| SWISH-CONFIG(1) | SWISH-E Documentation | SWISH-CONFIG(1) |
SWISH-CONFIG - Configuration File Directives
This document lists the available configuration directives available in Swish-e.
What files Swish-e indexes and how they are indexed, and where the index is written can be controlled by a configuration file.
The configuration file is a text file composed of comments, blank lines, and configuration directives. The order of the directives is not important. Some directives may be used more than once in the configuration file, while others can only be used once (e.g. additional directives will overwrite preceding directives). Case of the directive is not important -- you may use upper, lower, or mixed case.
Comments are any line that begin with a "#".
# This is a comment
As of 2.4.3 lines may be continued by placing a backslas as the last character on the line:
IgnoreWords \
am \
the \
foo
Directives may take more than one parameter. Enclose single parameters that include whitespace in quotes (single or double). Inside of quotes the backslash escapes the next character.
ReplaceRules append "foo bar" <- define "foo bar" as a single parameter
If you need to include a quote character in the value either use a backslash to escape it, or enclose it in quotes of the other type.
Backslashes also have special meaning in regular expressions.
FileFilterMatch pdftotext "'%p' -" /\.pdf$/
This says that the dot is a real dot (instead of matching any character). If you place the regular expression in quotes then you must use double-backslashes.
FileFilterMatch pdftotext "'%p' -" "/\\.pdf$/"
Swish-e will convert the double backslash into a single backslash before passing the parameter to the regular expression compiler.
Commented example configuration files are included in the conf directory of the Swish-e distribution.
Some command line arguments can override directives specified in the configuration file. Please see also the SWISH-RUN for instructions on running Swish-e, and the SWISH-SEARCH page for information and examples on how to search your index.
The configuration file is specified to Swish-e by the "-c" switch. For example,
swish-e -c myconfig.conf
You may also split your directives up into different configuration files. This allows you to have a master configuration file used for many different indexes, and smaller configuration files for each separate index. You can specify the different configuration files when running from the command line with the "-c" switch (see SWISH-RUN), or you may include other Configuration file with the IncludeConfigFile directive below.
Typically, in a configuration file the directives are grouped together in some logical order -- that is, directives that control the source of the documents would be grouped together first, and directives that control how each document is filtered or its words index in another group of directives. (The directives listed below are grouped in this order).
The configuration file directives are listed below in these groups:
Alphabetical Listing of Directives
Directives that Control Swish
These configuration directives control the general behavior of Swish-e.
IncludeConfigFile /usr/local/swish/conf/site_config.config
This may be overridden from the command line via the "-v" switch (see SWISH-RUN).
0 = no report
1 = fatal errors
2 = errors
3 = warnings
Currently (as of 2.4.4 - early 2005) libxml2 only reports errors at level 2. The default as of 2.4.4 is "2" which should report any errors that might indicate a problem parsing a document.
The exception to this is UTF-8 to Latin-1 conversion errors are reported at level 3 (changed from 1 in 2.4.4). Although these errors indicate a problem indexing text, they are only reported at level 3 because they can be very common.
It is recommended that you index at ParserWarnLevel 3 when first starting out to see what errors and warnings are reported. Then reduce the level when you understand what documents are causing parsing problems and why.
IndexFile /usr/local/swish/site.index
<meta name="robots" content="noindex">
The default is to ignore these meta tags and index the document. This tag is described at http://www.robotstxt.org/wc/exclusion.html.
Note: This feature is only available with the libxml2 HTML parser.
Also, if you are using the libxml2 parser (HTML2 and XML2) then you can use the following comments in your documents to prevent indexing:
<!-- SwishCommand noindex -->
<!-- SwishCommand index -->
and/or these may be used also:
<!-- noindex -->
<!-- index -->
For example, these are very helpful to prevent indexing of common headers, footers, and menus.
NOTE: This following items are currently not available. These items require Swish-e to parse the configuration file while searching.
Enable alternate search syntax. Allows the usage of a basic "Altavista(c)", "Lycos(c)", etc. like search syntax. This means a search query can contain "+" and "-" as syntax parameter.
Example:
swish-e -w "+word1 +word2 -word3 word4 word5"
"+" = following word has to be in all found documents
"-" = following word may not be in any document found
" " = following word will be searched in documents
Using this config directive you can change the boolean search operators of Swish-e, e.g. to adapt these to your language. The default is: AND OR NOT
Example (german):
SwishSearchOperators UND ODER NICHT
"SwishSearchDefaultRule" defines the default Boolean operator to use if none is specified between words or phrases. The default is "AND".
The word you specify must match one of the available "SwishSearchOperators".
Example:
SwishSearchOperators UND ODER NICHT
# Make it act like a web search engine
SwishSearchDefaultRule ODER
The output of Swish-e can be defined by specifying a format string with the "-x" command line argument. Using "ResultExtFormatName" you can assign a predefined format string to a name.
Examples:
ResultExtFormatName moreinfo "%c⎪%r⎪%t⎪%p⎪<author>⎪<publishyear>\n"
Then when searching you can specify the format string's name
swish-e ... -x moreinfo ...
See the "-x" switch in
SWISH-RUN for more information about output formats.
Administrative Headers Directives
Swish-e stores configuration information in the header of the index file. This information can be retrieved while searching or by functions in the Swish-e C library. There are a number of fields available for your own use. None of these fields are required:
Examples:
IndexName "Linux Documentation"
IndexDescription "This is an index of /usr/doc on our Linux machine."
IndexPointer http://localhost/swish/linux/index.html
IndexAdmin webmaster
Document Source Directives
These directives control what documents are indexed and how they are accessed. See also Directives for the File Access method only and Directives for the HTTP Access Method Only for directives that are specific to those access methods.
The "-S" command line argument is used to select the file access method.
swish-e -c swish.config -S fs - file system
swish-e -c swish.config -S http - internal http spider
swish-e -c swish.config -S prog - external program of any type
For the fs method of access IndexDir is a space-separated list of files and directories to index. Use a forward slash as the path separator in MS Windows.
For the http method the IndexDir setting is a list of space-separated URLs.
For the prog method the IndexDir setting is a list of space-separated programs to run (which generate documents for swish to index).
You may specify more than one IndexDir directive.
Any sub-directories of any listed directory will also be indexed.
