Re: gl_MODULE_INDICATOR

[Eric Blake]

On 05/05/2011 08:28 AM, Bruno Haible wrote:
> Hi Eric,
> 
>> I have a condition (namely, test-fclose), where I only want to test
>> a particular behavior of fclose if I know that fflush was replaced at
>> the same level as fclose.  That is, if fclose is in lib, but fflush is
>> only in tests, then the test is likely to fail and so I want to compile
>> it out.
> 
> It seems an odd wish, and it seems odd to want to test for such a
> situation via
>     #if GNULIB_FFLUSH == GNULIB_FCLOSE
> 
> Rather, we should try to make things work also in situations where
> people have invoked gnulib-tool several times within the same
> configure.ac.
> 
> What is the semantic of fclose() that you want to test?
> Basically, you have two possible behaviours of fclose(), one is probably
> stricter POSIX compliant than the other.

1. fclose alone - guarantee that fdopen(sockfd) can be fclose'd
2. fclose + fflush - guarantee that fclose(stdin) properly positions the
file on seekable input

libvirt is in camp 1 - it can't import fflush (since that is GPL) but
must import fclose (since it uses socket fds on mingw).

> How can a user request the
> stricter one? By requiring 'fclose' and 'fflush' together? Why not
> introduce 'fclose-strict' (or 'fclose-posix' or whatever is appropriate)?
> This module would depend on 'fclose' and 'fflush'. Then, in the tests,
> you write
> 
>    #if GNULIB_TEST_FCLOSE_STRICT
> 
> (using the usual GNULIB_TEST_* macro).
> 
> Would that solve your use-case?

I suppose it might.  But if we just relicense fflush to be LGPLv2+, then
fclose can depend on fflush to begin with, and always solve both
problems at once, at which point I don't see the need for an fflush-strict.

-- 
Eric Blake   address@hidden    +1-801-349-2682
Libvirt virtualization library http://libvirt.org

signature.asc
Description: OpenPGP digital signature