Add some tests and improve others to raise coverage percentage. This
should lead to almost 100% for the AbuseFilterParser class. Aside from
this, a couple of changes:
* Remove an unused function
* Let equals_to_any return a genuine result with empty strings
* Remove an if which will never be true in skipOverBraces, since the
function is called after checking the same conditions.
Bug: T201193
Change-Id: I7020b2ed996236c38c5784d161ad98ec44163406
Check if the entry is deleted first, since it's the strongest deletion
here (oversight level). Bonus: don't use implicit conversion when
checking the return value of SpecialAbuseLog::isHidden.
Bug: T200644
Change-Id: Ie5c4575ad29fe3dcb85a26cc74f1c59207df2852
Add an AbuseFilter log link to the subtitle of history pages.
Bug: T28934
Co-authored-by: Matěj Suchánek
Depends-On: I2e0e9e92d3fc303135b0eb9acf06b5fd120178a5
Depends-On: I58a3039b3755648bb0c8aaf87db48ace96ce9344
Change-Id: Ib89c48f2b8f3121ead184844844acee436e2fdd6
The following sniffs are failing and were disabled:
* MediaWiki.Commenting.FunctionAnnotations.UnrecognizedAnnotation
* MediaWiki.Usage.InArrayUsage.Found
Change-Id: I46e414246c6597dd78b069f753d686c0d1c1c09d
A month ago SpecialBlock::getSuggestedDurations has been
modified, and now it also returns an "other" key. Since we don't need it
and it would break thing up, add a parameter to avoid dealing with that.
Depends-On: Ic2dbc961f7eebad11da53724b9cce2f804ffad39
Change-Id: Ica37ba7015a04445c2cbafebcc85726368e23cb0
This settles almost everything, leaving the tags part ready to be
further improved in the follow-ups.
Also, replaced some fields with totally different ones, improved the
warn preview area and improved a bit nojs experience by hiding unusable
buttons.
Bug: T132284
Bug: T154749
Change-Id: I7a5caa862a32f9792140c6a4d9708a2d20472672
For the counting part I used this a relatively simple approach. It might
not be the best one, but should work without changing too much code. As
for the exception, I added it to every function which takes a single
parameter. Plus a couple of minor fixes: removed an unused function and
replaced "__METHOD__" with function names.
Bug: T198300
Change-Id: I484fe2994292970276150d2e417801453339e540
This line first used to be just an "strlen". Then we merged
Iaeae672dca66ffc745054daabd6f0eae7dfbc648 to clean input and this caused
some "undefined index" notices. These were in turn fixed in Ibebedb566da705e77ffb831ebda6476adba07c93 by adding an "empty". However, this slightly changed the range of accepted parameters, for instance refusing 0 and '0'. Those should never be used, so this is just a theoretical problem, but we'd better be consistent and simplify this line.
Change-Id: I4643d0632acf5926ac8de5da9bcb3e5dc715fdc1
Otherwise, if the user is warned for e.g. trying to move a page, and
after the warning he tries to delete it, he won't be warned again. Since
filtered action (edit, move, delete...) can be really different, we
should repeat the warning if the action changed.
Bug: T199621
Change-Id: Ia481b2bf552e16de8485c246aa5612d5bb2cd6ca
The patch adds the logid parameter to the queryAbuseLog API, so that
users will be able to retrieve a single result with the given logid.
Bug: T36731
Change-Id: I9160c3690e86ea40560f6fa7721918965234c29e
The function used to determine if a row is hidden has three possible
return values: true, false and "implicit". While the first and the
second one refer to AF own suppressing system, 'implicit' means that the
revision associated with the log entry is deleted. However, we checked
for such return value with a boolean cast, which caused true and
'implicit' to be equally treated, thus hiding revdel'ed revisions to
sysops. Bonus: fixed a comment typo.
Bug: T191699
Change-Id: I87d3a6437bb966198175e4bfd063e30ed79c345f
We still had three entries of "LogPage", which is legacy and has some
problems (I7bb0e92b2906a2511fc4290bdc76fc39ec4617fe). This patch updates
two of them to ManualLogEntry. The last one is handled separately in
Ic23e724997e4748c8d0da8138aa73d31b17b7064.
