[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnulib] Re: fnmatch.c includes alloca.h conditionally?

From: Paul Eggert
Subject: Re: [Bug-gnulib] Re: fnmatch.c includes alloca.h conditionally?
Date: Tue, 20 Apr 2004 11:47:27 -0700
User-agent: Gnus/5.1006 (Gnus v5.10.6) Emacs/21.3 (gnu/linux)

Derek Robert Price <address@hidden> writes:

> Is there a reason the Autoconf manual doesn't include the _MSC_VER switch?

Most likely it's because we didn't know about the problem when that
text was written.  Also, there's understandable reluctance to
complicate a common idiom for every little compiler idiosyncrasy that
we run into.  Bruno's (1b) suggestion (of arranging things so that
"#include <alloca.h>" always works) is the best way I know to get out
of this mess in the long run.

> Does this mean that you think it would be hard to persuade the
> libiberty folks to use approach 1b?

It might take some time, but I don't think we need to wait for that.
Please see the last two lines of this message for a suggestion that
should work regardless of whether the libiberty folks adopt 1b.

> 1.  Ok, vasnprintf.c includes alloca.h almost unconditionally, so it
> almost makes sense to include the alloca.a module, but on closer
> inspection, it appears broken.  Assuming HAVE_ALLOCA_H is not defined
> so that the GNULIB alloca.h header gets built and included, I would
> guess that HAVE_ALLOCA will also usually be undefined, as is the case
> on Windows.  Including alloca.h can then sometimes create a working
> alloca substitute, for instance on Windows where alloca is #defined to
> _alloca, and I presume something similar is happening in the AIX & HP
> ifdefs.  It then goes unused since HAVE_ALLOCA was never defined,
> rendering vasnprintf's dependency on alloca pointless since it is
> never used despite the header file's inclusion,

It's not pointless, since the maintainer of vasprintf can include
<alloca.h> unconditionally without worrying about whether there's a
system alloca.h; this simplifies the code a bit.  The maintainer can
also use HAVE_ALLOCA to determine whether to use an efficient alloca.
However, I admit that it's a bit confusing for those who don't know
all the twists and turns here.  Perhaps it would be slightly clearer
to replace this:

   #ifndef IN_LIBINTL
   # include <alloca.h>

by this:

   #ifdef HAVE_ALLOCA
   # include <alloca.h>

in vasnprintf.  That IN_LIBINTL is a bit of a mystery anyway -- at
least to me -- as the code later uses alloca if HAVE_ALLOCA, not if

While we're on the subject, the following code in lib/alloca_.h
is a bit weird:

#    include <alloca.h>

This "#include" cannot possibly be reached (which is a good thing as
the file would be including itself if it did).  Wouldn't it clarify
things a bit to remove that "#if" and "#include"?

> 2.  Why should regex depend on the alloca module then?

I don't know.  It doesn't seem to be necessary as things stand now.
argp and fnmatch are similar: they don't seem to need the alloca
module now.

> The fnmatch module doesn't currently work as specified on Windows.  It
> skips the _MSC_VER stuff that the alloc.h header does.

I assume you mean gnulib's "alloca_.h" here, not "alloc.h".

> Ok, but how about adding the full switch from alloca.h, including the
> _MSC_VER portion, to fnmatch.c?

Wouldn't it be simpler to change gl_FUNC_ALLOCA to always define HAVE_ALLOCA_H?

reply via email to

[Prev in Thread] Current Thread [Next in Thread]