if a field search is performed, the index prefix is computed and the match is only performed against appropriately prefixed terms in the index.
At query time, the field can be displayed inside the result list by using the appropriate directive in the definition of the result list paragraph format. All fields are displayed on the fields screen of the preview window (which you can reach through the right-click menu). This is independant of the fact that the search which produced the results used the field or not.
You can find more information in the section about the
fields file, or in
comments inside the file.
You can also have a look at the example on the Wiki, detailing how one could add a page count field to pdf documents for displaying inside result lists.
Recoll versions after 1.11 define a Python programming interface, both for searching and creating/updating an index.
The search interface is used in the Recoll Ubuntu Unity Lens and the Recoll Web UI. It can run queries on any Recoll configuration.
The index update section of the API may be used to create and update Recoll indexes on specific configurations (separate from the ones created by recollindex). The resulting databases can be queried alone, or in conjunction with regular ones, through the GUI or any of the query interfaces.
The search API is modeled along the Python database API specification. There were two major changes along Recoll versions:
The basis for the Recoll API changed from Python database API version 1.0 (Recoll versions up to 1.18.1), to version 2.0 (Recoll 1.18.2 and later).
became a package (with an internal
recoll module) as of Recoll version 1.19, in order
to add more functions. For existing code, this only
changes the way the interface must be imported.
We will describe the new API and package structure here. A paragraph at the end of this section will explain a few differences and ways to write code compatible with both versions.
The Python interface can be found in the source
directory contains the usual
setup.py. After configuring the main
Recoll code, you can use
the script to build and install the Python module:
python setup.py build
python setup.py install
As of Recoll 1.19, the module can be compiled for Python3.
The normal Recoll installer installs the Python2 API along with the main code. The Python3 version must be explicitely built and installed.
When installing from a repository, and depending on the distribution, the Python API can sometimes be found in a separate package.
As an introduction, the following small sample will
run a query and list the title and url for each of the
results. It would work with Recoll 1.19 and later. The
directory contains several examples of Python programming
with Recoll, exercising
the extension more completely, and especially its data
#!/usr/bin/env python from recoll import recoll db = recoll.connect() query = db.query() nres = query.execute("some query") results = query.fetchmany(20) for doc in results: print(doc.url, doc.title)
A few elements in the interface are specific and and need an explanation.
This data value (set as a field in the Doc
object) is stored, along with the URL, but not
indexed by Recoll.
Its contents are not interpreted by the index
layer, and its use is up to the application. For
example, the Recoll file system indexer
store the part of the document access path internal
to (possibly imbricated) container documents.
ipath in this case is
a vector of access elements (e.g, the first part
could be a path inside a zip file to an archive
member which happens to be an mbox file, the second
element would be the message sequential number
inside the mbox etc.).
ipath are returned in every search
result and define the access to the original
empty for top-level document/files (e.g. a PDF
document which is a filesystem file). The
Recoll GUI knows
about the structure of the
ipath values used by the
filesystem indexer, and uses it for such functions
as opening the parent of a given document.
document identifier) identifies a document. Because
of limitations inside the index engine, it is
restricted in length (to 200 bytes), which is why a
regular URI cannot be used. The structure and
contents of the
defined by the application and opaque to the index
engine. For example, the internal file system
indexer uses the complete document path (file path
+ internal path), truncated to length, the
suppressed part being replaced by a hash value. The
udi is not explicit in
the query interface (it is used "under the hood" by
module), but it is an explicit element of the
If this attribute is set on a document when
entering it in the index, it designates its
physical container document. In a multilevel
hierarchy, this may not be the immediate parent.
optional, but its use by an indexer may simplify
index maintenance, as Recoll will automatically
delete all children defined by
parent_udi == udi when the
document designated by
udi is destroyed. e.g. if a
Zip archive contains
entries which are themselves containers, like
mbox files, all the
subdocuments inside the
Zip file (mbox, messages, message
attachments, etc.) would have the same
parent_udi, matching the
udi for the
Zip file, and all
would be destroyed when the
Zip file (identified by its
udi) is removed from
the index. The standard filesystem indexer uses
inside the Recoll
configuration defines which document fields are
either "indexed" (searchable), "stored"
(retrievable with search results), or both.
contains two modules:
contains functions and classes used to query (or
update) the index. This section will only
describe the query part, see further for the
module contains functions and classes used to
access document data.
function connects to one or several
index(es) and returns a
may specify a configuration directory.
The usual defaults apply.
is a list of additional indexes (Xapian
decides if we can index new data through
This call initializes the recoll module, and it should always be performed before any other call or object creation.
A Db object is created by a
connect() call and holds a
connection to a Recoll index.
Closes the connection. You can't do
anything with the
Db object after this.
These aliases return a blank
Query object for this
Set the parameters used to build snippets
(sets of keywords in context text fragments).
the maximum total size of the abstract.
defines how many terms are shown around the
Expand an expression against the index
term list. Performs the basic function from
the GUI term explorer tool.
match_type can be either of
stem. Returns a
list of terms expanded from the input
(equivalent to a cursor in the Python DB API) is
created by a
Db.query() call. It is used to
execute index searches.
Sort results by
ascending or descending order. Must be called
before executing the search.
Starts a search for
search language string.
Starts a search for the query defined by the SearchData object.
Fetches the next
Doc objects in the current
search results, and returns them as an array
of the required size, which is by default the
value of the
arraysize data member.
Fetches the next
Doc object from the current
Closes the query. The object is unusable after the call.
Adjusts the position in the current result
mode can be
Retrieves the expanded query terms as a list of pairs. Meaningful only after executexx In each pair, the first entry is a list of user terms (of size one for simple terms, or more for group and phrase clauses), the second a list of query terms as derived from the user terms and used in the Xapian Query.
Return the Xapian query description as a Unicode string. Meaningful only after executexx.
Will insert <span "class=rclmatch">,
</span> tags around the match areas in
the input text and return the modified text.
ishtml can be
set to indicate that the input text is HTML
and that HTML special characters should not
methods if set should be an
object with methods startMatch(i) and
endMatch() which will be called for each
match and should return a begin and end
Create a snippets abstract for
Doc object) by
selecting text around the match terms. If
methods is set, will also perform
highlighting. See the highlight method.
