Re: A running list of questions from "porting" Slackware to musl

[Weldon Goree <weldon@langurwallah.org>]

On 09/30/2014 09:16 PM, Rich Felker wrote:
>
> Like glibc, musl supports the gettext API natively. There is no need
> for libintl and it should not be installed. libintl.h is just the
> header for the gettext API.
>
> I'm not sure what's happening with symbol names in your libintl. I
> suspect it was either built incorrectly or has some invalid glibc
> assumptions, but I know other users (even Alpine Linux) have used it
> successfully with musl in the past, before musl had gettext.
>

Nor I, but I've tried it with all the permutations of older gettext 
versions and no gettext versions, and several GNU programs fail to link 
if they aren't told to --disable-nls. I'm not a programmer, just a 
sysadmin, and I'm interested in musl because it might make my life 
easier in the future, but I'm in no place (yet) to figure out what I've 
done wrong.

> I don't think gcc/ld search /lib when linking, so putting libc.so
> there (and not having it in /usr/lib) would probably leave things
> broken. If the core issue is that you want the actual file on /, you
> could swap the direction of the symlink (make ld-musl the real file)
> or make both ld-musl and /usr/lib/libc.so symlinks to /lib/libc.so or
> similar.
>

When I cross-compile a native gcc it "wants" the linker to be in /lib 
(or I should say, Gregor's patches tell it to look for the symlink in 
/lib, and my being a sysadmin tells me that whatever /lib/ld-foo links 
to should not be in /usr because I do have to boot my server occasionally).

>
> If you're "porting" the whole system to build against musl, you
> shouldn't need the musl-gcc wrapper, and in fact you're going to have
> a hard time using it as soon as you try to build C++ software. You
> really want a native compiler toolchain that supports musl. See the
> musl-cross project for the necessary patches to GCC.

Granted: my slackware port doesn't use musl-gcc, but I'm maintaining the 
community port of musl for slackware-gnu users too.

>
>> 4. Is Pth a lost cause?
>
> Yes. But there's a new portable pth (sorry, couldn't resist) in the
> works that several projects (iirc gnupg) are eagar to switch to, since
> pth is so bad and has no future (it's incompatible with programs that
> use real threads, and likely with most/all future systems). It's
> API-compatible but based on pthread so it works anywhere. No idea what
> the status is, though.
>

I'll look into it. xorg was happy with the stub lib they wrote years ago 
so I haven't looked past that because it did what I needed; that was 
more out of curiosity.

>
> I'd need more context to know what, if anything, you're doing wrong,
> or if the packages you're compiling just insist on using stack
> protector. In any case having it on should not break anything, but you
> may need libssp_nonshared.a (from a gcc natively targeting musl) for
> ssp to link right in shared libraries. libssp.so should not exist, and
> libssp.a should be an empty .a file (ar rc libssp.a to make it).
>

RPM is the candidate this time; I only need it to build rpm2cpio to get 
the linuxdoc sources (I found a workaround), but it also took some 
finagling to get Perl to stop "helping" me by adding -f-stack-protector. 
This was again a point of curiosity.

Cheers,
Weldon