Bug#789927: Anthy library breakage
Hi, On Tue, Sep 05, 2017 at 09:00:52AM +0900, NIIBE Yutaka wrote: > Osamu Aoki wrote: > > Thanks. Now, I understand (I didn't know that fully). I should have > learned before uploading to Sid. After I failed to make ibus transition like your case, I made a memo as: https://www.debian.org/doc/manuals/debmake-doc/ch05.en.html#libnames This sites are useful: https://debconf4.debconf.org/talks/dependency-hell/img0.html https://wiki.debian.org/Teams/ReleaseTeam/Transitions I will check the whole situation a bit carefully to make recommendation later, if I can. If upstream didn't bump so name for this UTF-8 change, that complicates things. Did he? > I have been using Anthy since its birth, but I only use through Emacs > with egg. Even if I don't use, I should be careful when I touch Debian > package. > > Today, the Emacs module egg requires total rewrite for Emacs 25, due to > the change of Emacs (display property thing). I was considering making > a loadable module for Emacs for Anthy. That was another pressure. I see. I may need to update im-config once you do this. > Any suggestions are welcome. It seems for me that it would be better > to change the -dev package name. I think "Use the versioned -dev package names: libfoo7-dev and libfoo8-dev" is the way to go for this case. But I need to check more. Use the versioned -dev package names: libfoo7-dev and libfoo8-dev This is the special case. Two versions of the library source packages are allowed simultaneously in the archive. * Make all dependent packages depend on libfoo-dev. * Make both libfoo7-dev and libfoo8-dev provide libfoo-dev. * The source package needs to be renamed as libfoo7-7.0.tar.gz and libfoo8-8.0.tar.gz respectively from libfoo-?.0.tar.gz. * The package specific install file path including libfoo7 and libfoo8 respectively for header files etc. needs to be chosen to make them co-installable. Regards, Osamu
Bug#789927: Anthy library breakage
Osamu Aoki wrote: > If what NOKUBI Takatsugu's comment is correct, I wonder why anthy is > upgraded without coordinating with key users: > ibus > uim > fcitx > > (Please note these have many dependence packages so the testing > migration is slow if package version dependency is correctly recorded.) I'm sorry. It was me who upload the anthy package for Sid. Apparently, I was conscious of the impact when I was uploading it for experimental in 2015. In August 2017, I forgot this issue when I was asked uploading new Anthy for Debian. I had thought as if it's only anthy's own problem. The original plan in 2010 was doing the migration in Debian around 2010. It didn't occur because of our human resources (and troubles). What we did in 2010 was: We made a team as pkg-anthy on Alioth and revived upstream development to collect patches floating around and to merge. Then, in 2015, I uploaded it in experimental. > I agree moving to utf8 is right thing to do but this breakage is > something we should avoid. Yes. > If -dev package version is bumped, we can have slow migration without > breaking packages depending on old anthy. But if we share the same -dev > package name, all related package needs to be uploaded together > (otherwise we suffer long broken sid system). Thanks. Now, I understand (I didn't know that fully). I should have learned before uploading to Sid. > But if library change its API spec from non-utf8 to utf8, this needs to > be coordinated carefully. I learned. > It's not simple. What does people think is the right way to fix. I have been using Anthy since its birth, but I only use through Emacs with egg. Even if I don't use, I should be careful when I touch Debian package. Today, the Emacs module egg requires total rewrite for Emacs 25, due to the change of Emacs (display property thing). I was considering making a loadable module for Emacs for Anthy. That was another pressure. Any suggestions are welcome. It seems for me that it would be better to change the -dev package name. --
Bug#789927: Anthy library breakage
Hi, https://bugs.debian.org/873860#10 If what NOKUBI Takatsugu's comment is correct, I wonder why anthy is upgraded without coordinating with key users: ibus uim fcitx (Please note these have many dependence packages so the testing migration is slow if package version dependency is correctly recorded.) I agree moving to utf8 is right thing to do but this breakage is something we should avoid. If -dev package version is bumped, we can have slow migration without breaking packages depending on old anthy. But if we share the same -dev package name, all related package needs to be uploaded together (otherwise we suffer long broken sid system). I know we tend to forget about it. (I did this for ibus killing ibus-mozc for few months alienating many uses). But if library change its API spec from non-utf8 to utf8, this needs to be coordinated carefully. I haven't checked all the facts but it seems we should have bumped package version and set conflict with -dev etc. Just uploading new package may fix it in sid but this kind of api change can't be detected by symbol file and probable break testing. It's not simple. What does people think is the right way to fix. Osamu
