[gnutls-dev] Guile needs 1.8?

Simon Josefsson simon at josefsson.org
Fri Jun 29 16:32:28 CEST 2007


ludo at gnu.org (Ludovic Courtès) writes:

> Hi,
>
> Simon Josefsson <simon at josefsson.org> writes:
>
>> Hi!  I think it is common in Debian to allow multiple versions of
>> libraries to be installed, and that they don't conflict with *-dev
>> packages unless the ABI isn't backwards incompatible.  In this case,
>> doesn't the guile 1.8 libraries support the 1.6 API/ABI?
>
> Oh, you're right: 1.8 does support (in deprecated form) the 1.6 API (and
> ABI too, I think).

Ok.

>> My view is that the Guile M4 macros are broken here, the autoconf
>> approach is to test for the features you need instead of relying on
>> version information.
>
> The Guile M4 macros don't check for versions.  I added such a check in
> GnuTLS to address your (valid) concerns.

Thanks.

> Guile has a versioning policy similar to that of GnuTLS: odd-numbered
> branches are unstable and even-numbered branches are stable; there *are*
> significant API changes between two subsequent stable branches, although
> the previous stable API is still supported (and deprecated).
>
> The switch from 1.6 to 1.8 was such a major change.  New features were
> added and the API was overhauled.  The GnuTLS bindings are written
> against the 1.8 API.  They also use features that were not available in
> 1.6 and never will, such as SRFI-4 vectors.  There's no way they could
> someday "work" with 1.6.
>
> Therefore, checking for *specific* features (such or such macro or
> function) is irrelevant: we really want the whole shebang, not just one
> macro.  Checking for `scm_from_locale_string ()' allows us to know that
> we really get 1.8 (or some later backward-compatible version): this is
> one of the functions that was introduced in 1.8 and that will definitely
> never be available in 1.6.
>
>> I'm not sure, do the GnuTLS guile bindings uses and require the
>> scm_from_locale_string function?
>
> Yes, they do require this specific function.  But even if they didn't,
> that would still make sense to check for it, for the reasons outlined
> above.
>
> I hope this clarifies the rationale for the autoconf tests.

Yes, thanks for explaining.  I installed the patch.

/Simon



More information about the Gnutls-dev mailing list