Bug#1016418: transition: fmtlib
Control: tags -1 confirmed Control: tags 1017412 confirmed On 2022-08-17 03:42:18 +0800, Shengjing Zhu wrote: > On Wed, Aug 17, 2022 at 3:07 AM Sebastian Ramacher > wrote: > > > > On 2022-08-15 20:35:49 +0800, Shengjing Zhu wrote: > > > CC spdlog maintainer as well. > > > > > > On Mon, Aug 15, 2022 at 01:45:32PM +0200, Sebastian Ramacher wrote: > > > > Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib > > > > transition. Please fix this issue first and remove the moreinfo tag once > > > > that's done. > > > > > > What's your suggestions to fix the spdlog ABI breakage? > > > > The ABI breakage was caused by adding a new argument to some functions > > with a default argument. This could be fixed by keeping the old > > functions around. > > > > TBH, given libspdlog1.10 is in experimental, I feel it's more correct > to just do a spdlog transition. Please go ahead with both uploads. Cheers -- Sebastian Ramacher
Bug#1016418: transition: fmtlib
On Wed, Aug 17, 2022 at 3:07 AM Sebastian Ramacher wrote: > > On 2022-08-15 20:35:49 +0800, Shengjing Zhu wrote: > > CC spdlog maintainer as well. > > > > On Mon, Aug 15, 2022 at 01:45:32PM +0200, Sebastian Ramacher wrote: > > > Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib > > > transition. Please fix this issue first and remove the moreinfo tag once > > > that's done. > > > > What's your suggestions to fix the spdlog ABI breakage? > > The ABI breakage was caused by adding a new argument to some functions > with a default argument. This could be fixed by keeping the old > functions around. > TBH, given libspdlog1.10 is in experimental, I feel it's more correct to just do a spdlog transition. -- Shengjing Zhu
Bug#1016418: transition: fmtlib
On 2022-08-15 20:35:49 +0800, Shengjing Zhu wrote: > CC spdlog maintainer as well. > > On Mon, Aug 15, 2022 at 01:45:32PM +0200, Sebastian Ramacher wrote: > > Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib > > transition. Please fix this issue first and remove the moreinfo tag once > > that's done. > > What's your suggestions to fix the spdlog ABI breakage? The ABI breakage was caused by adding a new argument to some functions with a default argument. This could be fixed by keeping the old functions around. Cheers > > Currently, > > + spdlog/1.9 ftbfs with fmtlib/9 > + Upstream doesn't want to fix the ABI breakage in spdlog/1.10, the opening > issue is closed without fix. > > Solutions? > > + revert spdlog to 1.9 and let it use a embedded copy of fmtlib, so it won't > block fmtlib/9 transition. > > But I think it needs to use soversion libspdlog1, not libspdlog1-fmt8 as > well? > > + bump spdlog so version in Debian only without coordination from upstream. > > For example, -> libspdlog1.10-fmt8 > > + kick spdlog out of testing. -- Sebastian Ramacher
Bug#1015742: Bug#1016418: transition: fmtlib
On Mon, Aug 15, 2022 at 8:36 PM Shengjing Zhu wrote: > > CC spdlog maintainer as well. > > On Mon, Aug 15, 2022 at 01:45:32PM +0200, Sebastian Ramacher wrote: > > Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib > > transition. Please fix this issue first and remove the moreinfo tag once > > that's done. > > What's your suggestions to fix the spdlog ABI breakage? > > Currently, > > + spdlog/1.9 ftbfs with fmtlib/9 > + Upstream doesn't want to fix the ABI breakage in spdlog/1.10, the opening > issue is closed without fix. > > Solutions? > > + revert spdlog to 1.9 and let it use a embedded copy of fmtlib, so it won't > block fmtlib/9 transition. This seems much more complicated since packages may use spdlog and fmt together. > > But I think it needs to use soversion libspdlog1, not libspdlog1-fmt8 as > well? > > + bump spdlog so version in Debian only without coordination from upstream. > > For example, -> libspdlog1.10-fmt8 > So I will upload NMU to bump spdlog soversion to 1.10. It's unfortunate that we diverted from upstream with the soversion. But we'll align with upstream once spdlog 1.11 is released. @spdlog maintainer, if you disagree with this, there is still enough time to retract my NMU since it hits the NEW queue. > + kick spdlog out of testing. This is impossible as spdlog is a key package. -- Shengjing Zhu
Bug#1016418: transition: fmtlib
CC spdlog maintainer as well. On Mon, Aug 15, 2022 at 01:45:32PM +0200, Sebastian Ramacher wrote: > Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib > transition. Please fix this issue first and remove the moreinfo tag once > that's done. What's your suggestions to fix the spdlog ABI breakage? Currently, + spdlog/1.9 ftbfs with fmtlib/9 + Upstream doesn't want to fix the ABI breakage in spdlog/1.10, the opening issue is closed without fix. Solutions? + revert spdlog to 1.9 and let it use a embedded copy of fmtlib, so it won't block fmtlib/9 transition. But I think it needs to use soversion libspdlog1, not libspdlog1-fmt8 as well? + bump spdlog so version in Debian only without coordination from upstream. For example, -> libspdlog1.10-fmt8 + kick spdlog out of testing.