Change-Id: I2a4f18ea6baebdc114078c57d8937ce4ca2aace5
I've been noticing this problem for a long time: sometimes, when the
filter editor stays open for a long time and you try to click "save filter",
the page is scrolled and the edit isn't save (while it is indeed saved
when clicking save again). I found out that this is due to edit token
not matching. If that happens and the request was posted, warn user to
re-save the edit.
Change-Id: Id0c5600bf22632f57d237a19b492cc9c297be736
To generate an OOUI-friendly array with dropdown values, we need to
rearrange the array we already get from AbuseFilter::getBuilderValues().
Right now we do it in a pretty dirty way, which also causes errors if
external values (e.g. Flow variables) are in the list. With this patch,
such conversion is simplified, explained in a comment, and doesn't
output errors anymore.
Change-Id: I1063865aeff2dfb637e95d7b2ff30da39ceeab67
This is something that hasn't been working since January 2009, when AF
didn't have arrays and all variables were computed non-lazily. In fact,
when reverting "degroup", we used to take old groups from edit vars, but
the variable may not have been computed for such edit. Plus, we treated
the var collection as an array instead of an AbuseFilterVariableHolder
object, and exploded user_groups since it was a string. With this patch
everything should start working as intended.
Change-Id: I76917b2e331291bd42daeef8d048507dc38048cb
The message 'abusefilter-edit-notallowed' is used twice and outputted
as plain text. This makes it really, really hard to notice. Wrap it in a
block-level errorbox to make sure users see it.
Change-Id: I6e5579f9a5e33f05520001e10ffdde928ffdcff0
Introduced in I75ce47d247cf6949117370c8c78ab7c6980538f3, the message name
was misspelled in the code and thus the message doesn't show.
Change-Id: Iad515c48035259340c4824d456a14010c977e7a8
Html::warningBox makes use of Html::rawElement, where as noted in docblock the given html must *not* be escaped. Plus, bold text was broken due to escaping.
This reverts commit 7dfe4bfcfd.
Change-Id: I505be036291d4c6ff33c0c4fed4dd83a5bb56c54
This fixes the following minor issues:
* In HistoryPager's getQueryInfo, afh_id was listed twice
* In AbuseFilter::translateFromHistory a field named "af_" was produced
if no actions were in use
* The topnav link "Recent filter changes" wasn't STRONGed on pages like
"Special:AbuseFilter/history/123"
* In checkAllFilters and AbuseFilter::getFilter, select from DB only the
fields that will be used.
* Simplify some inline comments and remove superfluous ones
Change-Id: If72b18bedac5e580487406e696aea1fd172ae45b
Right now it can manually be added when creating filters. Since the
distinction is interal to AbuseFilter, we can't use hooks to achieve the
goal (the tag isn't already usable from outside AF). Also making
isAllowedTag public to make it testable.
Change-Id: I75ce47d247cf6949117370c8c78ab7c6980538f3
Trying to write unit tests, there are some things in the code that make
it not well testable. Here, two of them are corrected:
1 - Use class constants instead of static variables inside a non-static
method. Otherwise such variables won't be reset between tests. The
change is made so that there'll be less impact on blame.
2 - Set af_enabled to true even in af_deleted is true as well. For three
reasons: the first is that we already perform validation for this, so no
need to secretly change the option to whatever we think would make
sense. Second, this redundant validation makes some tests fail. Third:
this way, if the user selects both enabled and deleted, when the warning
is shown he'll indeed see that both checkboxes are selected. Before, he
would only see wpFilterEnabled as selected.
Change-Id: Ib7a0335fa7fb3b8a21765438a720205656c1ea09
Actually, it seems like I almost got it right at the first try. I tested
every validation scenario and it worked as espected, so ready for
review.
Bug: T193596
Change-Id: I7fd1798030d83292ce46543e25c0c431ec345a28
All uses of "throw" inside AbuseFilterParser are now covered.
Bonus: added a standard suppresswarning when checking regex validity.
Change-Id: Iacb8f7a361079e3e117dc6845597c7bd8473e54a
doLevel- functions are currently documented in AFPToken. This patch
copies such comments on docblocks in AbuseFilterParser, the place where
this docs can really be helpful.
