Bug comment: Set PCRE_MULTILINE on spamblacklist regexes. $ on spam blacklist regex should match the end of the url (not of the text) so it can be used to match only the mainpage. Since the candidate urls are already joined with a new-line separator, it's just setting PCRE_MULTILINE on the regex.
An API edit attempt with Spam Blacklist firing will now output something instead of crashing:
<?xml version="1.0"?><api><edit spamblacklist="http://blacklistme.example.com"
result="Failure" /></api>
Lines with "\" at the end would silently break both that and the following line in the batch, without triggering the overall parse errors.
Added a specific check for this case to skip the bad lines when building, and to check for them and report a warning during editing.
function_exists( 'Http::get' ) always returns false, since that's not a function name.
If the current state of the extension only works on modern versions of MW, then just call Http::get() without checking. If it still works on versions that predate it, it's best to keep the existing compat code probably.
* Use the new EditFilterMerged hook if available, for faster link finding
* Random bits of code were leaking out of the body file into the loader, poked them back in.
* Handle bad regexes more gracefully:
- The batched regexes are tested for validity, and if one is bad, the lines from that source are broken out line-by-line. This is slower, but the other lines in that source will still be applied correctly.
- Suppress warnings and be more verbose in the debug log.
- Check for bad regexes when a local blacklist page is edited, and prompt the user to fix the bad lines.
* Caching issues:
- Cache the full regexes per-DB instead of per-site; this should be friendlier to shared environments where not every wiki has the same configuration.
- Hopefully improve the recaching of local pages, which looked like it would preemptively apply the being-edited text to the cache during the filter callback, even though something else might stop the page from being saved. Now just clearing the cache after save is complete, letting it re-load later.
* Split out some of the regex batch functions for clarity.
There are probably still issues with caching of HTTP bits, and in general the local DB loading looks verrrry fragile.
Test this a bit more before syncing. :)
Should save some trouble for annoyed people. :)
The regular message cache behavior is used for this message, so it'll also update immediately, without waiting for the shared caches to time out.
Additionally, added a fix for configurations which don't hardcode the PHP include_path by using $IP in an include for HttpFunctions.php.
