Re: [musl] musl libc version

[Dmitry Samersoff <dms@samersoff.net>]

Hello Rich,

 > With that said, ability to test for some things is presently
 > inadequate, and there's a proposal I've made (based on previous
 > discussions with others) on libc-coord to improve this in a cross-libc
 > manner:

I like your proposal, but I'm in doubt that it's possible to coordinate 
it across all involved parties.

Absence of musl/libc-version.h (similar to gnu/*) and corresponding api 
creates a burden for support folks (as Alexander already mentioned) and 
people maintaining cross platform software like JVM.

 > If you have input into what kinds of things you need to be able to
 > test for it would be very helpful.

*For example*, on another topic we discussed a difference in behavior of 
dlopen() between glibc and musl. if sometimes musl will change its 
behavior, workaround code in java launcher could be deprecated and 
placed under if musl_version() < ...

*Another example is less technical but more effort driven*, absence of 
dlvsym() in musl leads to significant changes in libnuma handling code 
in JVM.
   Numa support have big performance impact and all related changes have 
to be carefully tested in a different environment.
   So I would prefer to put these changes under if musl() (or ever 
#ifdef MUSL) rather than test for dlvsym() presence at runtime. It 
allows me to be sure that we are always on the known and tested scenario.

-Dmitry

On 27.05.2020 18:57, Rich Felker wrote:
> On Wed, May 27, 2020 at 06:44:38PM +0300, Alexander Scherbatiy wrote:
>> Hello,
>>
>> Using glibc it is possible to get GNU C library version using
>> confstr() function and _CS_GNU_LIBC_VERSION constant.
>>
>> In musl the confstr() returns EINVAL for _CS_GNU_LIBC_VERSION.
>> Is there a way to get libc version in musl?
> 
> Can you describe what you're trying to achieve with a version number?
> musl specifically doesn't publish version numbers via macros or
> runtime interfaces because they're a poor proxy for checking for the
> particular feature/property/whatever you care about. For instance,
> distros may backport the thing you want to an older version, and even
> if not, you end up hard-coding knowledge about particular libcs that
> quickly bitrots rather than making something robust and future-proof.
> 
> With that said, ability to test for some things is presently
> inadequate, and there's a proposal I've made (based on previous
> discussions with others) on libc-coord to improve this in a cross-libc
> manner:
> 
> https://www.openwall.com/lists/libc-coord/2020/04/22/1
> 
> If you have input into what kinds of things you need to be able to
> test for it would be very helpful.
> 
> Rich
>