Fallback for user language and content language is easiest to
resolve server-side. Also saves sending a lot of data to the
client that it doesn't need. Similar to how ResourceLoader only
sends 1 set of message values.
Bug: 50431
Bug: 52922
Change-Id: If8317ed6522a05d5a48a210ff43c97277b950a97
gzdecode() has been introduced in PHP 5.4 although gzencode() has been
there for a while. Wikimedia is still using PHP 5.3, so the
TemplateDataBlob::newFromDatabase() would trigger a fatal error whenever
it is passed compressed JSON :-(
The test does not fix the root cause (that needs a fallback function in
the TemplateData extension), but it does highlight the issue in PHP 5.3.x.
Updated a @return comment, correcting the class being returned.
bug: 54058
Change-Id: I8357b474165f4317982b9c924cf0afe68650d677
We only have 65535 bytes, let's use them wisely.
Prior to Ib2db241a in core (1.22wmf12) even binary page properties
were sometimes shown as plaintext (e.g. on Special:PagesWithProp).
Thus this introduces a "soft dependency" on that change.
No backwards-incompatible changes were made in public methods.
Added two new ones:
* Add TemplateDataBlob#getJSONForDatabase which returns compressed
JSON, use it where applicable.
* Add TemplateDataBlob::newFromDatabase which accept compressed and
uncompressed JSON, use it where applicable.
Use a long pseudorandom string in the test.
Bug: 51740
Change-Id: Ie66b0dd6b6dab6f8648e78595c41e52d9c704d57
This might not be necessary on non-MySQL databases, but I'm sure that
at least their limits aren't lower than MySQL's, so let's stay on the
safe side.
Bug: 51740
Change-Id: I9c7115a951f126844bc70755bff12e8cd5fc2be1
Previously we only compared output, we now evaluate
status as well.
For example test {} => {} was misleading since the reason
it roundtripped was because it is *invalid* and we replace
invalid blobs with an empty object in blob storage.
But in the API this'll never get used because getStatus
will have a fatal error in it. Which overrides the data.
The tests were previously looking in the data whilst ignoring
the return status.
Change-Id: I9809f126c1615c9bc9c28ce2d8bb7953058cbf6e
InterfaceText now defaults to null instead of {en:""} which
was awkward to deal with in the frontend.
As specified:
- label is InterfaceText
- type is a string and must be a one of the recognized types
Updated example for the hypothetical variant of Template:Unsigned
and removed other no longer needed example.
HTML output has been revised per conversation with Trevor,
James and Timo:
- Not sortable.
- Add label to html output.
- Aliases in the main parameter column (one per line),
but muted in styling.
- Add type to html output.
The css module styles content from the server, not content
generated by javascript. Moved module to position => top to
fix flash of unstyled content.
Change-Id: I16d3f9e460c5513935b9b55fe4cec0092b38e6c2
I ran into countless random exceptions and errors locally due to
cache I built up over time that was no longer valid. Though in
approved commits in the repository there is only 1 version as of
now, I'm sure we will make changes in the future that require a
cache invalidation since getHtml needs to be able to make
assumptions about what properties exist.
Also fixed a bug in the parse() method where the 'deprecated'
property is asserted the wrong way around. Added a test for it.
Change-Id: I5f61d6030578a711909435c8b996373e9aaa5178
* Add unit tests for all types of invalid input we check for.
* Add unit tests for all types of input we expand or otherwise
normalise.
* Implement InterfaceText expansion/normalisation.
* Fix bug that caused a string value in the root description property
to be considered invalid (it only accepted an object, it should
accept both).
Change-Id: I5a15080f1f924451a9dde8af96ea2922011981ec
