bug#62051: Early detection of derivations with unreadable builder scripts

[Josselin Poiret via Bug reports for GNU Guix <bug-guix@gnu.org>]

[-- Attachment #1: Type: text/plain, Size: 2080 bytes --]

Hi Chris,

Christopher Baines <mail@cbaines.net> writes:

> This is part of the following builder script:
>
>   (cons "--enable-mpi-java" #<gexp  gnu/packages/mpi.scm:233:24 7f366e0cd930>)
>
> from: /gnu/store/yngxnpcs4s6y8acxf4nwx5pcpj0j6q6i-java-openmpi-4.1.4-builder
>
> And when attempting to build that derivation, you get the following
> error.
>
>   ice-9/read.scm:126:4: In procedure read-expr*:
>   /gnu/store/yngxnpcs4s6y8acxf4nwx5pcpj0j6q6i-java-openmpi-4.1.4-builder:1:3820: Unknown # object: "#<"
>
>
> It would be nice if Guix could detect this category of problems and
> raise an error at the time the derivation is created, rather than the
> error occuring only when you build the derivation.
>
> This would be helpful particularly for the Guix Data Service since
> currently it ends up storing these useless derivations, often many times
> since the builder includes some often changing string (7f366e0cd930 in
> the example above), so this is a common cause of spurious changes
> between revisions (as often noted on qa.guix.gnu.org).

We could probably modify sexp->string, or the builder bind in
gexp->derivation so that the sexp is sanity-checked for non-printable
things (we could even work on a whitelist basis).  However, the
docstring of sexp->string talks about performance, and indeed "write" is
pure C code and very fast.  I'd be reluctant to introduce a performance
hit that would be too heavy here.

This particular example though was caused by non-gexp #:phase arguments,
so another option could be to sanity check sexps given to sexp->gexp,
but again, the docstring talks about performance, so I'm not sure what
we should do here.  In general, things written only with G-Exps should
work well, because you can't insert random stuff into them, but S-Exps
are more dangerous, hence why I think this option would be a better
middle ground.

Paging Ludo wrt. the performance cost of this (I can write a patch for
it adding a whitelist of what is allowed in a sexp->gexp sexp).

Best,
-- 
Josselin Poiret

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 682 bytes --]