Bug#1016418: transition: fmtlib
Control: tags -1 moreinfo On 2022-07-31 20:18:31, Shengjing Zhu wrote: > Package: release.debian.org > Severity: normal > User: release.debian@packages.debian.org > Usertags: transition > X-Debbugs-Cc: z...@debian.org > > Ben file: > > title = "fmtlib"; > is_affected = .depends ~ "libfmt8" | .depends ~ "libfmt9"; > is_good = .depends ~ "libfmt9"; > is_bad = .depends ~ "libfmt8"; > > fmtlib 9 has been release for a while, I'd like to start the transition. > > I have built the rerverse-depends, and all ftbfs bugs have been filed. > See the affected packages in > https://bugs.debian.org/cgi-bin/pkgreport.cgi?repeatmerged=no=fmtlib > > Most packages have been fixed, or fixed-upstream, or have patches. > > The left packages are > + #1014677 src:opendht, I think the root cause is restinio, so I will re-test > after the new restinio is uploaded. > + #1014678 src:openimageio, it's not in testing. > > One difficult situation is spdlog. > > I have uploaded spdlog/1.10 last few days, but it causes a regression, since > upstream > breaks the ABI without bumping SOVERSION (#1016408). > > After filling bugs on upstream (https://github.com/gabime/spdlog/issues/2454), > the response seems depressed. They will start to use the package minior > version in > soversion, like 1.11, instead of maintaining a stable soversion. > > OTOH, spdlog in Debian already doesn't use soversion to track transition, > since it embeds > fmtlib version in it's shared library. Currently it's libspdlog1-fmt8. > > With the fmt transition, spdlog will also transit to libspdlog1-fmt9. So it > can solve the ABI > regression by spdlog/1.10. Let's note hide spdlog's ABI breakage that is unreleated to the fmtlib transition. Please fix this issue first and remove the moreinfo tag once that's done. Cheers -- Sebastian Ramacher
Bug#1016418: transition: fmtlib
Package: release.debian.org Severity: normal User: release.debian@packages.debian.org Usertags: transition X-Debbugs-Cc: z...@debian.org Ben file: title = "fmtlib"; is_affected = .depends ~ "libfmt8" | .depends ~ "libfmt9"; is_good = .depends ~ "libfmt9"; is_bad = .depends ~ "libfmt8"; fmtlib 9 has been release for a while, I'd like to start the transition. I have built the rerverse-depends, and all ftbfs bugs have been filed. See the affected packages in https://bugs.debian.org/cgi-bin/pkgreport.cgi?repeatmerged=no=fmtlib Most packages have been fixed, or fixed-upstream, or have patches. The left packages are + #1014677 src:opendht, I think the root cause is restinio, so I will re-test after the new restinio is uploaded. + #1014678 src:openimageio, it's not in testing. One difficult situation is spdlog. I have uploaded spdlog/1.10 last few days, but it causes a regression, since upstream breaks the ABI without bumping SOVERSION (#1016408). After filling bugs on upstream (https://github.com/gabime/spdlog/issues/2454), the response seems depressed. They will start to use the package minior version in soversion, like 1.11, instead of maintaining a stable soversion. OTOH, spdlog in Debian already doesn't use soversion to track transition, since it embeds fmtlib version in it's shared library. Currently it's libspdlog1-fmt8. With the fmt transition, spdlog will also transit to libspdlog1-fmt9. So it can solve the ABI regression by spdlog/1.10.