So that things like
for doc in query: will
Default number of records processed by fetchmany (r/w).
Number of records returned by the last execute.
Next index to be fetched from results.
Normally increments after each fetchone()
call, but can be set/reset before the call to
effect seeking (equivalent to using
Starts at 0.
contains index data for a given document. The data
is extracted from the index when searching, or set
by the indexer program when updating. The Doc
object has many attributes to be read or set by its
user. It matches exactly the Rcl::Doc C++ object.
Some of the attributes are predefined, but,
especially when indexing, others can be set, the
name of which will be processed as field names by
the indexing configuration. Inputs can be specified
as Unicode or strings. Outputs are Unicode objects.
All dates are specified as Unix timestamps, printed
as strings. Please refer to the
rcldb/rcldoc.h C++ file for a
description of the predefined attributes.
At query time, only the fields that are defined
stored either by
default or in the
fields configuration file will be
meaningful in the
object. Especially this will not be the case for
the document text. See the
rclextract module for accessing
Retrieve the named doc attribute. You can
Set the the named doc attribute. You can
Retrieve the URL in byte array format (no transcoding), for use as parameter to a system call.
Set the URL in byte array format (no transcoding).
Return a dictionary of doc object keys/values
list of doc object keys (attribute names).
allows building a query by combining clauses, for
Query.executesd(). It can be used
in replacement of the query language approach. The
interface is going to change a little, so no
detailed doc for now...
Index queries do not provide document content (only
a partial and unprecise reconstruction is performed to
show the snippets text). In order to access the actual
document data, the data extraction part of the indexing
process must be performed (subdocument access and
format translation). This is not trivial in general.
currently provides a single class which can be used to
access the data content for result documents.
object is built from a
Doc object, output from a
Extract document defined by
object. The doc.text field has the document
text converted to either text/plain or
text/html according to doc.mimetype. The
typical use would be as follows:
qdoc = query.fetchone() extractor = recoll.Extractor(qdoc) doc = extractor.textextract(qdoc.ipath) # use doc.text, e.g. for previewing
Extracts document into an output file, which can be given explicitly or will be created as a temporary file to be deleted by the caller. Typical use:
qdoc = query.fetchone() extractor = recoll.Extractor(qdoc) filename = extractor.idoctofile(qdoc.ipath, qdoc.mimetype)
The following sample would query the index with a
user language string. See the
python/samples directory inside the
Recoll source for
other examples. The
recollgui subdirectory has a very
embryonic GUI which demonstrates the highlighting and
data extraction functions.
#!/usr/bin/env python from recoll import recoll db = recoll.connect() db.setAbstractParams(maxchars=80, contextwords=4) query = db.query() nres = query.execute("some user question") print "Result count: ", nres if nres > 5: nres = 5 for i in range(nres): doc = query.fetchone() print "Result #%d" % (query.rownumber,) for k in ("title", "size"): print k, ":", getattr(doc, k).encode('utf-8') abs = db.makeDocAbstract(doc, query).encode('utf-8') print abs print
The update API can be used to create an index from data which is not accessible to the regular Recoll indexer, or structured to present difficulties to the Recoll input handlers.
An indexer created using this API will be have equivalent work to do as the the Recoll file system indexer: look for modified documents, extract their text, call the API for indexing it, take care of purging the index out of data from documents which do not exist in the document store any more.
The data for such an external indexer should be stored in an index separate from any used by the Recoll internal file system indexer. The reason is that the main document indexer purge pass (removal of deleted documents) would also remove all the documents belonging to the external indexer, as they were not seen during the filesystem walk. The main indexer documents would also probably be a problem for the external indexer own purge operation.
While there would be ways to enable multiple foreign indexers to cooperate on a single index, it is just simpler to use separate ones, and use the multiple index access capabilities of the query interface, if needed.
There are two parts in the update interface:
Methods inside the
recoll module allow inserting
data into the index, to make it accessible by the
normal query interface.
An interface based on scripts execution is
defined to allow either the GUI or the
rclextract module to access
original document data for previewing or
The update methods are part of the
recoll module described above. The
connect() method is used with a
writable=true parameter to obtain a
Db object. The
methods are then available.
Add or update index data for a given document
string must define a unique id for the document.
It is an opaque interface element and not
interpreted inside Recoll.
doc is a
created from the data to be indexed (the main
text should be in
is set, this is a unique identifier for the
top-level container (e.g. for the filesystem
indexer, this would be the one which is an actual
Purge index from all data for
udi, and all documents (if any)
which have a matrching
Test if the index needs to be updated for the
document identified by
udi. If this call is to be used,
should contain a signature value when calling
then compares its parameter value with the stored
sig is an opaque value, compared
as a string.
The filesystem indexer uses a concatenation of the decimal string values for file size and update time, but a hash of the contents could also be used.
As a side effect, if the return value is false
(the index is up to date), the call will set the
existence flag for the document (and any
subdocument defined by its
parent_udi), so that a later
purge() call will
The use of
purge() is optional, and the
indexer may use another method for checking the
need to reindex or to delete stale entries.
Delete all documents that were not touched during the just finished indexing pass (since open-for-write). These are the documents for the needUpdate() call was not performed, indicating that they no longer exist in the primary storage system.
Recoll has internal
methods to access document data for its internal
(filesystem) indexer. An external indexer needs to
provide data access methods if it needs integration
with the GUI (e.g. preview function), or support for
The index data and the access method are linked by
rclbes (recoll backend
Doc field. You
should set this to a short string value identifying
your indexer (e.g. the filesystem indexer uses either
"FS" or an empty value, the Web history indexer uses
The link is actually performed inside a
backends configuration file (stored
in the configuration directory). This defines commands
to execute to access data from the specified indexer.
Example, for the mbox indexing sample found in the
Recoll source (which sets
[MBOX] fetch = /path/to/recoll/src/python/samples/rclmbox.py fetch makesig = path/to/recoll/src/python/samples/rclmbox.py makesig
makesig define two commands to execute
to respectively retrieve the document text and compute
the document signature (the example implementation uses
the same script with different first parameters to
perform both operations).
The scripts are called with three additional
ipath, stored with the document when
it was indexed, and may use any or all to perform the
requested operation. The caller expects the result data
The Recoll source tree has two samples of external
indexers in the
src/python/samples directory. The
more interesting one is
rclmbox.py which indexes a directory
files. It exercises most features in the update
interface, and has a data access interface.
