Re: PATCH: BZ #14664: changing prefix breaks "make check-abi-libc"

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Sun, Dec 9, 2012 at 8:36 PM, Carlos O'Donell <carlos@systemhalted.org> wrote:
> On Sun, Dec 9, 2012 at 8:00 PM, H.J. Lu <hjl.tools@gmail.com> wrote:
>> Hi,
>>
>> Size of _nl_default_dirname depends on the size of prefix.  This patch
>> updates libc.abilist from
>>
>> _nl_default_dirname D 0x12
>>
>> where the size of prefix (/usr) is 4, to
>>
>> _nl_default_dirname D 0xe + sizeof_prefix
>>
>> and changes scripts/abilist.awk to compute _nl_default_dirname size as
>>
>> [size of _nl_default_dirname symbol - length of prefix] + sizeof_prefix
>>
>> Any comments?
>
> Could you please explain how this is different than just deleting the symbol?

--prefix=XXX is the part of glibc ABI.  Glibcs with different prefixes are
orthogonal to each other.  You can install multi glibcs with different
prefixes on the same machine at the same.  Programs compiled
for one glibc won't use the other glibcs at run-time since ld.soes
are installed in different places.  That is the whole idea of configuring
glibc with --prefix=XXX other than --prefix=/usr.

>
> Your changes make the test for this particular symbol a tautology.
>
> You check to see that the size is the size you set which is always true.

Not if you change definition of _nl_default_dirname in glibc source
or there is a compile bug which changes it.

> The check you want is to see if the size you set matches the expected ABI size.

Exactly.

> It just so happens that Andi's ABI is different because he used
> different inputs to the build.

My patch makes sure that all glibcs configured with the same --prefix=XXX
have the same size of _nl_default_dirname.

> No matter what we do here the only way for the test to pass is to
> support allowing Andi to describe his ABI and use that as the
> baseline.
>


-- 
H.J.