Note: While processing directories, Swish-e will ignore any files or directories that begin with a dot ("."). You may index files or directories that begin with a dot by specifying their name with "IndexDir" or "-i".
Examples:
# Index this directory an any subdirectories
IndexDir /usr/local/home/http
# Index the docs directory in current directory
IndexDir ./docs
# Index these files in the current directory
IndexDir ./index.html ./page1.html ./page2.html
# and index this directory, too
IndexDir ../public_html
For the HTTP method of access specify the URL's from which you want the spidering to begin.
Example:
IndexDir http://www.my-site.com/index.html
IndexDir http://localhost/index.html
Obviously, using the HTTP method to index is much slower than indexing local files. Be well aware that some sites do not appreciate spidering and may block your IP address. You may wish to contact the remote site before spidering their web site. More information about spidering can be found in Directives for the HTTP Access Method Only below.
For the prog method of access IndexDir specifies the path to the program(s) to execute. The external program must correctly format the documents being passed back to Swish-e. Examples of external programs are provided in the prog-bin directory.
IndexDir ./myprogram.pl
See prog for details.
Note: Not all directives work with all methods.
If the file's type is HTML or HTML2 (as set by "IndexContents" or "DefaultContents") then the file will be parsed for a HTML title and that title will be indexed. Note that you must set the file's type with "IndexContents" or "DefaultContents": ".html" and ".htm" are NOT type HTML by default. For example:
IndexContents HTML* .htm .html
If a title is found, it will still be checked for "FileRules title", and the file will be skipped if a match is found. See "FileRules".
If the file's type is not HTML, or it is HTML and no title is found, then the file's path will be indexed.
For example, this will allow searching by image file name.
NoContents .gif .xbm .au .mov .mpg .pdf .ps
Note: Using this directive will not cause files with those suffixes to be indexed. That is, if you use "IndexOnly" to limit the types of files that are indexed, then you must specify in "IndexOnly" the same suffixes listed in "NoContents".
This does not work:
# Wrong!
IndexOnly .htm .html
NoContents .gif .xbm .au .mov .mpg .pdf .ps
A "-S prog" program may set the "No-Contents:" header to enable this feature for a specific document (although it would be smarter for the "-S prog" program to simply only send the pathname or title to be indexed.
For example, you may index your files locally (with the File system indexing method), yet return a URL in search results. This directive can be used to map the file names to their respective URLs on your web server.
There are five operations you can specify: replace, append, remove, prepend, and regex They will parse the pathname in the order you've typed these commands.
This directive uses C library regex.h regular expressions.
replace "the string you want replaced" "what to change it to"
remove "a string to remove"
prepend "a string to add before the result"
append "a string to add after the result"
regex "/search string/replace string/options"
Remember, quotes are needed if an expression contains white space, and backslashes have special meaning.
Regex is an Extended Regular Expression. The first character found is the delimiter (but it's not smart enough to use matched chars such as [], (), and {}).
The replace string may use substitution variables:
$0 the entire matched (sub)string
$1-$9 returns patterns captured in "(" ")" pairs
$` the string before the matched pattern
$' the string after the matched pattern
The options change the behavior of expression:
i ignore the case when matching
g repeat the substitution for the entire pattern
Examples:
ReplaceRules replace testdir/ anotherdir/
ReplaceRules replace [a-z_0-9]*_m.*\.html index.html
ReplaceRules remove testdir/
ReplaceRules prepend http://localhost/
ReplaceRules append .html
ReplaceRules regex !^/web/(.+)/!http://$1.domain.com/!
replaces a file path:
/web/search/foo/index.html
with
http://search.domain.com/foo/index.html
ReplaceRules regex #^#http://localhost/www#
ReplaceRules prepend http://localhost/www (same thing)
# Remove all extensions from C source files
ReplaceRules remove .c # ERROR! That "." is *any char*
ReplaceRules remove \.c # much better...
ReplaceRules remove "\\.c" # if in quotes you need double-backslash!
ReplaceRules remove "\.c" # ERROR! "\." -> "." and is *any char*
The XML2, HTML2, and TXT2 parsers are currently only available when Swish-e is configured to use libxml2.
You may use XML*, HTML*, and TXT* to select the parser automatically. If libxml2 is installed then it will be used to parse the content. Otherwise, Swish-e's internal parsers will be used.
Documents that are not assigned a parser with "IndexContents" will, by default, use the HTML2 parser if libxml2 is installed, otherwise will use Swish-e's internal HTML parser. The "DefaultContents" directive may be used to assign a parser to documents that do not match a file extension defined with the "IndexContents" directive.
Example:
IndexContents HTML* .htm .html .shtml
IndexContents TXT* .txt .log .text
IndexContents XML* .xml
HTML* is the default type for all files, unless otherwise specified (and this default can be changed by the DefaultContents directive. Swish-e parses titles from HTML files, if available, and keeps track of the context of the text for context searching (see "-t" in SWISH-RUN).
If using filters (with the "FileFilter" directive) to convert documents you should include those extensions, too. For example, if using a filter to convert .pdf to .html, you need to tell Swish-e that .pdf should be indexed by the internal HTML parser:
FileFilter .pdf pdf2html
IndexContent HTML .pdf
See also Document Filter Directives.
Note: Some of this may be changed in the future to use content-types instead of file extensions. See SWISH-3.0
The XML2, HTML2, and TXT2 parsers are currently only available when Swish-e is configured to use libxml2.
You may use XML*, HTML*, and TXT* to select the parser automatically. If libxml2 is installed then it will be used to parse the content. Otherwise, Swish-e's internal parsers will be used.
Example:
DefaultContents HTML
The "DefaultContents" directive should be used when spidering, as HTML files may be returned without a file extension (such as when requesting a directory and the default index.html is returned).
Setting FileInfoCompression to "yes" will compress the index file to save disk space. This may result in longer indexing times. The default is "no".
Also see the "-e" switch in
SWISH-RUN for saving RAM during indexing.
Document Contents Directives
These directives control what information is extracted from your source documents, and how that information is made available during searching.
If "ConvertHTMLEntities" is set "no" the entities will be indexed without conversion.
NOTE: Entities within XML files and files parsed with libxml2 (HTML2) are converted regardless of this setting.
swish-e -w subjects=foo
(See also the "-t" switch in SWISH-RUN for information about context searching in HTML documents.)