See the comments inside the file for more information.
The following code fragments can be used to ensure that code can run with both the old and the new API (as long as it does not use the new abilities of the new API of course).
Adapting to the new package structure:
try: from recoll import recoll from recoll import rclextract hasextract = True except: import recoll hasextract = False
Adapting to the change of nature of the
member. The same test can be used to choose to use the
scroll() method (new) or set
next value (old).
rownum = query.next if type(query.next) == int else \ query.rownumber
Recoll binary copies are always distributed as regular packages for your system. They can be obtained either through the system's normal software distribution framework (e.g. Debian/Ubuntu apt, FreeBSD ports, etc.), or from some type of "backports" repository providing versions newer than the standard ones, or found on the Recoll WEB site in some cases. The most up-to-date information about Recoll packages can usually be found on the Recoll WEB site downloads page
There used to exist another form of binary install, as pre-compiled source trees, but these are just less convenient than the packages and don't exist any more.
The package management tools will usually automatically deal with hard dependancies for packages obtained from a proper package repository. You will have to deal with them by hand for downloaded packages (for example, when dpkg complains about missing dependancies).
In all cases, you will have to check or install supporting applications for the file types that you want to index beyond those that are natively processed by Recoll (text, HTML, email files, and a few others).
You should also maybe have a look at the configuration section (but this may not be necessary for a quick test with default parameters). Most parameters can be more conveniently set from the GUI interface.
The Windows installation of Recoll is self-contained, and only needs Python 2.7 to be externally installed. Windows users can skip this section.
Recoll uses external applications to index some file types. You need to install them for the file types that you wish to have indexed (these are run-time optional dependencies. None is needed for building or running Recoll except for indexing their specific file type).
After an indexing pass, the commands that were found
missing can be displayed from the recoll File menu. The list is stored in the
missing text file inside the
A list of common file types which need external commands follows. Many of the handlers need the iconv command, which is not always listed as a dependancy.
Please note that, due to the relatively dynamic nature of this information, the most up to date version is now kept on http://www.recoll.org/features.html along with links to the home pages or best source/patches pages, and misc tips. The list below is not updated often and may be quite stale.
For many Linux distributions, most of the commands listed can be installed from the package repositories. However, the packages are sometimes outdated, or not the best version for Recoll, so you should take a look at http://www.recoll.org/features.html if a file type is important to you.
As of Recoll release 1.14, a number of XML-based formats that were handled by ad hoc handler code now use the xsltproc command, which usually comes with libxslt. These are: abiword, fb2 (ebooks), kword, openoffice, svg.
Now for the list:
Openoffice files need unzip and xsltproc.
PDF files need pdftotext which is
part of Poppler
(usually comes with the
poppler-utils package). Avoid the
original one from Xpdf.
Postscript files need pstotext. The original version has an issue with shell character in file names, which is corrected in recent packages. See http://www.recoll.org/features.html for more detail.
MS Word needs antiword. It is also useful to have wvWare installed as it may be be used as a fallback for some files which antiword does not handle.
MS Excel and PowerPoint are processed by internal Python handlers.
MS Open XML (docx) needs xsltproc.
Wordperfect files need wpd2html from the libwpd (or libwpd-tools on Ubuntu) package.
RTF files need unrtf, which, in its older versions, has much trouble with non-western character sets. Many Linux distributions carry outdated unrtf versions. Check http://www.recoll.org/features.html for details.
TeX files need untex or detex. Check http://www.recoll.org/features.html for sources if it's not packaged for your distribution.
dvi files need dvips.
djvu files need djvutxt and djvused from the DjVuLibre package.
Audio files: Recoll releases 1.14 and later use a single Python handler based on mutagen for all audio file types.
Pictures: Recoll uses the Exiftool Perl package to extract tag information. Most image file formats are supported. Note that there may not be much interest in indexing the technical tags (image size, aperture, etc.). This is only of interest if you store personal tags or textual descriptions inside the image files.
chm: files in Microsoft help format need Python and the pychm module (which needs chmlib).
ICS: up to Recoll 1.13, iCalendar files need Python and the icalendar module. icalendar is not needed for newer versions, which use internal code.
Zip archives need Python (and the standard zipfile module).
Rar archives need Python, the rarfile Python module and the unrar utility.
Midi karaoke files need Python and the Midi module
Konqueror webarchive format with Python (uses the Tarfile module).
Mimehtml web archive format (support based on the email handler, which introduces some mild weirdness, but still usable).
Text, HTML, email folders, and Scribus files are processed internally. Lyx is used to index Lyx files. Many handlers need iconv and the standard sed and awk.
If you can install any or all of the following through the package manager for your system, all the better. Especially Qt is a very big piece of software, but you will most probably be able to find a binary package.
You may have to compile Xapian but this is easy.
The shopping list:
libtool triad. Only
autoconf is needed up
C++ compiler. Up to Recoll version 1.13.04, its absence can manifest itself by strange messages about a missing iconv_open.
bison command (for Recoll 1.21 and later).
For building the documentation (for Recoll 1.21 and later). This
sometimes comes with the
libxslt package. And also the
Docbook XML and style sheet files.
Development files for Xapian core.
If you are building Xapian for an older CPU
(before Pentium 4 or Athlon 64), you need to add
flag to the configure command. Else all Xapian
application will crash with an
illegal instruction error.
Development files for Qt 4 . Recoll has not been tested with Qt 5 yet. Recoll 1.15.9 was the last version to support Qt 3. If you do not want to install or build the Qt Webkit module, Recoll has a configuration option to disable its use (see further).
Development files for X11 and zlib.
Development files for Python (or use
You may also need libiconv. On Linux systems, the iconv interface is part of libc and you should not need to do anything special.
Check the Recoll download page for up to date version information.