Change-Id: I4e47e760a56800faa9b0a1146e0d79f8955dca9a
Otherwise, a blank page will be considered as having a newline inside,
which won't be marked as added (or removed) in the diff. This requires
introducing a new method and leaving the old one for backward
compatibility, and may cause regressions.
Bug: T74329
Change-Id: I9a2397fd849544b499cad97a383e5331471e9d73
Disabled vars can currently be overwritten by assigning them custom
values (e.g. old_text := 'foo'). However, this shouldn't be allowed to
avoid confusion.
Change-Id: I49136bf19371aee1e8068a9ae621310e1ab97c86
Old_text and old_html were disabled a long time ago. With this patch,
the user will get a custom error message if trying to use them (instead
of the unrecognisedvar one), plus they'll stop appearing in /examine and
/details, unless they were computed for the examined edit (and in that case, their description message is now restored). Lastly, added a precisation to their messages.
Bug: T190698
Change-Id: Ife168522e6b1d8eb94ebbb8a16ae8831ec1dc497
Arrays were introduced with the name "lists". While it **may** look
user-friendlier and so on, it actually uses a wrong name: lists are
different from arrays. I ran a grep and I should've replaced
every occurrence, plus everything seems to work, however a double check
wouldn't be bad.
Change-Id: I6a858f02f5dd9250ba7e1abf9c6422fd98758c9e
This way, we can specify an action and it'll return only conditions for
that specific action. This is especially thought to make results
filterable by action type.
Change-Id: I2c51b695262b132a5c7cdfab20d56e36f43c7448
Now the required need will be abusefilter-modify OR
abusefilter-view-private for /tools, /test and /examine.
Bug: T193903
Change-Id: I3f1a91a2cc1df2272e5d4099cefd7c649a0683d5
The conversion to OOUI brought some extra whitespace that creates some
problems while viewing the page, especially with specific skins. This
patch compacts four different form fields in a single one, having the
side benefit of reducing the amount of used variables.
Bug: T189425
Change-Id: I75aa83e36d12db65d8b54c76b3ea14c8c797215e
With this patch, filters which are both enabled and throttled have an
"actions automatically disabled" label together with "enabled" and the
row is displayed in red. Plus, some minor changes like removing unused
fields from sortable ones and added a comment to getQueryInfo about used
columns (the idea is that it'll be easy to understand if a given column
is already there, plus if we'll need the missing one we may just replace
it with '*').
Bug: T154206
Change-Id: Iab157d094cbf2d50e9db537535fd48243e74af0b
Most of these are accidential, obsolete from a time when PHP4 required
these & to enforce passing by reference. This is the default since PHP5.
The issue with this & is that is (in theory) allows hook handlers to
replace the object with an entirely different one. Luckily this does
not work in all cases I'm aware of. But it is confusing, semantically.
Change-Id: If1e9e2723ef96308f9b4b27377398a5e497bfe70
This is the default for many years now. Returning true is not different
from returning nothing.
I'm not touching functions that can either return true or false.
Change-Id: I6c70b8ef44f17271201a69a85301a631b32763c0
In If67035991a0835ec3edc13be4543e6b40c76c3ea I changed a couple of links
to OOUI buttons, but forgot to add one of these to the output (and to
enable OOUI as well).
Change-Id: I7dd4b554bae406bc0c8326867298302ee10b47f2
With I5e3764dbec8ac21f20c460181ae78ed73eca92f6 I introduced a function
to check that two blocks with different wordings refer to the same
duration. While that functions works good 99.9% of the time, there's a
highly unlikely but actual problem: if one of the operand is parsed at
time x and the other at time x+1 (in seconds, and this may happen even
if it gets parsed 1 ms later), the 2 durations will be considered
different and this may be annoying. With this patch I introduce another
tiny function which uses strtotime to parse a duration, but uses the
second parameter (=0) to avoid relativeness to the current time. Again,
this isn't likely to occurr, but since the fix is straightforward we'd
better do it. Also, now global durations aren't parsed at every
iteration (previously they were due to the same problem, amplified by
time distance between the first and the last iteration).