The MetaNames directive is a space separated list. For example:
MetaNames meta1 meta2 keywords subjects
You may also use "UndefinedMetaTags" to specify automatic extraction of meta names from your HTML and XML documents, and also to ignore indexing content of meta tags.
META tags can have two formats in your HTML source documents:
<META NAME="meta1" CONTENT="some content">
and (if using the HTML2/libxml2 parser)
<meta1>
some content
</meta1>
But this second version is invalid HTML, and will generate a warning if ParserWarningLevel is set (libxml2 only).
And in XML documents, use the format:
<meta1>
Some Content
</meta1>
Then you can limit your search to just META meta1 like this:
swish-e -w 'meta1=(apples or oranges)'
You may nest the XML and the start/end tag versions:
<keywords>
<tag1>
some content
</tag1>
<tag2>
some other content
</tag2>
<keywords>
Then you can search in both tag2 and tag2 with:
swish-e -w 'keywords=(query words)'
Swish-e indexes all text as some metaname. The default is "swishdefault", so these two queries are the same:
swish-e -w foo
swish-e -w swishdefault=foo
When indexing HTML Swish-e indexes the HTML title as default text, so when searching Swish-e will find matches in both the HTML body and the HTML title. Swish also, by default, indexes content of meta tags. So:
swish-e -w foo
will find "foo" in the body, the title, or any meta tags.
Currently, there's no way to prevent Swish-e from indexing the title contents along with the body contents, but see "UndefinedMetaTags" for how to control the indexing of meta tags.
If you would like to search just the title text, you may use:
MetaNames swishtitle
This will index the title text separately under the built-in swish internal meta name "swishtitle". You may then search like
swish-e -w foo -- search for "foo" in title, body (and undefined meta tags)
swish-e -w swishtitle=foo -- search for "foo" in title only
In addition to swishtitle, you can limit searches to documents' path with:
MetaNames swishdocpath
Then to search for "foo" but also limit searches to documents that include "manual" or "tutorial" in their path:
swish-e -w foo swishdocpath=(manual or tutorial)
See also "ExtractPath".
MetaNames summary
MetaNameAlias summary description overview
Then all three tags will get indexed as meta tag "summary". You can then search all the fields as:
-w summary=foo
The Alias work at search time, too. So these will also limit the search to the "summary" meta name.
-w description=foo
-w overview=foo
MetaNamesRank 4 subject
MetaNamesRank 3 swishdefault
MetaNamesRank 2 author publisher
MetaNamesRank -5 wrongwords
This feature is still considered experimental. If you use it, please send feedback to the discussion list.
Example:
HTMLLinksMetaName links
Now, to limit searches to files with a link to "home.html" do this:
-w links='"home.html"'
The double quotes force a phrase search.
To make Swish-e index links as normal text, you may use:
HTMLLinksMetaName swishdefault
This feature is only available with the libxml2 HTML parser.
Example:
ImagesLinksMetaName images
Now, if you would like to find pages that include a nice image of a beach:
-w images='beach'
To make Swish-e index links as normal text, you may use:
ImageLinksMetaName swishdefault
This feature is only available with the libxml2 HTML parser.
For example, by specifying a tag name:
IndexAltTagMetaName bar
would make this markup:
<foo>
<img src="/someimage.png" alt="Alt text here">
</foo>
appear like
<foo>
<bar>Alt text here</bar>
</foo>
Then the normal rules ("MetaNames" and "PropertyNames") apply to how that text is indexed.
If you use the special tag "as-text" then
<foo>
<img src="/someimage.png" alt="Alt text here">
</foo>
simply becomes
<foo>
Alt text here
</foo>
This feature is only available when using the libxml2 parser (HTML2 and XML2).
For example, say you wish to index image links under the metaname "images".
ImageLinksMetaName images
If an image is located in http://localhost/vacations/france/index.html and "AbsoluteLinks" is set to no, then a image within that document:
<img src="beach.jpeg">
will only index "beach.jpeg".
But, if you want more detail when searching, you can enable "AbsoluteLinks" and Swish-e will index "http://localhost/vacations/france/beach.jpeg". You can then look for images of beaches, but only in France:
-w images=(beach and france)
This also means you can search for any images within France:
-w images=(france)
This feature is only available with the libxml2 HTML parser.
<person age="23">
John Doe
</person>
look like the following to swish:
<person>
<person.age>
23
</person.age>
John Doe
</person>
What happens to the text "23" will depend on the setting of "UndefinedXMLAttributes":
See also the example below "XMLClassAttribues".
XMLClassAttributes class
<person class="first">
John
</person>
<person class="last">
Doe
</person>
Will appear to Swish-e as:
<person>
<person.first>
John
</person.first>
</person>
<person>
<person.last>
Doe
</person.last>
</person>
How the data is indexed depends on "MetaNames" and "UndefinedMetaTags".
Here's an example using the following configuration which combines the two directives "XMLClassAttributes" and "UndefinedXMLAttributes".
XMLClassAttributes class
UndefinedMetaTags auto
UndefinedXMLAttributes auto
IndexContents XML2 .xml
The source XML file looks like:
<xml> <person class="student" phone="555-1212" age="102"> John </person>
<person greeting="howdy">Bill</person> </xml>
Swish-e parses as:
./swish-e -c 2 -i 1.xml -T parsed_tags parsed_text -v 0
Indexing Data Source: "File-System"
<xml> (MetaName)
<person> (MetaName)
<person.student> (MetaName)
<person.student.phone> (MetaName)
555-1212
</person.student.phone>
<person.student.age> (MetaName)
102
</person.student.age>
John
</person>
<person> (MetaName)
<person.greeting> (MetaName)
howdy
</person.greeting>
Bill
</person>
</xml>
Indexing done!
One thing to note is that the first <person> block finds a class name "student" so all metanames that are created from attributes use the combined name "person.student". The second <person> block doesn't contain a "class" so, the attribute name is combined directly with the element name (e.g. "person.greeting").
The extracted string will be indexed under the specified meta name.
See "ReplaceRules" for a description of the various pattern replacement methods, but you will use the regex method.
For example, say your file system (or web tree) was organized into departments:
/web/sales/foo...
/web/parts/foo...
/web/accounting/foo...
And you wanted a way to limit searches to just documents under "sales".
ExtractPath department regex !^/web/([^/]+)/.*$!$1!
