Page MenuHomePhabricator

Check for possible regression in Wikidiff2 1.8.2
Closed, ResolvedPublic


AWight found cases of the new WD2 code apparently producing incorrect results. Given that this is being rolled out and a regression in compiled code can crash production servers, this was bugging me so much that I decided to look into it some more. Results so far:


This has a large number of changes, but the total count of added or deleted lines (i.e. candidates for moves) is below the threshold, so the move detection stuff is active. It may look confusing but this is intentional and OK.


This output is incorrect. The number of added+deleted lines crosses the default threshold (added+deleted lines >100). The move detection should be disabled, but it still runs.


This is also incorrect. Detection shouldn't run, same as 2).


This is the same diff as 3, but on test instead of beta, and the output is also wrong.

I imported 2) and 3) on my local installation with current Wikidiff2. Both show the correct output. Possible explanations:

  • Something related to PHP7, although I don't see how this could be an issue. My local installation runs PHP7.
  • Could it be some kind of caching issue?
  • The new Wikidiff2 version might not be properly rolled out, or
  • The server processes might not have been restarted after the rollout, so the PHP extension .so was not reloaded.

Event Timeline

WMDE-Fisch closed this task as Resolved.EditedJun 28 2019, 1:42 PM
WMDE-Fisch added a subscriber: WMDE-Fisch.

Just to update this task with the things discussed in the daily today. - The new version of Wikidiff2 will only be deployed to PHP7 servers since HVVM servers will be EOL in the upcoming weeks. The impacts of the new version could therefore not be validate when PHP7 was not explicitly requested.

Doing that in T223391#5292201 could confirm that everything works as expected. Therefore I'll close this ticket now. \o/