On 12/18/11 12:14, Amit Kulkarni wrote: > On Sun, Dec 18, 2011 at 10:57 AM, Nigel Taylor > <[email protected]> wrote: >> On 12/18/11 10:28, Stuart Henderson wrote: >>> On 2011/12/17 23:51, Nigel Taylor wrote: >>>> This update is a requirement for the next versions of Firefox, >>>> Thunderbird, etc, they require 3.7.7.1 or later, 3.7.9 is recommended >>>> for KDE 4.8beta1. Various updates to sqlite3 3.7.5 but have been sent to >>>> @ports but never committed. Chromium - chrome no longer uses the system >>>> sqlite3 as too far behind. >>> >>> We need to keep the regression tests (which the openbsd-wip version >>> loses) and the weak-aliased pthread code (which other diffs lost), >>> that is why they haven't been committed. >>> >>> In some ways it would make more sense to use an embedded copy of >>> sqlite for applications which often need a specific version, >>> but then we typically have symbols conflicting with a copy of >>> sqlite pulled in via a shared library dependency (and locally >>> renaming functions to avoid this in something like firefox is >>> not sustainable). >>> >>> >> Hi, >> >> I junked the openbsd-wip version and have gone back to 3.7.5, updated that >> to 3.7.9 downloading the sqlite-src-3070900.zip which includes the regression >> tests, rather than using sqlite-autoconf-3070900.tar.gz. >> >> Attached is a new diff, >> >> regression tests on amd64 >> 3.7.5 81 tests failed out of 132893 >> 3.7.9 1 test failed out of 129087, failed test is backup2-10, >> same test fails in 3.7.5 also. >> Both error on tcl tests as unsafe. >> >> regression test on i386 >> 3.7.9 1 test failed out of 128942, failed test is backup2-10. >> error on tcl tests as unsafe. >> >> Tests, Comments, Ok. > > this is unsupported and upstream recommends to use the autoconf tar.gz > and is why sthen & pea did the new port in the first place. > > i downloaded both sets and saw that there is no folder called tests in > the autoconf tar.gz while there is in the src .zip. upstream has > dropped the tests, we need to let them know to put them back. > > thanks
Last I talked with DRH about this, he was fairly adamant that since the sqlite people do plenty of testing, we should just be confident with the amalgamation. Not that I'm suggesting not telling them, just relating my past experience. I'll bring it up with him again. How hard/wacky would it be to build/use the amalgamation but run the tests from the other source tarball? Stu