Change-Id: I11a078f298aaed9631d7f422c6b9b722d28e73cc
While the change itself is simple, the only problem here is the desing,
since we're adding even more vertical space with this.
Bug: T164108
Change-Id: Ic5373dd4f0b85dc1311d90ac165d4520ac956e68
With I91a9c5cca55e540a6c95b750579c1c369a760b15 we replaced some globals
with Config and, in doing this, we added "$config->has()" to check if a
variable was null. However, "has" will always return true even if the
value is null (it only checks if it exists), and thus we end up showing
a global abusefilter pager even if no central DB is set.
Bug: T195022
Change-Id: I751fdefd29b6af1361021d4343ba67f16c99a037
Use TextContent::normalizeLineEndings instead of manually replacing
carriage returns, plus avoid the if with a simple string cast. This also
fixes some cases where a null edit isn't counted as such due to a "\n"
in new_wikitext which isn't trimmed.
Bug: T168736
Change-Id: Idfafab3fcf7912bf0aec22700d2c0137bdd6c3c8
With the introduction of custom block durations in Ib072433d19dabae48d8514e08be9893135b5d63c, the method which generates action display was enlarged in order to provide a more readable and complete message. However, for throttling we currently have an unreadable message like "Throttle: xx, yy, zzz". This is wrong for two reasons: first, those numbers need to be deciphered; second, the first number is the filter ID which is totally unuseful here.
Change-Id: I0ec6a27ff5f37aae864dfd91161bf44f0a217ef1
They were defaulted to false with
I93ad51ffe7bee597d2d127f4c5d6b2929ffc8f7e, which broke use cases where
the page field is NOT required, nor has a 'required' => false explicitly
declared.
Bug: T194425
Change-Id: I5ab768c02a30b6d053104e590729ef22bb4e0808
Pretty self-explanatory and straightforward, since recentchanges has a
dedicated column for bot edits.
Bug: T193994
Change-Id: I76d41e082aed262640e9fff856eeb97df49633d5
With If16975dd394cfdb3c57ff263366c2fc865de362a I broke flags checkboxes,
i.e. the one for enabling/deleting/etc. a filter. In fact, I
misunderstood the way cbReadOnlyAttribute was used (a dirty way,
actually) and this caused such checkboxes not to be disabled if the user
didn't have rights to edit the filter.
Change-Id: Ibf80b54e0f620734ad7767e4769a93bbf1feccff
Like we do for built-in values. If a blacklisted variable is overridden,
it still works, but there's no reason to allow it.
Bug: T191715
Change-Id: Ia4d42ec56dc4805454b96c52c2eace1924f6536c
Quite self-explanatory. If the talk page is blocked, a sentence is
added, otherwise it remains as it is. Plus, improved the way messages
are generated and reduced their reuse.
Bug: T193692
Change-Id: I01f5113ca586b94c25e1102c73d158ebb01c5a4b
The $deadActions array is populated but never used. At first I thought
it was about actions which aren't available, but this isn't right.
Instead, it's only used to keep track of available actions which aren't
used in the current filter. Which is some data that we don't need, nor
there's nothing we may do with that.
Bug: T188181
Change-Id: Ibdfeb92ccd790c0b1a4d79b382b053b9361459f8
We used to display the checkbox to block talk without checking if
it was defined. This caused a warning and an empty space with
wgBlockAllowsUTEdit set to false.
Change-Id: I97f82633e932de7e325615473c85245a406a55ef
After Id4dc1debf0240d5b336f4d9ab5b363c240f08807, the method has been
moved, and in doing that I forgot to change this line.
Change-Id: I20caf06f2c568605bd6a90c9cf2b425cd51512e7
Like we did for other links in /diff and /histories, there are some
links that we'd better display as OOUI buttons. Also, use the Html
class' specific method to show errorboxes.
Bug: T132284
Change-Id: If67035991a0835ec3edc13be4543e6b40c76c3ea
I'd like to have this reviewed by more than one user before merging, to avoid regressions of annoying typos.
Change-Id: I91a9c5cca55e540a6c95b750579c1c369a760b15
I found these vulnerabilities while trying to setup seccheck. Although
I'm not sure whether seccheck recognised them, I'm sure that they exist
since I did manual tests, and it's possible to inject custom scripts
with these.
