Re: pkg-config "Requires" fields and propagated inputs

[ludo@gnu.org (Ludovic Courtès)]

taylanbayirli@gmail.com (Taylan Ulrich "Bayırlı/Kammer") skribis:

> taylanbayirli@gmail.com (Taylan Ulrich "Bayırlı/Kammer") writes:
>
>> P.S.: I'll see if I can write a tool that compares the union of the
>> Requires[.private] fields of all .pc files in a package to the package's
>> propagated inputs, so we can detect mismatches automatically.
>
> So here's a first version.
>
> https://notabug.org/taylanub/guix/src/master/missing-propagated-inputs.scm
>
> Sample output from my system is below; evidently it's open to some
> improvement.

Nice!

> Maybe I should try to move to a strategy that calls pkg-config instead
> of parsing the files in Guile, but it might be nontrivial to get the
> PKG_CONFIG_PATH right for each call, and doing many execs might be too
> slow, so I'll first see how much further I can get with this strategy.

Yeah forking would be slow; parsing sounds good.  However, the ‘satisfy’
predicates could be based on actual package contents (when available):
looking for actual .pc files and libraries.

> $ ./pre-inst-env guile -c '((@ (guix missing-propagated-inputs) print-missing-propagated-inputs))'
> "fftw": ("-lfftw3" "-lm")
> "fftwf": ("-lfftw3f" "-lm")

This is a false positives: -lfftw3* is actually provided by the package
itself, and -lm is part of libc, so it’s always found (libc libraries
will have to be special-cased.)

> "apr": ("-ldl" "-lpthread" "-lcrypt" "-lrt" "-lapr-${APR_MAJOR_VERSION}")

Again libc libraries, and the library of APR itself.  Problem is that
this one requires almost a full reimplementation of pkg-config, to be
able to do variable substitution (probably not insurmountable though.)

> "guile": ("-lgc" "-lffi")

guile-2.0.pc has this:

  Libs: -L${libdir} -lguile-2.0 -L/gnu/store/dxfv00pgmnhxkj5jh3yxyp0q94l6i2ch-libgc-7.4.2/lib -lgc 

So here -lgc is a false positive: libgc is found by virtue of the -L
flag.

‘-lffi’ only appears in ‘Libs.private’ (static linking) so I would just
ignore it.


It looks like there are 3 classes of false positives here that would
need to be addressed to make the report more directly readable.

WDYT?

Thank you!

Ludo’.