Which says, extract out the department name (as substring $1) and index it as meta name "department". Then to limit a search to the sales department:
swish-e -w foo AND department=sales
Note that the "regex" method uses a substitution pattern, so to index only a sub-string match the entire document path in the regular expression, as shown above. Otherwise any part that is not matched will end up in the substitution pattern.
See the "ExtractPathDefault" option for a way to set a value if not patterns match.
Although unlikely, you may use more than one "ExtractPath" directive. More than one directive of the same meta name will operate successively (in order listed in the configuration file) on the path. This allows you to use regular expressions on the results of the previous pattern substitution (as if piping the output from one expression to the patter of the next).
ExtractPath foo regex !^(...).+$!$1!
ExtractPath foo regex !^.+(.)$!$1!
So, the third letter is indexed as meta name "foo" if both patterns match.
ExtractPath foo regex !^X(...).+$!$1!
ExtractPath foo regex !^.+(.)$!$1!
Now (not the "X"), if the first pattern doesn't match, the last character of the path name is indexed. You must be clear on this behavior if you are using more than one "ExtractPath" directive with the same metaname.
The document path operated on is the real path swish used to access the document. That is, the "ReplaceRules" directive has no effect on the path used with "ExtractPath".
The full path is used for each meta name if more than one "ExtractPath" directive is used. That is, changes to the path used in "ExtractPath foo" do not affect the path used by "ExtractPath bar".
For example, say your want to index each document with a metaname "department" based on the following path examples:
/web/sales/foo...
/web/parts/foo...
/web/accounting/foo...
But you are also indexing documents that do not follow that pattern and you want to search those separately, too.
ExtractPath department regex !^/web/([^/]+)/.*$!$1!
ExtractPathDefault department other
Now, you may search like this:
-w foo department=(sales) - limit searches to the sales documents
-w foo department=(parts) - limit searches to the parts documents
-w foo department=(accounting) - limit searches to the accounting documents
-w foo department=(other) - everything but sales, parts, and accounting.
This basically is a shortcut for:
-w foo not department=(sales or parts or accounting)
but you don't need to keep track of what was extracted.
Properties are useful for returning additional data from documents in search results -- this saves the effort of reading and parsing the source files while reading Swish-e search results, and is especially useful when the source documents are no longer available or slow to access (e.g. over http).
Another feature of properties is that Swish-e can use the PropertyNames for sorting the search results (see the "-s" switch).
PropertyNames author subjects
Two variations are available. "PropertyNamesCompareCase" and "PropertyNamesIgnoreCase". These tell Swish-e to either ignore or compare case when sorting results. The default for "PropertyNames" is to ignore the case.
PropertyNamesIgnoreCase subject
PropertyNamesCompareCase keyword
The defaults for "internal" properties are:
swishtitle -- ignore the case
swishdocpath -- compare case
swishdescription -- compare case
These can be overridden with "PropertyNamesCompareCase" and "PropertyNamesIgnoreCase".
PropertyNamesCompareCase swishtitle
Use of PropertyNames will increase the size of your index files, sometimes significantly. Properties will be compressed if Swish-e is compiled with zlib as described in the INSTALL manual page.
If Swish-e finds more than one property of the same name in a document the property's contents will be concatinated for strings, and a warning issues for numeric (or date) properties.
When printing properties with the swish-e binary newlines are replaced with a space character. Use the swish-e library (or SWISH::API perl module) to fetch properties without newlines replaced.
Swish-e uses strtoul(3) to convert the string into an unsigned long integer. Therefore, only positive integers can be stored.
Future versions of Swish-e may be able to store different property types (such as negative integers and real numbers). This directive may change in future releases of Swish.
Swish-e will not parse dates when indexing; you must use a timestamp.
PropertyNameAlias swishtitle title titel título titulo
Note that "swishtitle" is the built-in property used to store the title of a document, and therefore you do not need to specify it as a PropertyName before use.
If any of the property names do not exist they will be created (e.g. you do not need to define the property with PropertyNames first).
In general, this feature will only be useful when parsing HTML or XML with the libxml2 parser.
For example:
PropertyNamesMaxLength 1000 swishdescription
PropertyNameAlias swishdescription body
Is somewhat like
StoreDescription HTML <body> 1000
StoreDescription XML <body> 1000
StoreDescription HTML2 <body> 1000
StoreDescription XML2 <body> 1000
but StoreDescription allows setting the tag for each parser type.
PropertyNamesMaxLength 1000 headings
PropertyNameAlias headings h1 h2 h3 h4
collects all the heading text into a single property called "headings", not to exceed 1000 characters.
This setting is used when sorting during indexing, and perhaps when sorting while searching. It also effects the order when limiting to a range of values with the -L option.
If "PreSortedIndex" it is not present in the config file (default action), all the properties will be presorted at indexing time. If it is present without any parameter, no properties will be presorted. Otherwise, only the property names specified will be presorted.
For example, if you only wish to sort results by a property called "title":
PropertyNames title age time
PreSortedIndex title
The document type (XML, HTML and TXT) must match the document type currently being indexed as set by "IndexContents" or "DefaultContents". See those directives for possible values. A common problem is using "StoreDescription" yet not setting the document's type with "IndexContents" or "DefaultContents". Another problem is different types:
IndexContents HTML2 .html
StoreDescription HTML <body>
Then .html documents are assigned a type of HTML2 (and parsed by the libxml2 parser), but the description will not be stored since it is type HTML instead of HTML2.
For text documents you specify the type TXT (or TXT2 or TXT*) and the number of characters to capture.
StoreDescription TXT 20
The above stores only the first twenty characters from the text file in the Swish-e index file.
For HTML, and XML file types, specify the tag to use for the description, and optionally the number of characters to capture. If not specified will capture the entire contents of the tag.
StoreDescription HTML <body> 20000
StoreDescription XML <desc> 40
Again, note that documents must be assigned a document type with "IndexContents" or "DefaultContents" to use this feature.
Swish-e will compress the descriptions (or any other large property) if compiled to use zlib (see INSTALL). This is recommended when using StoreDescription and a large number of documents. Compression of 30% to 50% is not uncommon with HTML files.
The default depends on the default setting compiled with zlib, but is typically six.
This option is useful when using "StoreDescription" to store a large amount text in properties (or if using "PropertyNames" with large property sizes).
Properties must be over a value defined in config.h (100 is the default) before compression will be attempted. Swish-e will never store the results of the compression if the compressed data is larger than the original data.
