ApiStructureTest has a lot of logic for validating Action API settings
arrays during CI. Some of that logic should be part of ParamValidator
instead.
Bug: T242887
Change-Id: I3c3d23e38456de19179ae3e5855397316b6e4c40
Depends-On: I04de72d731b94468d8a12b35df67f359382b3742
This brings significant modularization to the Action API's parameter
validation, and allows the Action API and MW REST API to share
validation code.
Note there are several changes in this patch that may affect other code;
see the entries in RELEASE-NOTES-1.35 for details.
Bug: T142080
Bug: T232672
Bug: T21195
Bug: T34675
Bug: T154774
Change-Id: I1462edc1701278760fa695308007006868b249fc
Depends-On: I10011be060fe6d27c7527312ad41218786b3f40d
Done:
* Replace LanguageConverter::newConverter by LanguageConverterFactory::getLanguageConverter
* Remove LanguageConverter::newConverter from all subclasses
* Add LanguageConverterFactory integration tests which covers all languages by their code.
* Caching of LanguageConverters in factory
* Make all tests running (hope that's would be enough)
* Uncomment the deprecated functions.
* Rename FakeConverter to TrivialLanguageConverter
* Create ILanguageConverter to have shared ancestor
* Make the LanguageConverter class abstract.
* Create table with mapping between lang code and converter instead of using name convention
* ILanguageConverter @internal
* Clean up code
Change-Id: I0e4d77de0f44e18c19956a1ffd69d30e63cf51bf
Bug: T226833, T243332
This log message was added at the request of Wikimedia's Security team
in 2016, but as far as I can tell was never enabled on Wikimedia sites.
Rather than adding more logic to avoid logspam when browsers include an
Origin header for same-orign requests, let's just get rid of it
entirely.
Bug: T243908
Change-Id: I799b17a5f4891f9e739d6b62551a5736b8a6664e
This patch tries to make assertions in tests more readable by using more
self-documenting assertions as provided by modern PHPUnit versions. Among
a few others, these two main changes are done:
* I found a lot of assertions with the expected value being the *second*
parameter. I did not changed all of them. Only some that can be replaced
with assertNull() and such.
* I try to replace all `assertTrue( is_…() )` with dedicated assertions.
Change-Id: I1fc72188fbd0edacf13886e7f9a9eacbd85f13c2
Remove the deprecated Preprocessor_DOM class, which was hard-deprecated
in 1.34. This begins to simplify parser configuration and reduce redundant
code paths, but I've left two things for cleanup in a future patch:
1. The `preprocessorClass` configuration option to the parser, exposed
in `$wgParserConf`, ServiceWiring, ParserFactory, etc. There is no reason
for this to be exposed as configurable, but I've left this clean up to a
future patch.
2. The `$wgMaxGeneratedPPNodeCount` configuration, exposed also in
ParserOptions. Only Preprocessor_DOM calculated this count, and since
we are only using Preprocessor_Hash now, this configuration has no effect.
But since this value was exposed in ParserOptions and elsewhere, I've
deprecated where needed but left this clean up to a future patch.
Bug: T204945
Change-Id: I727f003f9a42d0c92bcbcce8a8289d5af6cd1298
The setting of $pageParams in that code path was assuming the page was
specified by title, breaking when it was passed a pageid.
Bug: T241362
Change-Id: Ia5b34ee392c94e5ae7693e95b58806c1dac2b24e
This patch fixes all PHPUnit 8 compat issues in the DBless suite, aside
from assertArraySubset.
Bug: T192167
Change-Id: Iea782386509b9e579f06d63687669e14bc437fad
Done automatically using the master version of MW codesniffer and
running composer fix.
Bug: T192167
Change-Id: If6b40f515fde32ab5eff074a90e821c30c791827
The immediate use case is for testing, where some tests need to use the
PHP implementation even when wikidiff2 is installed.
Bug: T237049
Change-Id: I41dc4c0933429065d7638f518ec31f0a056afc41
This commit splits changes from Ic14f5debc53e55d6714 to reduce it to
only strictly needed things. It can be merged immediately.
Bug: T192167
Change-Id: I8c541a66ea13421dbe7fa51d197d5455cc4786eb
The main purpose of this class is to generate filenames during
tests. The reason it doesn't use fixed filenames is not because
it wants to be a fuzzy test, but because it wants to allow
running against a development wiki (outside CI) multiple
times where one test may've failed and unable to clean up after
itself, or where you may've done something manually and thus
a conflicting file exists in the local wiki from before
the DB was cloned for testing.
