Hi Stéphane, Quoting Stéphane Glondu (2018-12-08 09:41:48) > Le 26/11/2018 à 09:12, Johannes Schauer a écrit : > > what do we want to do about this bug? Even the new upstream release doesn't > > yet > > get rid of the Obj.magic. > One easy thing to do is to disable the failing test. But it means some > programs buggy on i386 (probably using atd-stuff with floats) may slip in > production environments. What do you think?
I think the decision is between either shipping ocaml-atd with buggy float behaviour on i386 or not shipping ocaml-atd at all on i386. Since there are still perfectly valid uses for ocaml-atd on i386 (if one doesn't use floats) and because ocaml-atd has reverse dependencies that do not make use of floats, I think it would not be a good idea to remove it on i386. So I think the right thing to do here, would be to make sure that all reverse dependencies still work on i386 and that the users on i386 are made aware of the limitations of the package. As for the latter, one could install an architecture-specific NEWS.Debian file either by using dh-exec in the *.install files: #!/usr/bin/dh-exec [i386] debian/NEWS.Debian.i386 /usr/share/doc/ocaml-atd/NEWS.Debian Or by putting NEWS.Debian into an architecture-specific subdirectory and then adding architecture specific *.install files like a "libatdgen-ocaml-dev.install.i386" with content like: debian/NEWS.Debian.i386/NEWS.Debian /usr/share/doc/libatdgen-ocaml-dev > > Currently, botch cannot migrate to testing because of atd and the freeze is > > looming closer... > As reverse-dependencies of ocaml-atd, I see botch, nss-passwords and > belenios. The last two don't use atd stuff with floats so we can hope they > are not affected by this bug. What about botch? It seems that botch has a float datatype: https://sources.debian.org/src/botch/0.21-8/datatypes.atd/ But that type seems to be unused. I just patched the "fstat" line away and rebuilt botch and everything still works fine. The rest is integers. Thanks! cheers, josch
signature.asc
Description: signature