This option is only available when Swish-e is compiled with zlib support.
Example:
TruncateDocSize 10000000
The default is zero, which means read all data.
Warning: If you use TruncateDocSize, use it with care! TruncateDocSize is a safety belt only, to limit e.g. filteroutput, when accessing databases, or to limit "runnaway" filters. Truncating doc input may destroy document structures for Swish-e (e.g. swish may miss closing tags for XML or HTML documents).
TruncateDocSize does not currently work with the "prog" input source method.
It's a good idea to create both a normal index and a fuzzy index and allow your search interface select which index to use. Many people find the fuzzy searches to be too fuzzy.
The available fuzzy indexing options can be displayed by running
swish-e -T LIST_FUZZY_MODES
Available options include:
The installed stemmers can be viewed by running
swish-e -T LIST_FUZZY_MODES
For example, to use the Spanish stemming module:
FuzzyIndexingMode Stemming_es
Selects the legacy Swish-e English stemmer.
This is deprecated in favor of the Snowball English stemmer Stemming_en1.
Words are converted using the Porter stemming algorithm.
From: http://www.tartarus.org/~martin/PorterStemmer/
The Porter stemming algorithm (or Porter stemmer) is a
process for removing the commoner morphological and inflexional
endings from words in English. Its main use is as part of a
term normalisation process that is usually done when setting up
Information Retrieval systems.
This will help a search for "running" to also find "run" and "runs", for example.
The stemming function does not convert words to their root, rather programmatically removes endings on words in an attempt to make similar words with different endings stem to the same string of characters. It's not a perfect system, and searches on stemmed indexes often return curious results. For example, two entirely different words may stem to the same word.
Stemming also can be confusing when used with a wildcard (truncation). For example, you might expect to find the word "running" by searching for "runn*". But this fails when using a stemmed index, as "running" stems to "run", yet searching for "runn*" looks for words that start with "runn".
Soundex should not be used to search for sound-alike words. Metaphone would be more appropriate for generic sound matching of words. Soundex should only be used where you need to search multiple documents for proper names which sound similar. This is primarily used for indexing genealogical records. This may be useful for indexing other collections of data consisting mostly of names. Many common name variations are matched by Soundex. The only notable exception is the first letter of the name. The first letter is not matched for sound.
From: http://aspell.sourceforge.net/metaphone/
Lawrence Philips' Metaphone Algorithm is an algorithm which returns
the rough approximation of how an English word sounds.
The "DoubleMetaphone" mode will sometimes generate two different metaphones for the same word. This is supposed to be useful when a word may be pronounced more than one way.
A metaphone index should give results somewhere in between Soundex and Stemming.
UseStemming no
UseStemming yes
When UseStemming is set to "yes" every word is stemmed before placing it in to the index.
This option is deprecated. It has been superceded by "FuzzyIndexingMode".
This option is deprecated. It has been superceded by "FuzzyIndexingMode".
IgnoreTotalWordCountWhenRanking no
The default was changed from no to yes in version 2.2.
NOTE: must be set to no if you intend to use the -R 1 option when searching.
MinWordLimit 5
When indexing Swish-e uses WordCharacters to split up the document into words. Words are defined by any string of non-blank characters that contain only the characters listed in WordCharacters. If a string of characters includes a character that is not in WordCharacters then the word will be spit into two or more separate words.
For example:
WordCharacters abde
Would turn "abcde" into two words "ab" and "de".
Next, of these words, any characters defined in IgnoreFirstChar are stripped off the start of the word, and IgnoreLastChar characters are stripped off the end of the word. This allows, for example, periods within a word (www.slashdot.com), but not at the end of a word. Characters in IgnoreFirstChar and IgnoreLastChar must be in WordCharacters.
Finally, the resulting words MUST begin with one of the characters listed in BeginCharacters and end with one of the characters listed in EndCharacters. BeginCharacters and EndCharacters must be a subset of the characters in WordCharacters. Often, WordCharacters, BeginCharacters and EndCharacters will all be the same.
Note that the same process applies to the query while searching.
Getting these settings correct will take careful consideration and practice. It's helpful to create an index of a single test file, and then look at the words that are placed in the index (see the "-v 4", "-D" and "-k" searching switches).
Currently there is only support for eight-bit characters.
Example:
WordCharacters .abcdefghijklmnopqrstuvwxyz
BeginCharacters abcdefghijklmnopqrstuvwxyz
EndCharacters abcdefghijklmnopqrstuvwxyz
IgnoreFirstChar .
IgnoreLastChar .
So the string
Please visit http://www.example.com/path/to/file.html.
will be indexed as the following words:
please
visit
http
www.example.com
path
to
file.html
Which means that you can search for "www.example.com" as a single word, but searching for just "example" will not find the document.
Note: when indexing HTML documents HTML entities are converted to their character equivalents before being processed with these directives. This is a change from previous versions of Swish-e where you were required to include the characters "0123456789&#;" to index entities. See also "ConvertHTMLEntities"
Buzzwords are case insensitive.
Buzzwords should be separated by spaces and may span multiple directives. If the special format "File:filename" is used then the Buzzwords will be read from an external file during indexing.
Examples:
Buzzwords C++ TCP/IP
Buzzwords File: ./buzzwords.lst
If a Buzzword contains search operator characters they must be backslashed when searching. For example:
Buzzwords C++ TCP/IP web=http
./swish-e -w 'web\=http'
Buzzwords are found by splitting the text on whitespace, removing "IgnoreFirstChar" and "IgnoreLastChar" characters from the word, and then comparing with the list of "Buzzwords". Therefore, if adding "Buzzwords" to an index you will probably want to define "IgnoreFirstChar" and "IgnoreLastChar" settings.
Note: Buzzwords specific settings for "IgnoreFirstChar" and "IgnoreLastChar" may be used in the future.
Enabling this option can reduced the size of the index file, but at the expense of slower wildcard search times.
The default changed from YES to NO starting with version 2.4.3.
Words should be separated by spaces and may span multiple directives. If the special format "File:filename" is used then the stop words will be read from an external file during indexing.
In previous versions of Swish-e you could use the directive
IgnoreWords swishdefault - obsolete!
to include a default list of compiled in stopwords. This keyword is no longer supported.
