On 09.02.2014 22:26, Ben Reser wrote: > On 2/9/14, 8:06 AM, Branko Čibej wrote: >> I just tested the 1.8.5 release tarball in exactly the same environment. All >> tests pass (except for wc-queries-test #3, which we know did not work with >> SQLite 3.8.x in 1.8.5). >> >> As far as I'm concerned, this proves conclusively that 1.8.6 has a serious >> regression somewhere, and my -1 stands. Judging by the test failures, it >> appears to be related to the client-side move handling, which I know nothing >> about. >> >> It's a shame because there are a few nice improvements in this version; but >> I'm >> not comfortable with releasing until and unless we figure out what's wrong. > Found how to reproduce it without the homebrew setup. Homebrew is passing > -DSQLITE_ENABLE_STAT3 in CPPFLAGS. You can reproduce this with a shared > version of SQLite or the amalgamation by passing > CPPFLAGS="-DSQLITE_ENABLE_STAT3" to the appropriate configure command > (SQLite's > if building shared or ours if using the amalgamation). > > The setting is documented here: > https://sqlite.org/compile.html#enable_stat3 > > Homebrew does not provide an option to disable this so all their builds have > this. > > Believe that we have a regression in 1.8.6. Branko is right, we should fix > this before moving forward. > > 1.8.6 will not be released.
:( The only good thing about this is that the regression exists on trunk, too, so it's not a side effect of a backport. It must be fairly recent, because I (used to) run trunk tests fairly regularly with homebrewed SQLite. I just wonder when we started to rely on a particular query plan. -- Brane -- Branko Čibej | Director of Subversion WANdisco // Non-Stop Data e. br...@wandisco.com