Recoll has been built on Linux, FreeBSD, Mac OS X, and Solaris, most versions after 2005 should be ok, maybe some older ones too (Solaris 8 is ok). If you build on another system, and need to modify things, I would very much welcome patches.
will disable the code for phonetic matching of
enable the code for real time indexing. Inotify
support is enabled by default on recent Linux
will enable sending Zeitgeist events about the
visited search results, and needs the qzeitgeist package.
available from version 1.17 to implement the result
list with a Qt
QTextBrowser instead of a WebKit widget if you do
not or can't depend on the latter.
is available from version 1.19 to suppress
multithreading inside the indexing process. You can
also use the run-time configuration to restrict
using a single thread, but the compile-time option
may disable a few more unused locks. This only
applies to the use of multithreading for the core
index processing (data input). The Recoll monitor mode always
uses at least two threads of execution.
--disable-python-module will avoid
building the Python module.
prevent fetching data from file extended
attributes. Beyond a few standard attributes,
fetching extended attributes data can only be
useful is some application stores data in there,
and also needs some simple configuration (see
comments in the
fields configuration file).
will enable splitting
This is not enabled by default as it has the
unfortunate side-effect of making some phrase
searches quite confusing: ie,
"MySQL manual" would be matched by
"MySQL manual" and
"my sql manual" but
(only inside phrase searches).
Specify the version of the 'file' command to use
(ie: --with-file-command=/usr/local/bin/file). Can
be useful to enable the gnu version on systems
where the native one is bad.
Disable the Qt interface. Will allow building the
indexer and the command line search program in
absence of a Qt environment.
connection monitoring inside recollindex. Together
with --disable-qtgui, this allows building recoll
without Qt and
will avoid building the user manual. This avoids
having to install the Docbook XML/XSL files and the
TeX toolchain used for translating the manual to
up to 1.21 only) will compile Recoll with position-dependant
code. This is incompatible with building the KIO or
the Python or
but might yield very marginally faster code.
Of course the usual autoconf configure
Normal procedure (for source extracted from a tar distribution):
(practices usual hardship-repelling invocations)
When building from source cloned from the BitBucket
repository, you also need to install autoconf, automake, and libtool and you must execute
sh autogen.sh in the top
source directory before running
We did not test building the GUI on Solaris for recent versions. You will need at least Qt 4.4. There are some hints on an old web site page, they may still be valid.
Someone did test the 1.19 indexer and Python module build, they do work, with a few minor glitches. Be sure to use GNU make and install.
install or execute
in the root of the source tree. This will copy the
sample configuration files, scripts and other shared data
If the installation prefix given to recollinstall is
different from either the system default or the value
which was specified when executing configure (as in
/some/path), you will have to set the
variable to indicate where the shared data is to be found
(ie for (ba)sh:
You can then proceed to configuration.
Most of the parameters specific to the recoll GUI are set
through the Preferences menu
and stored in the standard Qt place (
You probably do not want to edit this by hand.
Recoll indexing options are set inside text configuration files located in a configuration directory. There can be several such directories, each of which defines the parameters for one index.
The configuration files can be edited by hand or through the Index configuration dialog (Preferences menu). The GUI tool will try to respect your formatting and comments as much as possible, so it is quite possible to use both approaches on the same configuration.
The most accurate documentation for the configuration parameters is given by comments inside the default files, and we will just give a general overview here.
For each index, there are at least two sets of
configuration files. System-wide configuration files are
kept in a directory named like
/usr/share/recoll/examples, and define
default values, shared by all indexes. For each index, a
parallel set of files defines the customized
The default location of the customized configuration is
.recoll directory in your
home. Most people will only use this directory.
This location can be changed, or others can be added
environment variable or the
option parameter to recoll and recollindex.
In addition (as of Recoll version 1.19.7), it is possible
to specify two additional configuration directories which
will be stacked before and after the user configuration
directory. These are defined by the
RECOLL_CONFMID environment variables. Values
from configuration files inside the top directory will
override user ones, values from configuration files inside
the middle directory will override system ones and be
overriden by user ones. These two variables may be of use
to applications which augment Recoll functionality, and need to add
configuration data without disturbing the user's files.
Please note that the two, currently single, values will
probably be interpreted as colon-separated lists in the
future: do not use colon characters inside the directory
does not exist when recoll or recollindex are started,
it will be created with a set of empty configuration files.
give you a chance to edit the configuration file before
starting indexing. recollindex will proceed
immediately. To avoid mistakes, the automatic directory
creation will only occur for the default location, not if
RECOLL_CONFDIR were used (in the latter
cases, you will have to create the directory).
All configuration files share the same format. For example, a short extract of the main configuration file might look as follows:
# Space-separated list of directories to index. topdirs = ~/docs /usr/share/doc [~/somedirectory-with-utf8-txt-files] defaultcharset = utf-8
There are three kinds of lines:
Comment (starts with #) or empty.
Parameter affectation (name = value).
Section definition ([somedirname]).
Long lines can be broken by ending each incomplete part
with a backslash (
Depending on the type of configuration file, section definitions either separate groups of parameters or allow redefining some parameters for a directory sub-tree. They stay in effect until another section definition, or the end of file, is encountered. Some of the parameters used for indexing are looked up hierarchically from the current directory location upwards. Not all parameters can be meaningfully redefined, this is specified for each in the next section.
When found at the beginning of a file path, the tilde character (~) is expanded to the name of the user's home directory, as a shell would do.
Some parameters are lists of strings. White space is used for separation. List elements with embedded spaces can be quoted using double-quotes. Double quotes inside these elements can be escaped with a backslash.
No value inside a configuration file can contain a newline character. Long lines can be continued by escaping the physical newline with backslash, even inside quoted strings.
astringlist = "some string \ with spaces" thesame = "some string with spaces"
Parameters which are not part of string lists can't be quoted, and leading and trailing space characters are stripped before the value is used.
Encoding issues. Most of the configuration parameters are plain ASCII. Two particular sets of values may cause encoding issues:
File path parameters may contain non-ascii characters and should use the exact same byte values as found in the file system directory. Usually, this means that the configuration file should use the system default locale encoding.
parameter should be encoded in UTF-8. If your system
locale is not UTF-8, and you need to also specify
non-ascii file paths, this poses a difficulty because
common text editors cannot handle multiple encodings
in a single file. In this relatively unlikely case,
you can edit the configuration file as two separate
text files with appropriate encodings, and
concatenate them to create the complete
Defines the main configuration directory.
Locations for temporary files, in this order of
priority. The default if none of these is set is to
temporary files may be created during indexing,
mostly for decompressing, and also for processing,
e.g. email attachments.
