[6.3] Drop xm-*.h from GDB - move to pure autoconf

[6.3] Drop xm-*.h from GDB - move to pure autoconf

[Andrew Cagney]

Hello,

Since 1995 (4.15 according to NEWS) GDB's been using autoconf, and since 
2000, GDB's had systems (NetBSD/PPC was first) that did not need an 
xm-*.h file.

With it now >=9 years since autoconf's introduction and with autoconf 
proving that it has made the xm files redundant, it's time to cut our 
losses and remove any remaining xm file support.

At this stage, the only issue I know of blocking the xm file removal, is 
DOS file support.  Both DJGPP and CYGWIN appear to have configury not 
yet handled by autoconf.

For 6.3, it would be good to see GDB:

- have DJGPP and CYGWIN use autoconf for DOS file support
- remove all xm-*.h files
- (oh and) switch to autoconf 2.5x

Andrew

Re: [6.3] Drop xm-*.h from GDB - move to pure autoconf

[Eli Zaretskii]

> Date: Thu, 29 Jul 2004 20:04:48 -0400
> From: Andrew Cagney <cagney@gnu.org>
> 
> At this stage, the only issue I know of blocking the xm file removal, is 
> DOS file support.  Both DJGPP and CYGWIN appear to have configury not 
> yet handled by autoconf.

??? Could you spell out the problems?

There are dozens of GNU projects, all of them using Autoconf (and also
libtool, for that matter), that were ported to DJGPP by running the
configure script with minimal changes (akin the config.sed script used
by GDB).  I'm not aware of any problems that would stop the show for
you, but perhaps I'm missing something.

> - have DJGPP and CYGWIN use autoconf for DOS file support

What do you mean by that?  What ``DOS file support'' in GDB needs to
be autoconfiscated?

Re: [6.3] Drop xm-*.h from GDB - move to pure autoconf

[Andrew Cagney]

>>Date: Thu, 29 Jul 2004 20:04:48 -0400
>>> From: Andrew Cagney <cagney@gnu.org>
>>> 
>>> At this stage, the only issue I know of blocking the xm file removal, is 
>>> DOS file support.  Both DJGPP and CYGWIN appear to have configury not 
>>> yet handled by autoconf.
> 
> 
> ??? Could you spell out the problems?
> 
> There are dozens of GNU projects, all of them using Autoconf (and also
> libtool, for that matter), that were ported to DJGPP by running the
> configure script with minimal changes (akin the config.sed script used
> by GDB).  I'm not aware of any problems that would stop the show for
> you, but perhaps I'm missing something.

Right.

>>> - have DJGPP and CYGWIN use autoconf for DOS file support
> 
> 
> What do you mean by that?  What ``DOS file support'' in GDB needs to
> be autoconfiscated?

Check the contents of config/i386/xm-{cygwin,djgpp}.h.  Instead of being 
detected by autoconf, those values are being hardwired.  Easy to fix (I 
just fixed HOST_FLOAT_FORMAT).

Andrew

Re: [6.3] Drop xm-*.h from GDB - move to pure autoconf

[Eli Zaretskii]

> Date: Fri, 30 Jul 2004 09:23:48 -0400
> From: Andrew Cagney <cagney@gnu.org>
> 
> Check the contents of config/i386/xm-{cygwin,djgpp}.h.  Instead of being 
> detected by autoconf, those values are being hardwired.  Easy to fix

Ah, that...  Your message made it sound as if there's some inherent
problem that prevents autoconfiscation, so I was surprised.

Yes, it should be easy to fix, except for the gdb.ini part: if one
builds the DJGPP port on Windows or cross-compiles on Unix, the
filesystem will allow that name.  So the test needs to be dependent on
the fact that a DJGPP port is being built, no matter what the
underlying OS has to say about .gdbinit.

Note that include/filenames.h has some constants similar to the `;' vs
`:' stuff, so maybe some of GDB sources could use the results of
autoconfiscation instead.

Re: [6.3] Drop xm-*.h from GDB - move to pure autoconf

[Andrew Cagney]

>>Date: Fri, 30 Jul 2004 09:23:48 -0400
>>> From: Andrew Cagney <cagney@gnu.org>
>>> 
>>> Check the contents of config/i386/xm-{cygwin,djgpp}.h.  Instead of being 
>>> detected by autoconf, those values are being hardwired.  Easy to fix
> 
> 
> Ah, that...  Your message made it sound as if there's some inherent
> problem that prevents autoconfiscation, so I was surprised.
> 
> Yes, it should be easy to fix, except for the gdb.ini part: if one
> builds the DJGPP port on Windows or cross-compiles on Unix, the
> filesystem will allow that name.  So the test needs to be dependent on
> the fact that a DJGPP port is being built, no matter what the
> underlying OS has to say about .gdbinit.

I've zapped most of the xm files, but the cygwin/djgpp ones are still there.

> Note that include/filenames.h has some constants similar to the `;' vs
> `:' stuff, so maybe some of GDB sources could use the results of
> autoconfiscation instead.

Yes.

Andrew