If the request URL was not normalized, for example having a double slash
in it, this could cause it to fail to match in the PathRouter. But the
canonicalizing redirect was using the normalized URL, causing a redirect
loop exception.
So:
* If the PathRouter fails to match with the original URL, try matching
against the normalized URL. This allows it to still work for
normalized URLs with a double slash in the title part of the path.
* Have WebRequest::getFullRequestURL() always return the URL without
removing dot segments or interpreting double slashes. Just append
the path to the server.
* Make MediaWikiTest.php use WebRequest instead of FauxRequest, allowing
it to reproduce the exception in question. Add relevant test.
* Add tests for the new PathRouter behaviour.
Bug: T100782
Change-Id: Ic0f3a0060904abc364f75dae920480b81175d52f
Find: /isset\(\s*([^()]+?)\s*\)\s*\?\s*\1\s*:\s*/
Replace with: '\1 ?? '
(Everywhere except includes/PHPVersionCheck.php)
(Then, manually fix some line length and indentation issues)
Then manually reviewed the replacements for cases where confusing
operator precedence would result in incorrect results
(fixing those in I478db046a1cc162c6767003ce45c9b56270f3372).
Change-Id: I33b421c8cb11cdd4ce896488c9ff5313f03a38cf
The return value for ApiFormatBase::getFilename() was formerly
documented as "must be encoded for inclusion in a Content-Disposition
header's filename parameter." While this is ok for the common use case
where the module is returning a constant string or can assume whatever
it gets back from getExtensionsForType() is ok, it's not in general a
good idea to make all callers handle that.
Further, it's not possible to represent characters outside of the
ISO-8859-1 character set in a 'filename' parameter. You have to use
'filename*' to do that (see RFC 5987 and RFC 6266).
So, this patch changes the definition of getFilename() to remove the
encoding requirement, and adds code to properly convert and escape the
value for the 'filename' and (if necessary) 'filename*' parameters.
Note this may give unexpected results (double encoding) if any module
actually is returning an encoded filename. I don't see any such cases in
core or in extensions in Gerrit.
Change-Id: I0c2749a847b639f228efff4e1917a61612a1f7d1
Ensuring proper behavior of the base class lets comprehensive tests of
subclasses be simpler.
This also adjusts ApiFormatTestBase to be a bit more usable, passing an
array of options through to encodeData() instead of just a class name.
And removes the unused 'SKIP' from testGeneralEncoding, but allows
expecting an exception (for use in I63ce42dd).
Change-Id: Ib2a1fa0b04860b09105376881ff8411f9534c453