Allow adding configuration directories with priorities below and above the user directory (see above the Configuration overview section for details).
Help for setting up external indexes. See this paragraph for explanations.
Defines replacement for the default location of
Recoll data files, normally found in, e.g.,
Defines replacement for the default location of
Recoll filters, normally found in, e.g.,
aspell program to
use for creating the spelling dictionary. The
result has to be compatible with the
libaspell which Recoll is using.
Space-separated list of files or directories to recursively index. Default to ~ (indexes $HOME). You can use symbolic links in the list, they will be followed, independantly of the value of the followLinks variable.
Files and directories which should be ignored. White space separated list of wildcard patterns (simple ones, not paths, must contain no / ), which will be tested against file and directory names. The list in the default configuration does not exclude hidden directories (names beginning with a dot), which means that it may index quite a few things that you do not want. On the other hand, email user agents like Thunderbird usually store messages in hidden directories, and you probably want this indexed. One possible solution is to have '.*' in 'skippedNames', and add things like '~/.thunderbird' '~/.evolution' to 'topdirs'. Not even the file names are indexed for patterns in this list, see the 'noContentSuffixes' variable for an alternative approach which indexes the file names. Can be redefined for any subtree.
List of name endings (not necessarily dot-separated suffixes) for which we don't try MIME type identification, and don't uncompress or index content. Only the names will be indexed. This complements the now obsoleted recoll_noindex list from the mimemap file, which will go away in a future release (the move from mimemap to recoll.conf allows editing the list through the GUI). This is different from skippedNames because these are name ending matches only (not wildcard patterns), and the file name itself gets indexed normally. This can be redefined for subdirectories.
Paths we should not go into. Space-separated list of wildcard expressions for filesystem paths. Can contain files and directories. The database and configuration directories will automatically be added. The expressions are matched using 'fnmatch(3)' with the FNM_PATHNAME flag set by default. This means that '/' characters must be matched explicitely. You can set 'skippedPathsFnmPathname' to 0 to disable the use of FNM_PATHNAME (meaning that '/*/dir3' will match '/dir1/dir2/dir3'). The default value contains the usual mount point for removable media to remind you that it is a bad idea to have Recoll work on these (esp. with the monitor: media gets indexed on mount, all data gets erased on unmount). Explicitely adding '/media/xxx' to the topdirs will override this.
Set to 0 to override use of FNM_PATHNAME for matching skipped paths.
skippedPaths equivalent specific to real time indexing. This enables having parts of the tree which are initially indexed but not monitored. If daemSkippedPaths is not set, the daemon uses skippedPaths.
Space-separated list of wildcard expressions for names that should be ignored inside zip archives. This is used directly by the zip handler, and has a function similar to skippedNames, but works independantly. Can be redefined for subdirectories. Supported by recoll 1.20 and newer. See https://bitbucket.org/medoc/recoll/wiki/Filtering%20out%20Zip%20archive%20members
Follow symbolic links during indexing. The default is to ignore symbolic links to avoid multiple indexing of linked files. No effort is made to avoid duplication when this option is set to true. This option can be set individually for each of the 'topdirs' members by using sections. It can not be changed below the 'topdirs' level. Links in the 'topdirs' list itself are always followed.
Restrictive list of indexed mime types. Normally not set (in which case all supported types are indexed). If it is set, only the types from the list will have their contents indexed. The names will be indexed anyway if indexallfilenames is set (default). MIME type names should be taken from the mimemap file. Can be redefined for subtrees.
List of excluded MIME types. Lets you exclude some types from indexing. Can be redefined for subtrees.
Size limit for compressed files. We need to decompress these in a temporary directory for identification, which can be wasteful in some cases. Limit the waste. Negative means no limit. 0 results in no processing of any compressed file. Default 50 MB.
Size limit for text files. Mostly for skipping monster logs. Default 20 MB.
Index the file names of unprocessed files Index the names of files the contents of which we don't index because of an excluded or unsupported MIME type.
Use a system command for file MIME type guessing as a final step in file type identification This is generally useful, but will usually cause the indexing of many bogus 'text' files. See 'systemfilecommand' for the command used.
Command used to guess MIME types if the internal methods fails This should be a "file -i" workalike. The file path will be added as a last parameter to the command line. 'xdg-mime' works better than the traditional 'file' command, and is now the configured default (with a hard-coded fallback to 'file')
Decide if we process the Web queue. The queue is a directory where the Recoll Web browser plugins create the copies of visited pages.
Page size for text files. If this is set, text/plain files will be divided into documents of approximately this size. Will reduce memory usage at index time and help with loading data in the preview window at query time. Particularly useful with very big files, such as application or system logs. Also see textfilemaxmbs and compressedfilemaxkbs.
Size limit for archive members. This is passed to the filters in the environment as RECOLL_FILTER_MAXMEMBERKB.
Decide if we store character case and diacritics in the index. If we do, searches sensitive to case and diacritics can be performed, but the index will be bigger, and some marginal weirdness may sometimes occur. The default is a stripped index. When using multiple indexes for a search, this parameter must be defined identically for all. Changing the value implies an index reset.
Decides if terms will be generated for numbers. For example "123", "1.5e6", 192.168.1.4, would not be indexed if nonumbers is set ("value123" would still be). Numbers are often quite interesting to search for, and this should probably not be set except for special situations, ie, scientific documents with huge amounts of numbers in them, where setting nonumbers will reduce the index size. This can only be set for a whole index, not for a subtree.
Determines if we index 'coworker' also when the input is 'co-worker'. This is new in version 1.22, and on by default. Setting the variable to off allows restoring the previous behaviour.
Decides if specific East Asian (Chinese Korean Japanese) characters/word splitting is turned off. This will save a small amount of CPU if you have no CJK documents. If your document base does include such text but you are not interested in searching it, setting nocjk may be a significant time and space saver.
This lets you adjust the size of n-grams used for indexing CJK text. The default value of 2 is probably appropriate in most cases. A value of 3 would allow more precision and efficiency on longer words, but the index will be approximately twice as large.
Languages for which to create stemming expansion data. Stemmer names can be found by executing 'recollindex -l', or this can also be set from a list in the GUI.
