Re: [Gcl-devel] Does the --enable-ansi config option work?

[Camm Maguire]

Greetings!

Yes, thanks for pointing this out!  This is our first example of a
'gotcha' with dlopen.  In general, one cannot build images which load
.o files and save them with save-system when using dlopen.  Instead,
one compiles the .o files, and uses the system linker (ld or gcc) to
link the final image.  Its just an extra step, as far as I can see,
but conventional lisp usage often invokes the save-system apparently. 

As of this point, there are a few architectures which *must* use
dlopen, and therefore cannot build the ansi image as currently
compiled.  So we either have to make a relatively simple change to the
pcl and clcs makefiles to use ld in (likely) all cases for simplicity,
or generate the bfd patches for these other architectures so that they
can relocate objects without dlopen.  I'd prefer the latter, but we
shall see if it proves too much work.

BTW, I'm under the impression that Windows still has full
functionality in this regard, as it apparently is able to use the
relocation code in sfasl.c, which refers to a relocate() function
provided in rel_coff.c.  So this should not explain Mike's difficulty
with the ansi build on Windows.  BTW, what is the status here?



"Paul F. Dietz" <address@hidden> writes:

> Camm Maguire wrote:
> 
> > OK, this is the infamous buggy redhat gcc snapshot, never intented for
> > release by the gcc people.  Please try *any* other released gcc
> > version -- 2.95, 3.0, or 3.1.  I'll wager this clears everything up,
> > but if not, kindly please compile gcl with -g (in linux.defs), omit
> > the -fomit-frame-pointer flag, and then run the compilation command
> > governing macros.lisp in gdb, and report where the segfault is.
> > 
> > Come to think of it, I've just increased the level of optimization
> > passed to gcc by gcl in the command-line flags.  If I recall, there
> > was a work around for this buggy redhat gcc by lowering the
> > optimization level.  You could also try that, if you want -- take '-O6
> > -fomit-frame-pointer' down to '-O2' or '-O', for example.  But in
> > general, gcl probably shouldn't waste too much time supporting this
> > mistaken gcc release.
> 
> I had configured with --enable-ansi --enable-dlopen.  When I tried
> just --enable-ansi the build succeeded.
> 
> I'm getting a machine with Mandrake 8.2 preinstalled in a few days,
> so I'll rebuild on that when it arrives.
> 
>       Paul
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah