I have reproduced at least one instance of the message "ParserOutput
does not specify a language and no page language set in helper." by
trying to query a rest.php html endpoint on a module with the
scribunto-doc-page-does-not-exist message deactivated (which is the case
on wikidata).
I do not know if that path is the one that triggers the current large
amount of these messages coming from Wikidata, but it looks reasonable
to fix that one at least.
I'm also not sure this fix is the right one - I'll let people with more
Scribunto and ContentHandler experience chime on that :)
Bug: T362312
Change-Id: I26c1683c803f5cafd08de83e8b428031414d1c4e
Mostly unspecific arrays that can be made more specific. This patch
intentionally doesn't touch any code, only comments.
Change-Id: I5b4690e6160c6c543f9fcb2f62c41f6329d48bdb
Returning true is the same as returning nothing. It's only meaningful
when a hook handler can also return false. Some actually do this.
I'm not touching these.
See Icccf60b for the reasoning why the added `@return void` are
beneficial.
Change-Id: I6de7addee853ff183058e6c84e87a5b275c785e8
No other access in the Title class contains a `get` prefix, so this
makes the title methods consistent.
Bug: T373047
Change-Id: I4a3a7498462b7b3b6143dc61f529e526ccb112e2
Added `categories` property in the lua title class which fetches page
categories. Analogous to getContent() which fetches page content. This
increments the expensive function count.
This enables category-specific editnotices (T85372) and other use-cases.
Bug: T50175
Change-Id: Ie8e0762c537374e6086abe9f9916b8200582776a
Made the earlier semantically-correct Hindi localization an alias
and added the suggested Hindi localization preferred by the Hindi
Wikipedia community members the primary Hindi localization.
Bug: T366235
Followup-To: I811594311abb4e1d3e44ce3b284d11f72836cbdd
Change-Id: I2f8c7b1dfe22f210773507e86f4b12edd3fd196b
When modules are first loaded by `require` in package.lua, it checks
for a return value with `if res then`. If that check fails, the module
will be cached with the default value `true` at line 109, and that
value is then returned.
That check should be changed to `if res ~= nil then`, because
otherwise a module with a return value of `false` will be cached with
the value `true`.
This also necessitates a corresponding change to the check at line 93,
which is when `package.loaded` is initially checked to see if a package
has been loaded before, as that value could be `false`.
Bug: T362045
Change-Id: Id42e26a922c13971c25ed8654e822dc6fdcb6ca7
`mw.clone` is called on the global table for each new `{{#invoke:}}`,
which is a major contributor to loading times on large pages. This
change introduces various speed optimisations that make it around
15% faster:
* `recursiveClone` is declared outside of the main function, to avoid
generating a new closure each time `mw.clone` is called.
* This means `tableRefs` has to be passed as an argument to
`recursiveClone` instead of being an upvalue, which also decreases
access time.
* `recursiveClone` is only called after a `type` then `tableRefs` check
confirms a value is an unseen table, meaning it's only called when
needed, not for every key/value pair (which add noticeable overhead
when cloning large tables).
* `getmetable` is only called once, not twice, by using a local
variable.
* `tableRefs[elt]` is only accessed once, by using a ternary operator.
* Any global accesses have been replaced by local ones.
Bug: T357199
Change-Id: I51593a971f0be728255723ca300d1ce692ead256
`mw.loadJsonData` in mw.lua throws an error if the input argument is
not a string. The error message should refer to the type of the input
argument `module`, not that of the variable `arg` (which is usually
undeclared, giving the type `nil`).
This also necessitates updating test `mw.loadJsonData, bad title (1)`
in CommonTests.lua, which checks for the error message if the input
is the number 0; the correct message should say "got number", not
"got nil".
Bug: T364987
Change-Id: I8362e91489a8ddc4c890e2799892f692c497c9b9