Default character set. This is used for files which do not contain a character set definition (e.g.: text/plain). Values found inside files, e.g. a 'charset' tag in HTML documents, will override it. If this is not set, the default character set is the one defined by the NLS environment ($LC_ALL, $LC_CTYPE, $LANG), or ultimately iso-8859-1 (cp-1252 in fact). If for some reason you want a general default which does not match your LANG and is not 8859-1, use this variable. This can be redefined for any sub-directory.
A list of characters, encoded in UTF-8, which should be handled specially when converting text to unaccented lowercase. For example, in Swedish, the letter a with diaeresis has full alphabet citizenship and should not be turned into an a. Each element in the space-separated list has the special character as first element and the translation following. The handling of both the lowercase and upper-case versions of a character should be specified, as appartenance to the list will turn-off both standard accent and case processing. The value is global and affects both indexing and querying. Examples: Swedish: unac_except_trans = ää Ää öö Öö üü Üü ßss œoe Œoe æae Æae ﬀff ﬁfi ﬂfl åå Åå . German: unac_except_trans = ää Ää öö Öö üü Üü ßss œoe Œoe æae Æae ﬀff ﬁfi ﬂfl In French, you probably want to decompose oe and ae and nobody would type a German ß unac_except_trans = ßss œoe Œoe æae Æae ﬀff ﬁfi ﬂfl . The default for all until someone protests follows. These decompositions are not performed by unac, but it is unlikely that someone would type the composed forms in a search. unac_except_trans = ßss œoe Œoe æae Æae ﬀff ﬁfi ﬂfl
Overrides the default character set for email messages which don't specify one. This is mainly useful for readpst (libpst) dumps, which are utf-8 but do not say so.
Set fields on all files (usually of a specific fs area). Syntax is the usual: name = value ; attr1 = val1 ; [...] value is empty so this needs an initial semi-colon. This is useful, e.g., for setting the rclaptg field for application selection inside mimeview.
Use mtime instead of ctime to test if a file has been modified. The time is used in addition to the size, which is always used. Setting this can reduce re-indexing on systems where extended attributes are used (by some other application), but not indexed, because changing extended attributes only affects ctime. Notes: - This may prevent detection of change in some marginal file rename cases (the target would need to have the same size and mtime). - You should probably also set noxattrfields to 1 in this case, except if you still prefer to perform xattr indexing, for example if the local file update pattern makes it of value (as in general, there is a risk for pure extended attributes updates without file modification to go undetected). Perform a full index reset after changing this.
Disable extended attributes conversion to metadata fields. This probably needs to be set if testmodifusemtime is set.
Define commands to gather external metadata, e.g. tmsu tags. There can be several entries, separated by semi-colons, each defining which field name the data goes into and the command to use. Don't forget the initial semi-colon. All the field names must be different. You can use aliases in the "field" file if necessary. As a not too pretty hack conceded to convenience, any field name beginning with "rclmulti" will be taken as an indication that the command returns multiple field values inside a text blob formatted as a recoll configuration file ("fieldname = fieldvalue" lines). The rclmultixx name will be ignored, and field names and values will be parsed from the data. Example: metadatacmds = ; tags = tmsu tags %f; rclmulti1 = cmdOutputsConf %f
Top directory for Recoll data. Recoll data directories are normally located relative to the configuration directory (e.g. ~/.recoll/xapiandb, ~/.recoll/mboxcache). If 'cachedir' is set, the directories are stored under the specified value instead (e.g. if cachedir is ~/.cache/recoll, the default dbdir would be ~/.cache/recoll/xapiandb). This affects dbdir, webcachedir, mboxcachedir, aspellDicDir, which can still be individually specified to override cachedir. Note that if you have multiple configurations, each must have a different cachedir, there is no automatic computation of a subpath under cachedir.
Maximum file system occupation over which we stop indexing. The value is a percentage, corresponding to what the "Capacity" df output column shows. The default value is 0, meaning no checking.
Xapian database directory location. This will be created on first indexing. If the value is not an absolute path, it will be interpreted as relative to cachedir if set, or the configuration directory (-c argument or $RECOLL_CONFDIR). If nothing is specified, the default is then ~/.recoll/xapiandb/
Name of the scratch file where the indexer process updates its status. Default: idxstatus.txt inside the configuration directory.
Directory location for storing mbox message offsets cache files. This is normally 'mboxcache' under cachedir if set, or else under the configuration directory, but it may be useful to share a directory between different configurations.
Minimum mbox file size over which we cache the offsets. There is really no sense in caching offsets for small files. The default is 5 MB.
Directory where we store the archived web pages. This is only used by the web history indexing code Default: cachedir/webcache if cachedir is set, else $RECOLL_CONFDIR/webcache
Maximum size in MB of the Web archive. This is only used by the web history indexing code. Default: 40 MB. Reducing the size will not physically truncate the file.
The path to the Web indexing queue. This is hard-coded in the plugin as ~/.recollweb/ToIndex so there should be no need or possibility to change it.
Aspell dictionary storage directory location. The aspell dictionary (aspdict.(lang).rws) is normally stored in the directory specified by cachedir if set, or under the configuration directory.
Directory location for executable input handlers. If RECOLL_FILTERSDIR is set in the environment, we use it instead. Defaults to $prefix/share/recoll/filters. Can be redefined for subdirectories.
Directory location for icons. The only reason to change this would be if you want to change the icons displayed in the result list. Defaults to $prefix/share/recoll/images
Threshold (megabytes of new data) where we flush from memory to disk index. Setting this allows some control over memory usage by the indexer process. A value of 0 means no explicit flushing, which lets Xapian perform its own thing, meaning flushing every $XAPIAN_FLUSH_THRESHOLD documents created, modified or deleted: as memory usage depends on average document size, not only document count, the Xapian approach is is not very useful, and you should let Recoll manage the flushes. The default value of idxflushmb is 10 MB, and may be a bit low. If you are looking for maximum speed, you may want to experiment with values between 20 and 80. In my experience, values beyond 100 are always counterproductive. If you find otherwise, please drop me a note.
Maximum external filter execution time in seconds. Default 1200 (20mn). Set to 0 for no limit. This is mainly to avoid infinite loops in postscript files (loop.ps)
Maximum virtual memory space for filter processes (setrlimit(RLIMIT_AS)), in megabytes. Note that this includes any mapped libs (there is no reliable Linux way to limit the data space only), so we need to be a bit generous here. Anything over 2000 will be ignored on 32 bits machines.
