This changes the query planner, so it can change perf. We should get this early in this cycle to evaluate it.

uploading test patch to be pushed to the try server to test performance.

I've pushed the upgrade to the Places branch to gather Talos numbers, will report my findings. Unfortunately I'll have to backout and reland even if positive, since I was so cool to insert the wrong bug # in the commit message :\ But well, that's the easy part. Then I'll be also able to include approval for the update. Notice the patch in comment 2 is missing test_quota.c changes, and that those changes break compilation (casts have been wrongly removed, waiting an answer from SQLite team).

(In reply to comment #3) > Notice the patch in comment 2 is missing test_quota.c changes, and that those > changes break compilation (casts have been wrongly removed, waiting an answer > from SQLite team). I told her to ignore the test for the time being, FWIW.

a first result, is that the update makes random failure bug 610789 to happen much more often, it's almost permaorange. The only relevant result from talos seems a ts_shutdown change, but I'm respinning a bunch of tests to check. Notice 3.7.7 will be release at the end of June, and it includes a really interesting fix for us: [db7e500f69] Enhance WAL mode so that the WAL file honors the journal size limit set by PRAGMA journal_size_limit. If we end up near that date we could evaluate skipping 3.7.6 (it's probably matter of checking if we can take 3.7.6 in FF6 and 3.7.7 in FF7)

I'd say that most of the variations in talos are inside noise, the only blocking issue may be the almost permaorange. http://tinyurl.com/6985hpx

I think there is poor value in taking 3.7.6 as of now, 3.7.7 has more useful changes, especially for Mobile, and it's not far. Unless anyone disagree I'd suggest to wontfix this update.

Sounds like a good plan to me.