Change-Id: I97804be8352a1b784d483195edb29e363a0c616e
'disallow' rules with a missing first parameters are perfectly fine (and
quite common), so don't throw notices when that happens.
This broke Flow's unit tests, and caused exceptions for all api.php
edits that triggered a 'disallow' rule.
Change-Id: Ibebedb566da705e77ffb831ebda6476adba07c93
This seems like the logical way to be sure that multiple mass actions
won't be counted as one, thus reaching the conditions limit. I tried to
test this locally, but I actually had troubles to simply replicate the
issue of the reached limit in a stable manner, so I'm not totally sure.
Anyway, this shouldn't do any harm.
Bug: T193374
Change-Id: Icdc172f76705870ee502339a53e912e15a3bd31d
The variable was declared in the "if" branch but also used in the "else"
one. This caused the rules textarea to not have the readonly attribute
if the user wasn't allowed and CodeEditor wasn't installed.
Change-Id: I2bf69dc0f2d24efac41d1ac6100ed7e286e3afa4
This is taken from I6a57a28f22600aafb2e529587ecce6083e9f7da4 and makes
all the needed changes to make phan pass. Seccheck will instead fail,
but since it's not clear how to fix it (and it is non-voting), for the
moment we may merge this and enable phan on IC.
Bug: T192325
Change-Id: I77648b6f8e146114fd43bb0f4dfccdb36b7ac1ac
Follow-up of Iaeae672dca66ffc745054daabd6f0eae7dfbc648. Some actions
were still marked with red, specifically the ones with block inside. The
reason is that we stored the 'blocktalk' parameter as an emtpy string if
false, which wasn't filtered when loading request. Changing the empty
string to something different is enough to fix the problem, hopefully
without regressions. Note that this isn't retroactive and needs an edit
to become effective.
Bug: T189681
Change-Id: I7d7f0606fc23bad5ba342076066ab0e935680b3f
Introduce a new function which can be used to group multiple comparisons
in a single condition. In particular, equals_to_any(S, A, B) is the
equivalent of S === A || S === B. This is especially useful in checking
for multiple namespaces, as proposed in the Community health initiative.
Change-Id: I9dcfe303eb5e51e1882fe4a65fa876aa93db7686
I left as ToDo the checks between an array and something else. With this
patch, it'll work like PHP: the result will be true iff the comparison
is loose, the array is empty and the other operand is either false or
null.
Change-Id: Idc5cadb697ed4fc7f4856967274169f77495ed9f
I added searchEnabled in I0771fa048d21031ed1e0f8a6909213bdb869a5ed, but
forgot to pass it as parameter when there's an error with the regex.
This means that, if you try to make a search with a wrong regex, when
the page is reloaded the fields for searching aren't shown and you get a
PHP warning. Here I also added warning suppressions as usually done when
checking regex validity to avoid unnecessary PHP warnings.
Change-Id: Ibc3110c30959c99d0825e1e3d7edb1e96dd9d536
Basically, with this we always start with a functioning textarea. If JS
is enabled (and CodeEditor installed), it gets then replaced by the Ace
editor.
Bug: T192241
Change-Id: Id4dc1debf0240d5b336f4d9ab5b363c240f08807
As discussed in the task, wgTitle was used (overridden) since it was null in API
calls. However, the problem has been fixed in api.php in 2009, so we
don't need to deal with it anymore. This also means that we may remove
anything else that was added to restore the original title at the end of
the function. At last, this was the only remaining exception for PHPCS.
Bug: T178007
Change-Id: Id043c74ec8d57c5fb0ab22f54acf6a31fe6b6f06
This should fix every error with excluded rules, leaving only the one
for $wgTitle. A double check would be nice in order to avoid regressions
due to stupid mistakes.
Bug: T178007
Change-Id: I22c179f3a01d652640304b59e43fcb5b5a9abac3
IP::isInRange() can return true for invalid IPs so this can
cause false positives. Instead of letting this happen, don't
allow it in the first place.
See also Ibfe55c2ebac0fccfa8329436
Bug: T124117
Change-Id: Id10552e117ce2b231504e41627b44f8cfb0d4329