Stage input queues configuration. There are three internal queues in the indexing pipeline stages (file data extraction, terms generation, index update). This parameter defines the queue depths for each stage (three integer values). If a value of -1 is given for a given stage, no queue is used, and the thread will go on performing the next stage. In practise, deep queues have not been shown to increase performance. Default: a value of 0 for the first queue tells Recoll to perform autoconfiguration based on the detected number of CPUs (no need for the two other values in this case). Use thrQSizes = -1 -1 -1 to disable multithreading entirely.
Number of threads used for each indexing stage. The three stages are: file data extraction, terms generation, index update). The use of the counts is also controlled by some special values in thrQSizes: if the first queue depth is 0, all counts are ignored (autoconfigured); if a value of -1 is used for a queue depth, the corresponding thread count is ignored. It makes no sense to use a value other than 1 for the last stage because updating the Xapian index is necessarily single-threaded (and protected by a mutex).
Log file verbosity 1-6. A value of 2 will print only errors and warnings. 3 will print information like document updates, 4 is quite verbose and 6 very verbose.
Log file destination. Use 'stderr' (default) to write to the console.
Override loglevel for the indexer.
Override logfilename for the indexer.
Override loglevel for the indexer in real time mode. The default is to use the idx... values if set, else the log... values.
Override logfilename for the indexer in real time mode. The default is to use the idx... values if set, else the log... values.
Indexing process current directory. The input handlers sometimes leave temporary files in the current directory, so it makes sense to have recollindex chdir to some temporary directory. If the value is empty, the current directory is not changed. If the value is (literal) tmp, we use the temporary directory as set by the environment (RECOLL_TMPDIR else TMPDIR else /tmp). If the value is an absolute path to a directory, we go there.
Script used to heuristically check if we need to retry indexing files which previously failed. The default script checks the modified dates on /usr/bin and /usr/local/bin. A relative path will be looked up in the filters dirs, then in the path. Use an absolute path to do otherwise.
Additional places to search for helper executables. This is only used on Windows for now.
Length of abstracts we store while indexing. Recoll stores an abstract for each indexed file. The text can come from an actual 'abstract' section in the document or will just be the beginning of the document. It is stored in the index so that it can be displayed inside the result lists without decoding the original file. The idxabsmlen parameter defines the size of the stored abstract. The default value is 250 bytes. The search interface gives you the choice to display this stored text or a synthetic abstract built by extracting text around the search terms. If you always prefer the synthetic abstract, you can reduce this value and save a little space.
Truncation length of stored metadata fields. This does not affect indexing (the whole field is processed anyway), just the amount of data stored in the index for the purpose of displaying fields inside result lists or previews. The default value is 150 bytes which may be too low if you have custom fields.
Language definitions to use when creating the aspell dictionary. The value must match a set of aspell language definition files. You can type "aspell dicts" to see a list The default if this is not set is to use the NLS environment to guess the value.
Additional option and parameter to aspell dictionary creation command. Some aspell packages may need an additional option (e.g. on Debian Jessie: --local-data-dir=/usr/lib/aspell). See Debian bug 772415.
Set this to have a look at aspell dictionary creation errors. There are always many, so this is mostly for debugging.
Disable aspell use. The aspell dictionary generation takes time, and some combinations of aspell version, language, and local terms, result in aspell crashing, so it sometimes makes sense to just disable the thing.
Auxiliary database update interval. The real time indexer only updates the auxiliary databases (stemdb, aspell) periodically, because it would be too costly to do it for every document change. The default period is one hour.
Minimum interval (seconds) between processings of the indexing queue. The real time indexer does not process each event when it comes in, but lets the queue accumulate, to diminish overhead and to aggregate multiple events affecting the same file. Default 30 S.
Timing parameters for the real time indexing. Definitions for files which get a longer delay before reindexing is allowed. This is for fast-changing files, that should only be reindexed once in a while. A list of wildcardPattern:seconds pairs. The patterns are matched with fnmatch(pattern, path, 0) You can quote entries containing white space with double quotes (quote the whole entry, not the pattern). The default is empty. Example: mondelaypatterns = *.log:20 "*with spaces.*:30"
ionice class for the real time indexing process On platforms where this is supported. The default value is 3.
ionice class parameter for the real time indexing process. On platforms where this is supported. The default is empty.
auto-trigger diacritics sensitivity (raw index only). IF the index is not stripped, decide if we automatically trigger diacritics sensitivity if the search term has accented characters (not in unac_except_trans). Else you need to use the query language and the "D" modifier to specify diacritics sensitivity. Default is no.
auto-trigger case sensitivity (raw index only). IF the index is not stripped (see indexStripChars), decide if we automatically trigger character case sensitivity if the search term has upper-case characters in any but the first position. Else you need to use the query language and the "C" modifier to specify character-case sensitivity. Default is yes.
Maximum query expansion count for a single term (e.g.: when using wildcards). This only affects queries, not indexing. We used to not limit this at all (except for filenames where the limit was too low at 1000), but it is unreasonable with a big index. Default 10000.
Maximum number of clauses we add to a single Xapian query. This only affects queries, not indexing. In some cases, the result of term expansion can be multiplicative, and we want to avoid eating all the memory. Default 50000.
Maximum number of positions we walk while populating a snippet for the result list. The default of 1,000,000 may be insufficient for very big documents, the consequence would be snippets with possibly meaning-altering missing words.
Attempt OCR of PDF files with no text content if both tesseract and pdftoppm are installed. The default is off because OCR is so very slow.
Enable PDF attachment extraction by executing pdftk (if available). This is normally disabled, because it does slow down PDF indexing a bit even if not one attachment is ever found.
This file contains information about dynamic fields
handling in Recoll. Some
very basic fields have hard-wired behaviour, and, mostly,
you should not change the original data inside the
fields file. But you can
create custom fields fitting your data and handle them
just like they were native ones.
fields file has
several sections, which each define an aspect of fields
processing. Quite often, you'll have to modify several
sections to obtain the desired behaviour.
We will only give a short description here, you should refer to the comments inside the default file for more detailed information.
Field names should be lowercase alphabetic ASCII.
A field becomes indexed (searchable) by having a prefix defined in this section.