Using dictionary based random names is needlessly complex,
and has the additional downside of actually not avoiding
conflicts very well.
Replace all this by using a timestamp instead, which is much
more stable over time and basically will never conflict with
itself unless two tests were run on the same machine and started
in the same second (or if the clock went backwards/broken).
That seems unlikely enough to not need to support magically.
But to counter-act it a bit still, also add a 3-char random hex
in front of it. This also helps with file listings so that
when you run it three times, it's easier to see the files generated
by the same run close to each other (by not having subsequent
tests start with a mostly identical prefix).
Bug: T222416
Change-Id: Iec1d89324ff2fa53d1712796157adaf4ed34bdfe
Languages with variants no longer can override the parent's constructor
(which is now used for injecting services). Instead, they need to
override Language::newConverter().
Bug: T201405
Change-Id: I923400d61763cf1db88cb0c3f684c9c10e58032d
RandomImageGenerator::getRandomLines() is trying to select a random
subset of lines from a dictionary file, but the algorithm has a few bugs
that cause it to potentially leave entries in the returned array as null
and to be biased against the first few lines in the file.
This patch corrects the bugs to implement a Fisher–Yates (or Knuth)
shuffle,[1] run in reverse so that N does not need to be known ahead of
time and only maintaining the desired subset of the array in memory.
In local testing over many trials, it has produced output without nulls
100% of the time and the aggregate results do not have any obvious bias.
[1]: https://en.wikipedia.org/wiki/Fisher%E2%80%93Yates_shuffle
Bug: T222416
Change-Id: If136928bda2165198d3957a3536de85d465321fc
This reverts commit 5f06efb318, which
reverted 9335363789, which makes
the deprecated property AbstractBlock::mReason private.
After 9335363789, AbstractBlock::mReason is obsolete, since the block
reason is now stored as a CommentStoreComment, AbstractBlock::reason.
Change-Id: Ica0a74be90383689ca8e4cfe6d0fb25c9a5942c5
This reverts commit 9335363789.
Reason for revert: It's full of code accessing AbstractBlock::mReason
out there, see [1]. Also, it was never hard deprecated. While that may
be acceptable under some circumstances, it's definitely not OK to remove
code when there are consumers around. I'd have fixed it right now without
reverting if it were a single repo, but there's just too many.
[1] - https://codesearch.wmflabs.org/search/?q=-%3EmReason&i=nope&files=&repos=
Change-Id: I8669f502b50cff89e28dada0f65fe2b130ae9b37
AbstractBlock::setReason now accepts a string, Message or
CommentStoreComment. The CommentStoreComment is accessed via
AbstractBlock::getReasonComment.
AbstractBlock::getReason returns the reason as a string, with
the language and format consistent with how block reasons were
built before this commit. This method is deprecated, since it
makes assumptions about the language and format needed. The
deprecated mReason property is no longer public.
Doing this (and T227005) will remove the implicit dependency of
BlockManager::getUserBlock on language, which causes a recursion
error if the block is checked before the user has loaded. It also
provides a mechanism for getting the block reason in a language
specified by the caller. (This does not apply to DatabaseBlock
reasons entered via the Special:Block form, which were not and
are still not translatable.)
This commit also updates authentication classes to return the
translated reason.
Bug: T227007
Change-Id: Iec36876e930dff96a256aebbdc39cbfb331c244e
Add public, protected or private to function missing a visibility
Enable the tests folder for the phpcs sniff
Change-Id: Ibefce76ea9984c47e08c94889ea2eafca7565e2c
These are static methods that have to do with processing language names
and codes. I didn't include fallback behavior, because that would mean a
circular dependency with LocalisationCache.
In the new class, I renamed AS_AUTONYMS to AUTONYMS, and added a class
constant DEFINED for 'mw' to match the existing SUPPORTED and ALL. I
also renamed fetchLanguageName(s) to getLanguageName(s).
There is 100% test coverage for the code in the new class.
This was previously committed as 2e52f48c2e and reverted because it
depended on e4468a1d6b, which had to be reverted for performance
issues. There should be no changes other than rebasing.
Bug: T201405
Change-Id: Ifa346c8a92bf1eb57dc5e79458b32b7b26f1ee8a
