[Bug libc/15756] New: Check ABI for completeness

[Bug libc/15756] New: Check ABI for completeness

[schwab@linux-m68k.org]

http://sourceware.org/bugzilla/show_bug.cgi?id=15756

            Bug ID: 15756
           Summary: Check ABI for completeness
           Product: glibc
           Version: 2.18
            Status: NEW
          Severity: normal
          Priority: P2
         Component: libc
          Assignee: unassigned at sourceware dot org
          Reporter: schwab@linux-m68k.org
                CC: drepper.fsp at gmail dot com

There should be a check that an architecture has all the expected interfaces in
its abilist files, to avoid repeating the mistakes with fallocate64 in 2.10
(missing for all 32-bit archtectures) or mcount in 2.17 (missing for aarch64).

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15756] Check ABI for completeness

[joseph at codesourcery dot com]

http://sourceware.org/bugzilla/show_bug.cgi?id=15756

--- Comment #1 from joseph at codesourcery dot com <joseph at codesourcery dot com> ---
Such a tools may well be useful to find existing problems and for checking 
new architectures (with testing, overlapping tests for similar things can 
often be useful), but as I said in 
<http://sourceware.org/ml/libc-alpha/2013-07/msg00380.html> I think this 
is mainly an API issue not an ABI one.

Taking bug 15666 (alpha __sqrt*_finite) as an example, the real bug isn't 
the missing symbols, it's that calling certain functions in code built 
with -ffinite-math-only would have resulted in undefined symbols.  And 
it's not a bug that IA64 doesn't have those symbols, instead having its 
own bits/math-finite.h that avoids needing them.  Nor is it a bug that 
architectures where long double functions are aliases for double functions 
don't have *l_finite at all - bits/math-finite.h ensures the right 
functions get used in that case.  If anything, it's an ABI mistake that 
IA64 *does* have the *_finite symbols for Bessel functions (since it gets 
those functions from generic libm), even though its bits/math-finite.h 
won't use them.  (Since those are part of the public ABI now, the only 
sensible fix would be to make IA64 bits/math-finite.h use those functions 
rather than completely disabling *_finite.)

For all of those architecture variations that aren't bugs, a test that 
verifies calling functions with -ffinite-math-only works would work 
smoothly in the presence of such variations without needing any special 
cases, while detecting the actual bug on alpha before it was fixed, but a 
test based on comparing ABI lists would need special cases or manual 
architecture maintainer review to determine that the differences are OK.

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug libc/15756] Check ABI for completeness

[fweimer at redhat dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=15756

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
              Flags|                            |security-

-- 
You are receiving this mail because:
You are on the CC list for the bug.

[Bug build/15756] Check ABI for completeness

[jsm28 at gcc dot gnu.org]

https://sourceware.org/bugzilla/show_bug.cgi?id=15756

Joseph Myers <jsm28 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |carlos at redhat dot com
          Component|libc                        |build

-- 
You are receiving this mail because:
You are on the CC list for the bug.