A field becomes stored (displayable inside results) by having its name listed in this section (typically with an empty value).
This section defines lists of synonyms for the
canonical names used inside the
This section also defines aliases for the canonic field names, with the difference that the substitution will only be used at query time, avoiding any possibility that the value would pick-up random metadata from documents.
Some input handlers may need specific
configuration for handling fields. Only the email
message handler currently has such a section (named
[mail]). It allows
indexing arbitrary email headers in addition to the
ones indexed by default. Other such sections may
appear in the future.
Here follows a small example of a personal
fields file. This would
extract a specific email header and use it as a
searchable field, with data displayable inside result
lists. (Side note: as the email handler does no decoding
on the values, only plain ascii headers can be indexed,
and only the first occurrence will be used for headers
that occur several times).
[prefixes] # Index mailmytag contents (with the given prefix) mailmytag = XMTAG [stored] # Store mailmytag inside the document data record (so that it can be # displayed - as %(mailmytag) - in result lists). mailmytag = [queryaliases] filename = fn containerfilename = cfn [mail] # Extract the X-My-Tag mail header, and use it internally with the # mailmytag field name x-my-tag = mailmytag
Recoll versions 1.19 and later process user extended file attributes as documents fields by default.
Attributes are processed as fields of the same name,
after removing the
prefix on Linux.
section of the
file allows specifying translations from extended
attributes names to Recoll field names. An empty
translation disables use of the corresponding attribute
mimemap specifies the
file name extension to MIME type mappings.
For file names without an extension, or with an
unknown one, a system command (file
-i, or xdg-mime) will be
executed to determine the MIME type (this can be switched
off, or the command changed inside the main configuration
The mappings can be specified on a per-subtree basis,
which may be useful in some cases. Example: okular notes have a
.xml extension but should be handled
specially, which is possible because they are usually all
located in one place. Example:
[~/.kde/share/apps/okular/docdata] .xml = application/x-okular-notes
mimemap variable has been
noContentSuffixes, while keeping the
same function, as of Recoll version 1.21. For older
Recoll versions, see the
noContentSuffixes but use
mimeconf specifies how
the different MIME types are handled for indexing, and
which icons are displayed in the recoll result
Changing the parameters in the [index] section is probably not a good idea except if you are a Recoll developer.
The [icons] section allows you to change the icons
which are displayed by recoll in the result
lists (the values are the basenames of the png images
directory (specified in
mimeview specifies which
programs are started when you click on an Open link in a result list. Ie: HTML is
normally displayed using firefox, but you may prefer
might be named oofice instead of
Changes to this file can be done by direct editing, or through the recoll GUI preferences dialog.
If Use desktop preferences to
choose document editor is checked in the
Recoll GUI preferences,
mimeview entries will
be ignored except the one labelled
application/x-all (which is set to use
In this case, the
xallexcepts top level variable defines a
list of MIME type exceptions which will be processed
according to the local entries instead of being passed to
the desktop. This is so that specific Recoll options such as a page number
or a search string can be passed to applications that
support them, such as the evince viewer.
As for the other configuration files, the normal usage
is to have a
inside your own configuration directory, with just the
non-default entries, which will override those from the
central configuration file.
All viewer definition entries must be placed under a
The keys in the file are normally MIME types. You can
add an application tag to specialize the choice for an
area of the filesystem (using a
localfields specification in
mimeconf). The syntax for
the key is
entry, (placed at the top level, outside of the
[view] section), holds a
list of MIME types that should not be uncompressed before
starting the viewer (if they are found compressed, ie:
The right side of each assignment holds a command to be executed for opening the file. The following substitutions are performed:
%D. Document date
%f. File name. This may be the name of a temporary file if it was necessary to create one (ie: to extract a subdocument from a container).
%i. Internal path, for subdocuments of containers. The format depends on the container type. If this appears in the command line, Recoll will not create a temporary file to extract the subdocument, expecting the called application (possibly a script) to be able to handle it.
%M. MIME type
%p. Page index. Only significant for a subset of document types, currently only PDF, Postscript and DVI files. Can be used to start the editor at the right page for a match or snippet.
%s. Search term. The value will only be set for documents with indexed page numbers (ie: PDF). The value will be one of the matched search terms. It would allow pre-setting the value in the "Find" entry inside Evince for example, for easy highlighting of the term.
In addition to the predefined values above, all
will be replaced by the value of the field named
fieldname for the document.
This could be used in combination with field
customisation to help with opening the document.
query-time path translations. These can be useful in
The file has a section for any index which needs translations, either the main one or additional query indexes. The sections are named with the Xapian index directory names. No slash character should exist at the end of the paths (all comparisons are textual). An exemple should make things sufficiently clear
[/home/me/.recoll/xapiandb] /this/directory/moved = /to/this/place [/path/to/additional/xapiandb] /server/volume1/docdir = /net/server/volume1/docdir /server/volume2/docdir = /net/server/volume2/docdir
Imagine that you have some kind of file which does
not have indexable content, but for which you would
like to have a functional Open link in the result list (when
found by file name). The file names end in
.blob and can be
displayed by application
You need two entries in the configuration files for this to work:
~/.recoll/mimemap), add the
.blob = application/x-blobapp
Note that the MIME type is made up here, and
you could call it
application/x-blobapp = blobviewer %f
We are supposing that
a file name parameter here, you would use
%u if it liked URLs
If you just wanted to change the application used by
Recoll to display a
MIME type which it already knows, you would just need
entries you add in your personal file override those in
the central configuration, which you do not need to
mimeview can also
be modified from the Gui.
Let us now imagine that the above
.blob files actually
contain indexable text and that you know how to extract
it with a command line program. Getting Recoll to index the files is easy.
You need to perform the above alteration, and also to
add data to the
file (typically in
section, add the following line (more about the
indexing script later):
application/x-blobapp = exec rclblob
section, you should choose an icon to be
displayed for the files inside the result lists.
Icons are normally 64x64 pixels PNG files which
[categories] section, you should
add the MIME type where it makes sense (you can
also create a category). Categories may be used
for filtering in advanced search.
rclblob handler should
be an executable program or script which exists inside
/usr/share/recoll/filters. It will be
given a file name as argument and should output the
text or html contents on the standard output.
The filter programming section describes in more detail how to write an input handler.