Hi Norbert, On Fri, Sep 18, 2015 at 12:02:58AM +0900, Norbert Preining wrote: > I think the prime problem is with tex-common, although you didn't > spot it. tex-common declares an interest on some files called > format files. TeX (and friends like metafont) create dump files, > memory images, which are generated via the trigger mechanism of > dpkg by tex-common's postinstall, see do_triggers which calls > basically fmtutil-sys.
I certainly lack intimate knowledge of TeX packaging. This is the primary reason why I didn't submit a bug with a patch. I do see the triggers into tex-common while installing/upgrading TeX packages, but I have no clue what they do. > So besides the format dump files, I don't see anything where the > arch might possibly leak in the way that it disturbs M-A. And > then again, the real problem is not mixing i386 and am64, but > big-endian and little-endian systems. Being endianess-aware can be a good reason for not marking something M-A:foreign. > > texlive-latex-base M-A:foreign has a very immediate benefit on cross > > building and being able to mix architectures in a multiarch enabled > > system. > > I am not sure here - the format dumps are generated by tex-common > via trigger, but the files that are used to generated the formats > are shipped by different packages. I mentioned this from a user's (actually cross builder's) perspective, i.e. what would be desirable. This was not meant to imply that it is actually correct. Sorry for the confusion. >From a multiarch pov, triggers count equally as maintainer scripts. It is very good that you mention them. Your knowledge of TeX is very helpful here. > In the case that the dump files are *not* compatbile, would it > still make sense to decalre all the non-tex-common packages > M-A: foreign, and only the tex-common package something else? > I guess that does not make sense, right? Let's distinguish two aspects here: Desirability and correctness. Having almost everything except tex-common is probably still desirable, because only few things depend or build-depend on it directly. Most of the relevant dependencies are on texlive-something. As for correctness, I am not so sure. Even though it is the tex-common postinst that runs as a trigger, the architecture awareness needs to be accounted to the package that invokes the trigger (like the locales example runs code from libc-bin). So my question here would be: How are these (potentially arch-dependent) dumps supposed to be used? > Thanks a lot for your continous support and patience with my > ignorant questions Please keep this going. It makes cross builders happy. Helmut
