Re: PATCH: Spell exp2l correctly

[Roland McGrath <roland at redhat dot com>]

> I'm afraid I'm not knowledgeable enough to help with that; I don't 
> understand well enough all the ways in which GLIBC has been built and 
> deployed.  

As if anyone does... ;-)

> The exp2l function is listed in libm.abilist, but the (bogus) expl2 does
> not appear.

We know that expl2 has not been exported before.  (The same typo would have
to have appeared in math/Versions as well.)  Thanks for reminding me about
the abilist file; I'd forgotten about my own invention that answers my
question!

exp2l appears in libm.abilist under:

GLIBC_2.1 i.86-.*-linux.*/notls i.86-.*-linux.*/thread i.86-.*-linux.*/tls m68.*-.*-linux.*/notls
| GLIBC_2.2.5 x86_64-.*-linux.*/tls
| GLIBC_2.2 ia64-.*-linux.*/tls

i.e., exp2l does not appear in the abilist for configurations other than
i.86, x86_64, ia64, and m68.*.  (Though in fact only a few past
configurations are represented in the abilist files.  When I created them,
I left it mostly to the architecture maintainers to send me updates, and
few did.)

So, we need to figure out the story for configurations not represented in
libm.abilist now.  powerpc{,64} and sh[34] look to be the only other ones
there, that don't have an exp2l export (i.e., they were using the generic
file with its typos).  As things work now, we'd have to add a
sysdeps/CPU/Versions file for each affected platform to put exp2l in the
new version set.  Or else, put the GLIBC_2.1 { exp2l; } in each known-good
platforms' sysdeps/CPU/Versions, and change it in the generic file (which
might be more in line with the sysdeps/... code selection for exp2l that
leads to the difference in the first place).  

This all presupposes we decide we care.  We could just decide that new
applications using exp2l won't ever care about running on old libraries on
the affected platforms, and punt.


Thanks,
Roland