Such searches are not useful and unnecessarily creates processing load.
The length limit 100 should cover the longest language names.
Bug: T293749
Change-Id: Ide32704cca578b9aecbce34bdcc0ac25c2a09a4d
The following sniffs are failing and were disabled:
* MediaWiki.VariableAnalysis.UnusedGlobalVariables.UnusedGlobal$wgWBClientSettings
Change-Id: Ia5423c3d7ea419b3f073f35736de7a9379d4429a
The following sniffs are failing and were disabled:
* MediaWiki.Commenting.LicenseComment.InvalidLicenseTag
The following sniffs now pass and were enabled:
* MediaWiki.Commenting.FunctionComment.MissingParamComment
Change-Id: I06e0542d737cec5e2500aad6d85f72951f8b584d
1. Do string comparison for equality early in the method so that we can
do early return if it passes.
2. Move the zero length check for string up for early return. This may
not have any significant change in performance though.
Change-Id: I86bdd612a4a31c5ebfac6bcd7687b829acc69cda
* Store prefixes and infixes separately in the data
* First match language code, then prefixes, then infixes
* Try to use suggestion either in user language or autonym first
* use formatversion=2 to avoid escaping Unicode
Using Language::fetchLanguageName might can have a small
performance impact. On the other hand there is now check
to skip languages we already found, avoiding some fuzzy
matching.
This is in a preparation for a change in jquery.uls to use
the search API more, while trying to reduce the amount of
weird autocompletion suggestions we show to the user.
Bug: T73891
Change-Id: Id94c5352d9a591969bf90144d1d2d5e758d08301
To keep the average and maximum bucket size low, I made codepoints
< 4000 more granular and code points >= 4000 less granular. This
could be tweaked further for sure to reach more even sized buckets.
Bucket stats before:
- 773 buckets
- smallest has 1 entries
- largest has 1804 entries
- median size is 66 entries
- average size is 45.394566623545 entries
Bucket stats after:
- 698 buckets
- smallest has 1 entries
- largest has 1792 entries
- median size is 16 entries
- average size is 50.272206303725 entries
Change-Id: Id62d93658117564b05294c2fe36ca7c182784859
Serialized format is no longer in style for data. PHP files can
take advantage of AutoLoader and caching so they can even be faster
than serialized files. As side bonus we can have readable diffs
for updates.
Only downside is that the file generation takes about ten lines of
ugly string manipulation.
Change-Id: If09704d1172daa13c72a308814534cac1fe9899f
With this MEÄNKELI with typos=1 finds results.
Updated test case for lowercased result. Renamed variables in test
file for clarity. Updated the default value for MW_INSTALL_PATH to
work with the default layout.
Change-Id: Id93c84d308705f55b4d2378fc8c7b7f243e1b53f
Also removed some dead code that never ran, there is no variable named
"$buckets" so it'll never have an offset.
Bug: 45327
Change-Id: I1f70ef0ec4f2434f9f072e718140ff8050b81ba3
* Update .gitignore to ignore .idea.
* Removed unused local variables.
* use local context and Message class instead of deprecated wfMsg* methods.
* Remove redundant px in CSS where possible.
* Combine CSS statements where possible.
* Replace b by strong.
Change-Id: I9d5ed7b7ce585a1c101044254bcbdfc33d42afc1
* Introduce Levenshtein algorithm
* New API param 'typos' to give number of typos allowed
* test cases
Change-Id: I22bf34d08a910d1509d7eab5adc292eadc9a7c7d