Examples:
IgnoreWords www http a an the of and or
IgnoreWords File: ./stopwords.de
You can specify a list of words following the directive (you may specify more than one "UseWords" directive in a config file), and/or use the "File:" form to specify a path to a file containing the words:
UseWords perl python pascal fortran basic cobal php
UseWords File: /path/to/my/wordlist
Please drop the Swish-e list a note if you actually use this feature. It may be removed from future versions.
IgnoreLimit 50 1000
Swish-e must do extra processing to adjust the entire index when this feature is used. It is recommended that instead of using this feature that you decided what words are stopwords and add them to IngoreWords in your configuration file. To do this, use IgnoreLimit one time and note the stop words that are found while indexing. Add this list to IgnoreWords, and then remove IgnoreLimit from the configuration file.
This option is useful to avoid indexing specific data from a file. For example:
<person>
<first_name>
William
</first_name> <last_name>
Shakespeare
</last_name> <updated_date>
April 25, 1999
</updated_date>
</person>
In the above example you might not want to index the updated date, and therefore prevent finding this record by searching
-w 'person=(April)'
This is solved by:
IgnoreMetaTags updated_date
See also "UndefinedMetaTags".
This experimental feature can be used to define a set of characters that describe a number. If a word is found to contain only those characters it will not be indexed. The characters listed must be part of "WordCharacters" settings. In other words, the "word" checked is a word that Swish-e would otherwise index.
For example,
IgnoreNumberChars 0123456789$.,
Then Swish-e would not index the following:
123
123,456.78
$123.45
You might be tempted to avoid indexing hex numbers with:
IgnoreNumberChars 0123456789abcdef
which will not index 0D31, but will also not index the word "bad".
This is an experimental feature that may change in future versions. One possible change is to use regular expressions instead.
IndexComments yes
Note: This is a change in the default behavior prior to version 2.2.
For example:
# This will index a_b as a-b and ámo as amo
TranslateCharacters _á -a
"TranslateCharacters :ascii7:" is a predefined set of characters that will translate eight bit characters to ascii7 characters. Using the :ascii7: rule will translate "Ääç" to "aac". This means: searching "Çelik", "çelik" or "celik" will all match the same word.
TranslateCharacters is done early in the indexing process, after converting HTML entities but before splitting the input text into words based on WordCharacters. So characters you are translating from do not need to be listed in word characters.
The same character translations take place when searching.
For example, if you have a tag:
<subjects>
computer programming ⎪ apple computers
</subjects>
You might want to prevent matching "programming apple" in that meta name.
BumpPositionCounterCharacters ⎪
There is no default, and you may list a string of characters.
For example,
<person>
<first_name>
William
</first_name>
<last_name>
Shakespeare
</last_name>
<updated_date>
April 25, 1999
</updated_date>
</person>
In the configuration file:
DontBumpPositionOnEndTags first_name
DontBumpPositionOnStartTags last_name
This configuration allows this phrase search
-w 'person=("william shakespeare")'
but this phrase search will fail
-w 'person=("shakespeare april")'
Directives for the File Access method only
Some directives have different uses depending on the source of the documents. These directives are only valid when using the File system method of indexing.
# Only index .html .htm and .q files
IndexOnly .html .htm .q
"IndexOnly" checks that the file end in the characters listed. It does not check "extensions". "IndexOnly" is tested right before "FileRules" is processed.
FollowSymLinks no
FollowSymLinks yes
Note that when set to "no" extra stat(2) system calls must be made for each file. For large number of files you may see a small reduction in indexing time by setting this to "yes".
See also the "-l" switch in SWISH-RUN.
Except for "FileRules title ...", this feature is only available for file access method (-S fs), which is the default indexing mode. Also, any pathname modification with "ReplaceRules" happens after the check for "FileRules". (It's unlikely that you would exclude files with "FileRules" based on text you added with "ReplaceRules"!)
The regular expression is a C regex.h extended regular expression. You may supply more than one regular expression per line, or use separate directives. Preceding the regular expression with the word "not" negates the match.
The regular expression is compared against [type] as described below.
For historical reasons, you can specify "contains" or "is". "is" simply forces the regular expression to match at the start and end of the string (by internally prepending "^" and appending "$" to the regular expression).
The "regex" option requires delimiter characters:
FileRules title regex /^private/i
The only advantage of "regex" is if you want to do case insensitive matches, or simply like your regular expressions to look like perl regular expressions. You must use matching delimiters; (), {}, and [], are not currently supported for no good reason other than laziness.
Use quotes (" or ') around a pattern if it contains any white space. Note that the backslash character becomes the escape character within quotes.
For example, these sets generate the same regular expressions.
FileRules title is hello
FileRules title contains ^hello$
FileRules title regex /^hello$/
These all need quotes due to the included space character
FileRules title is "hello there"
FileRules title contains "^hello there$"
FileRules title regex "!^hello there$!"
These show how the backslash must be doubled inside of quotes. Swish-e converts a double-backslash into a single backslash, and then passes that single onto the regular expression compiler.
FileRules filename regex /\.pdf/
FileRules filename regex "/\\.pdf/"
FileRules filename regex !hello\\there! # need double for real backslash
FileRules filename regex "!hello\\\\there!" # need double-double inside of quotes
Matching Types
The following types of match strings my be supplied:
FileRules pathname
FileRules dirname
FileRules filename
FileRules directory
FileRules title
FileMatch pathname
FileMatch filename
FileMatch dirname
FileMatch directory
pathname matches the regular expression against the current pathname. The pathname may or may not be absolute depending on what you supplied to "IndexDir".
Example:
# Don't index paths that contain private or hidden
FileRules pathname contains (private⎪hidden)
# Same thing
FileRules pathname regex /(private⎪hidden)/
# Don't index exe files
FileRules pathname contains \.exe$
dirname and filename split the path name by the last delimiter character into a directory name, and a file name. Then these are compared against the patterns supplied. Directory names do not have a trailing slash. All path names use the forward slash as a delimiter within Swish-e.
Example:
# Same as last example - don't index *.exe files.
FileRules filename contains \.exe$
# Don't index any file called test.html files
FileRules filename contains ^test\.html$
# Same thing
FileRules filename is test\.html
# Don't index any directories that contain "old" (/usr/local/myold/docs)
FileRules dirname contains old
# Don't index any directories that contain the path segment "old" (/usr/local/old/foo)
FileRules dirname contains /old/
# Index only .htm, .html, plus any all-digit file names
IndexOnly .htm .html
FileMatch filename contains ^\d+$
# Same as previous, but maybe a little slower
FileRules filename regex not !\.(htm⎪html)$!
FileMatch filename contains ^\d+$
Swish-e checks these settings in the order of "pathname", "dirname", and "filename", and "FileMatch" patterns are checked before "FileRules", in general. This allows you to exclude most files with "FileRules", yet allow in a few special cases with "FileMatch". For example:
# Exclude all files of .exe, .bin, and .bat
FileRules filename contains \.(exe⎪bin⎪bat)$
# But, let these two in
FileMatch filename is baseball\.bat incoming_mail\.bin
# Same, but as a single pattern
FileMatch filename is (baseball\.bat⎪incoming_mail\.bin)
The "directory" type is somewhat unique. When Swish-e recurses into a directory it will compare all the files in the directory with the pattern and then decide if that entire directory should or should not be indexed (or recursed). Note that you are matching against file names in a directory -- and some of those names may be directory names.
A "FileRules directory" match will cause Swish-e to ignore all files and sub-directories in the current directory.
Warning: A match with "FileMatch directory" says to index everything in the *current* directory and ignore any FileRules for this directory.
Example:
# Don't index any directories (and sub directories) that contain
# a file (or sub-directory) called "index.skip"
FileRules directory contains ^index\.skip$
# Don't index directories that contain a .htaccess file.
FileRules directory contains ^\.htaccess
Note: While processing directories, Swish-e will ignore any files or directories that begin with a dot ("."). You may index files or directories that begin with a dot by specifying their name with "IndexDir" or "-i".
"title" checks for a pattern match in an HTML title.
Example:
FileRules title contains construction example pointers
# This example says to ignore case
FileRules title regex "/^Internal document/i"
Note: "FileRules title" works for any input method (fs, prog, or http) that is parsed as HTML, and where a title was found in the document.
In case all this seems a bit confusing, processing a directory happens in the following order.
First the directory name is checked:
FileRules dirname - reject entire directory if matches
Next the directory is scanned and each file name (which might be the name of a sub-directory) is checked:
FileRules directory - reject entire dir if *any* files match
FileMatch directory - accept entire dir if *any* files match
Then, unless "FileMatch directory" matched, each file is tested with FileMatch. A match says to index the file without further testing (i.e. overrides FileRules and IndexOnly):
FileMatch pathname \
FileMatch dirname - file is accepted if any match
FileMatch filename /
otherwise
IndexOnly - file is checked for the correct file extension
FileRules pathname \
FileRules dirname - file is rejected if any match
FileRules filename /
finally, the file is indexed.
Files (not directories) listed with "IndexDir" or "-i" are processed in a similar way:
FileMatch pathname \
FileMatch dirname - file is accepted if any match
FileMatch filename /
otherwise, the file is rejected if it doesn't have the correct extension or a FileRules matches.
IndexOnly - file is checked for the correct file extension
FileRules pathname \
FileRules dirname - file is rejected if any match
FileRules filename /
Note: If things are not indexing as you expect, create a
directory with some test files and use the "-T
regex" trace option to see how file names are checked. Start
with very simple tests!
Directives for the HTTP Access Method Only
The HTTP Access method is enabled by the "-S http" switch when indexing. It works by running a Perl program called SwishSpider which fetches documents from a web server.
Only text files (content-type of "text/*") are indexed with the HTTP Access Method. Other document types (e.g. PDF or MSWord) may be indexed as well. The SwishSpider will attempt to make use of the SWISH::Filter module (included with the Swish-e distribution) to convert documents into a format that Swish-e can index.
Note: The -S prog method of spidering (using spider.pl) can be a replacement for the -S http method. It offers more configuration options and better spidering speed.
These directives below are available when using the HTTP Access Method of indexing.
MaxDepth 5
Note: The default was changed from 5 to 0 in release 2.4.0
Delay 1
Note: The default was changed from 60 to 5 seconds in release 2.4.0
TmpDir /tmp/swish
If this directory does not exist or is not writable Swish-e will fail with an error during indexing.
Note, the environment variables of "TMPDIR", "TMP", and "TEMP" (in that order) will override this setting.
SpiderDirectory /usr/local/swish
EquivalentServer http://library.berkeley.edu http://www.lib.berkeley.edu
EquivalentServer http://sunsite.berkeley.edu:2000 http://sunsite.berkeley.edu
Directives for the prog Access Method Only
This section details the directives that are only available for the "prog" document source feature of Swish-e. The "prog" access method runs an external program that "feeds" documents to Swish-e. This allows indexing and filtering of documents from any source.
See prog - general purpose access method in the SWISH-RUN man page for more information.
A number of example programs for use with the "prog" access method are provided in the prog-bin directory. Please see those example if you have questions about implementing a "prog" input program.
SwishProgParameters /path/to/config hello there
IndexDir /path/to/program.pl
Then running:
swish-e -c config -S prog
Swish-e will execute "/path/to/program.pl" and pass "/path/to/config hello there" as three command line arguments to the program. This directive makes it easy to pass settings from the Swish-e configuration file to the external program.
For example, the "spider.pl" program (included in the "prog-bin" directory) uses the "SwishProgParameters" to specify what file to read for configuration information.
SwishProgParameters spider.config
IndexDir ./spider.pl
The "spider.pl" program also has a default action so you can avoid using a configuration file:
SwishProgParameters default http://www.swishe.org/ http://some.other.site/
IndexDir ./spider.pl
And the spider program will use default settings for spidering those sites.
Swish-e can read documents from standard input, so another way to run an external program with parameters is:
./spider.pl spider.conf ⎪ ./swish-e -S prog -i stdin
Notes when using MS Windows
You should use unix style path separators to specify your external program. Swish will convert forward slashes to backslashes before calling the external program. This is only true for the program name specified with "IndexDir" or the "-i" command line option.
In addition, Swish-e will make sure the program specified actually exists, which means you need to use the full name of the program.
For example, to run the perl spider program spider.pl you would need a Swish-e configuration file such as:
IndexDir e:/perl/bin/perl.exe
SwishProgParameters prog-bin/spider.pl default http://swish-e.org
and run indexing with the command:
swish-e -c swish.cfg -S prog -v 9
The "IndexDir" command tells Swish-e the name of the program to run. Under unix you can just specify the name of the script, since unix will figure out the program from the first line of the script.
