bug#54111: guile bundles (a compiled version of) UnicodeData.txt and binaries

["Ludovic Courtès" <ludo@gnu.org>]

Hi,

Maxime Devos <maximedevos@telenet.be> skribis:

> Ludovic Courtès schreef op zo 27-02-2022 om 14:52 [+0100]:

[...]

>> We could rewrite ‘unidata_to_charset.pl’ in Scheme, but then Guile would
>> still need to provide a pre-compiled version of srfi-14.i.c for
>> bootstrapping purposes.  Or we could rewrite it in Awk, since Guile
>> already depends on Awk anyway.
>> 
>> Thoughts?
>
> The ‘blob’ seems relatively harmless to the compilation process, so
> when there are bootstrapping problems, I think we can leave it in.
>
> However, all this Unicode is important for some other things (e.g. some
> DNS and filesystem things).  So it would be nice to validate that no
> attacker with access to the Guile repo stealthily introduced some wrong
> information in during an otherwise routine update of the Unicode
> information.

The threat model is that the repository is trusted (that’s a strong
assumption, but that’s how it is).  You cannot protect against someone
with access to the repository.

We could use ‘guix git authenticate’ to improve on that.

> Hence, the following proposal:
>
>   * Make perl an optional dependency of Guile (upstream) and add an
>     '--with-unicode-data=[...]' configure flag or something like that.
>
>     If perl is detected by './configure' and '--with-unicode-data=...'
>     is set, then let one of the makefiles run 'unidata_to_charset.pl'
>     and compare the 'new' srfi-14.i.c against the old srfi-14.i.c.
>
>     In case of a mismatch, bail out.
>
>     When there's no perl or --with-unicode-data, then just use the
>     bundled srfi-14.i.c.
>
>   * Add 'perl' (or 'perl-boot0' because that perl is probably good
>     enough?) to the native-inputs of guile.
>
> Actually, the second is already done in 'guile-final'.
> Optionally, this can be combined with rewriting it in Scheme
> or some other language.

It might be easier to rewrite in Awk in build srfi-14.i.c
unconditionally no?

We can also add ‘--with-unicode-data’, though that’s orthogonal.

Thanks,
Ludo’.