GNU clisp 2.49.90 available for beta-testing

[Bruno Haible]

Hi Sébastien,

> I made a new upload incorporating your patches:
> 
>  https://buildd.debian.org/status/package.php?p=clisp&suite=experimental

And I made a new prerelease:
  https://haible.de/bruno/gnu/clisp-2.49.92.tar.bz2

It incorporates the memory-map patches, adds memory-map patches also for
  arm
  powerpc
  powerpc64
  sparc64
and a couple of other fixes.

> - armel: lisp image generated, but fails to compile asdf (memory mapping issue)
> - ppc64el: lisp image generated, but fails to compile asdf (segfault)

These MAY be fixed in 2.49.92, due to the memory-map fixes.
If not, please try adding -DSAFETY=3 to the CFLAGS for these architectures.

> > > - armhf: can't run lisp image (PSEUDOCODE_ALIGNMENT is not fulfilled)
>
> Concerning armhf, note that the Debian package used to add the -marm flag in
> order to disable thumb mode, but I removed it this for this upload.

With this remark, you put me exactly on the right track. I had not understood
why function pointers were all odd, despite me giving -falign-functions=4.
And I had not understood the effect of -marm vs. -mthumb and that different
GCC builds have a different default in this respect.

This is now fixed in 2.49.92.

> When I reintroduce -marm, I get a segfault on the first run of the lisp.run binary

Fixed as well. It turned out, I was accidentally doing an mmap() that
annihilated the entire stack. This of course bombed out immediately.

> except for
> s390x where there is now a segfault (but clisp actually never compiled on s390x
> in Debian).

I'm doing regular testing in Debian 8.6 for s390x, both in 32-bit mode and in
64-bit mode, and it has been rock-solid for months.

Bruno