The "SwishProgParameters" are
the parameters passed to the program specified by
"IndexDir" (perl.exe in this case). The
first parameter is the perl script to run (prog-bin/spider.pl). Perl
passes the rest of the parameters directly to the perl script. The second
parameter default tells the spider.pl program to use default
settings for spidering (or you could specify a spider config file -- see
"perldoc spider.pl" for details), and
lastly, the URL is passed into the spider program.
Document Filter Directives
Internally, Swish-e knows how to parse only text, HTML, and XML documents. With "filters" you can index other types of documents. For example, if all your web pages are in gzip format a filter can uncompress these on the fly for indexing.
You may wish to read the Swish-e FAQ question on filtering before continuing here. How Do I filter documents?
There are two suggested methods for filtering.
Filtering with SWISH::Filter
The Swish-e distribution includes a Perl module called SWISH::Filter and individual filters located in the filters directory. This system uses plug-in filters to extend the types of documents that Swish-e can index. The plug-in filters do not actually do the filtering, but rather provide a standard interface for accessing programs that can filter or convert documents. The programs that do the filtering are not part of the Swish-e distribution; they must be downloaded and installed separately.
The advantage of this method is that new filtering methods can be installed easily.
This system is designed to work with the -S http and -prog methods, but may also be used with the "FileFilter" feature and -S fs indexing method. See $prefix/share/doc/swish-e/examples/filter-bin/swish_filter.pl for an example.
See the filters/README file for more information.
Filtering with the FileFilter feature
A filter is an external program that Swish-e executes while processing a document of a given type. Swish-e will execute the filter program for each file that matches the file suffix (extension) set in the FileFilter or FileFilterMatch directives. FileFilterMatch matches using regular expressions and is described below.
Filters may be used with any type of input method (i.e. -S fs, -S http, or -S prog). But because
Swish-e calls the external program passing as default arguments:
Swish-e can also pass other parameters to the filter program. These parameters can be defined using the FileFilter or FileFilterMatch directives. See Filter Options below.
The filter program must open the file, process its contents, and return it to Swish-e by printing to STDOUT.
Note that this can add a significant amount of time to the indexing process if your external program is a perl or shell script. If you have many files to filter you should consider writing your filter in C instead of a shell or perl script, or using the "prog" Access Method along with SWISH::Filter.
This is the path to a directory where the filter programs are stored. Swish-e looks in this directory to find the filter specified in the FileFilter directive.
This directive is not needed if the filter program can be found in your system's path. Even if your filter is not in your system's path you can specify the full path to the filter in the FileFilter or FileFilterMatch directives.
Example:
FilterDir /usr/local/swish/filters
On systems that have a working fork(2) system call the filter program is run by forking swish then executing the filter. This mean the shell is not used for running the filter and no arguments are passed through the shell.
On other systems (e.g. Windows) the arguments are double-quoted and popen(3) is used to run the program. This does pass argument though the shell and may be a security concern depending on the abilities of the shell.
Filter options:
Filter options are a string passed as arguments to the filter-prog. Filter options can contain variables, replaced by Swish-e. If you omit filter-options Swish-e will use default parameters for the options listed above.
Default: %p %P
Which means: pass "workfile path" and "documentfile path" to filter.
Variables in filter options:
%% = %
%P = Full document pathname (e.g. URL, or path on filesystem)
%p = Full pathname to work file (maybe a tmpfile or the real document path on filesystem)
%F = Filename stripped from full document pathname
%f = Filename stripped from "work" pathname
%D = Directoryname stripped from full document pathname
%d = Directoryname stripped from full "work" pathname
Examples of strings passed:
%P = document pathname: http://myserver/path1/mydoc.txt
%p = work pathname: /tmp/tmp.1234.mydoc.txt
%F = mydoc.txt
%f = tmp.1234.mydoc.txt
%D = http://myserver/path1
%d = /tmp
Notes when using MS Windows
Windows uses double quotes to escape shell metacharacters, so if you need to use quotes then use single quotes around the entire option string.
FileFiler .mydoc mydocfilter.exe '--title "text with spaces"'
You can specify the filter program using forward slashes (unix style). Swish will convert the slashes to backslashes before running your program.
FileFilter .mydoc c:/some/path/mydocfilter.exe '-d "%d" -example -url "%P" "%f"'
Examples of filters:
FileFilter .doc /usr/local/bin/catdoc "-s8859-1 -d8859-1 %p"
FileFilter .pdf pdftotext "%p -"
FileFilter .html.gz gzip "-c %p"
FileFilter .mydoc "/some/path/mydocfilter" "-d %d -example -url %P %f"
The above examples are running a binary filter program. For more complicated filtering needs you may use a scripting language such as Perl or a shell script. Here's some examples of calling a shell and perl script:
FileFilter .pdf pdf2html.sh
FileFilter .ps ghostscript-filter.pl
Using a scripting language (or any language that has a large startup cost) can greatly increase the indexing time. For small indexing jobs, this may not be an issue, but for large collections of files that require processing by a scripting language, you may be better off using the "-S prog" access method where the script will only be compiled once, instead of for each document.
Filters are probably easier to write than a "-S prog" program. Which you decide to use depends on your requirements. Examples of filter scripts can be found in the filter-bin directory, and examples of "-S prog" programs can be found in the prog-bin directory.
Examples:
FileFilterMatch ./pdftotext "%p -" /\.pdf$/
Note that will also match a file called ".pdf", so you may want to use something that requires a filename that has more than just an extension. For example:
FileFilterMatch ./pdftotext "%p -" /.\.pdf$/
To specify more than one extension:
FileFilterMatch ./check_title.pl "%p" /\.html$/ /\.htm$/
Or a few ways to do the same thing:
FileFilterMatch ./check_title.pl %p /\.(html⎪html)$/
FileFilterMatch ./check_title.pl %p /\.html?$/
And to ignore case:
FileFilterMatch ./check_title.pl %p /\.html?$/i
You may also precede an expression with "not" to negate regular expression that follow. For example, to match files that do not have an extension:
FileFilterMatch ./convert "%p %P" not /\..+$/
$Id: SWISH-CONFIG.pod 1846 2006-10-20 20:18:30Z whmoseley $
.
| 2009-04-04 | 2.4.7 |