Re: ["Eli Zaretskii" <address@hidden>] Re: configure: a.out.h vs nlist.h

[Andreas Schwab]

Akim Demaille <address@hidden> writes:

|> | It seems like AC_FUNC_GETLOADAVG does not define NLIST_STRUCT any more,
|> | even though getloadavg.c needs it and it is documentated.
|> | 
|> | Andreas.
|> | 
|> | From: "Eli Zaretskii" <address@hidden>
|> | Subject: Re: configure: a.out.h vs nlist.h
|> | To: Andreas Schwab <address@hidden>
|> | Cc: address@hidden, address@hidden
|> | Reply-To: Eli Zaretskii <address@hidden>
|> | 
|> | > From: Sam Steingold <address@hidden>
|> | > Date: 24 Oct 2001 16:24:28 -0400
|> | > 
|> | > > > From: Sam Steingold <address@hidden>
|> | > > > Date: 24 Oct 2001 12:55:05 -0400
|> | > > > 
|> | > > > On the current CVS emacs, I get this:
|> | > > > getloadavg.c:356:24: a.out.h: No such file or directory
|> | > > > after 'make distclean; ./configure; make bootstrap'
|> | > > > on Solaris 2.5.1, gcc 3.0, autoconf 2.52.
|> | > > 
|> | > > That's weird: according to "cvs log", getloadavg.c didn't change since
|> | > > June 2000.  Could you please look closer at this and see what changed
|> | > > since the last pretest you built (did you build any pretests on that
|> | > > machine?).
|> | > > 
|> | > > Is it possible that this is the first time some of the tools are used,
|> | > > or a new version of some tool?
|> | > 
|> | > I think I was not clear enough.
|> | > the problem is obviously with the recent changes in the configuration
|> | > process, i.e., the check for a.out.h vs nlist.h (see the subject line)
|> | > in configure.  I suspect that when aclocal.m4 was dropped, the check
|> | > went away too, so the default include (a.out.h) is used now always.
|> | 
|> | The problem is not with aclocal.m4, it's with the new Autoconf: its
|> | AC_FUNC_GETLOADAVG macro doesn't seem to define NLIST_STRUCT, and
|> | getloadavg.c needs that, at least on your system.
|> | 
|> | Andreas, could you please look into this?
|> | 
|> | ----------
|> 
|> I don't think this was meant :(
|> 
|> But looking at the code from 2.13:
|> 
|>   AC_CHECK_HEADER(nlist.h,
|>   [AC_DEFINE(NLIST_STRUCT)
|>   AC_CACHE_CHECK([for n_un in struct nlist], ac_cv_struct_nlist_n_un,
|>   [AC_TRY_COMPILE([#include <nlist.h>],
|> 
|> it seems useless: Just check HAVE_NLIST_H (the code in 2.50 does
|> define the symbol).  As a matter of fact, looking at the version in
|> the fileutils, I see:
|> 
|> # ifndef NLIST_STRUCT
|> #  if HAVE_NLIST_H
|> #   define NLIST_STRUCT
|> #  endif
|> # endif

According to getloadavg.c, NLIST_STRUCT actually means more than that:

   NLIST_STRUCT                 Include nlist.h, not a.out.h, and
                                the nlist n_name element is a pointer,
                                not an array.

|> Can't getloadavg.c be simplified instead?  What is its master?

There is a file /gd/gnu/loadavg/getloadavg.c on fencepost which seems to
be the master, although the emacs version is even newer.  The fileutils
version has to changes that do not appear in the other files:

  * some locale support
  * default NLIST_STRUCT from HAVE_NLIST_H

Otherwise it seems to be derived from an older version of the fencepost
file.  The emacs version has two further changes from the fencepost
version:

  * don't infer NLIST_STRUCT from system defines
  * fix #ifdef FD_SETFD typo.

IMHO the changes should be merged.

Andreas.

-- 
Andreas Schwab                                  "And now for something
address@hidden                          completely different."
SuSE Labs, SuSE GmbH, Schanzäckerstr. 10, D-90443 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5