Bug#545078: lintian: warn if symbol file contains tag but no build-dep on dpkg-dev (>= 1.15.3~)

[Jakub Wilk <ubanus@users.sf.net>]

[CC-ing people who implemented this feature, as I am not a dpkg-dev 
expert. Please correct me if I wrote something silly.]

* Russ Allbery <rra@debian.org>, 2009-09-13, 20:19:
> > dpkg-dev 1.15.3 introduced the following change:
>
> >    * Support tags before symbols in symbols file contained in source
> >      packages.  The first two tags are "optional" (use it to not fail if
> >      the symbol might disappear from the library without breaking the
> >      ABI) and "arch" to restrict the set of architectures where the
> >      symbol is supposed to exist.  Thanks to Modestas Vainius
> >      <modestas@vainius.eu> for the patch.  See dpkg-gensymbols(1) for
> >      more information. Closes: #521551
>
> > lintian should warn if symbol file contains tags but the package does
> > not build-depend on dpkg-dev (>= 1.15.3~).
>
> The tricky part of implementing this is that currently all of Lintian's
> analysis of symbol files is based on the binary package, and by that point
> dpkg-gensymbols has already analyzed and removed the tags.  Finding the
> symbol files in the source package is a bit trickier since they can be
> assembled on the fly if the package wishes to, and it would mean parsing
> them again.

I think we can accept the possibility of false negatives in case these 
files are generated on the fly.

Symbols in the source package should be checked anyway, because this is 
the only way to early catch issues in arch-specific symbol files. 
(Imagine one makes a typo in debian/symbols.s390 but don't have a S390 
handy to check if the package is actually built correctly.)

> What happens if a package with a symbol file using tags is built with an
> old version of dpkg-dev?  Does the tag end up in the installed symbol file
> as if it were a symbol? 

Yes, unless a tag contains spaces, it is considered by old dpkg-dev as
a part of symbol name. That is, in many cases, a package would simply 
FTBFS due to missing symbols...

> If so, Lintian could detect that fairly easily and warn about that.

Right, but that should be a separate check.

--
Jakub Wilk