This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Update powerpc ABI data


On Tue, 1 May 2012, Roland McGrath wrote:

> > Most of them were also wrong on x86 and sparc, but David only updated
> > the abi lists that he could test, I suppose.  Most of the missing
> > symbols are related to the long double switch.  The abi files of the
> > other architectures will need similar updating.
> 
> The files certainly were right at one time (like when I wrote all this
> stuff to begin with), and certainly changes like the long double switch
> should only have affected newer symbol versions, not preexisting ones at
> the time.  I wonder if there was some bug in the merge/extract scripts
> so the data David extracted isn't what was actually in use originally.

The extraction was (I presume) using exactly the same script that was used 
before by make check-abi to extract the data to compare against the lists 
of symbols from the newly built libraries.  Before

commit 00bbd29b35717f0de8b97e5b1e6fd3e979808ec3
Author: Ulrich Drepper <drepper@gmail.com>
Date:   Sat Jan 7 19:23:45 2012 -0500

    Update ABI information

the data hadn't been updated (other than adding one empty file) since 
April 2003 (between 2.3.2 and 2.3.3).  So only changes to 2.3.2 or older 
symbol versions would necessarily indicate problems (as opposed to the 
January update not being based on real data for most architectures).

I do think it's useful to compare the ABI baselines for architectures you 
maintain against old binaries of various past glibc versions, to see if 
the current baselines have differences in old symbol versions to what's in 
the old binaries - I've done that for ARM and MIPS (details of the results 
on libc-ports).  Even where the binaries are from distributors with local 
patches applied that might include backports of new symbols and symbol 
versions, such comparisons are still a useful starting point for looking 
for ABI breakages that crept in while check-abi wasn't maintained or run 
by default.

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]