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] Don't check ABI for _nl_default_dirname


On Sat, Dec 8, 2012 at 1:29 PM, Andi Kleen <andi@firstfloor.org> wrote:
>> The discussion above concluded that it would be a better idea to make
>> the abi check a little smarter, than to remove tests.  It would be
>> nicer if the abilist file format is extended to denote data symbols
>> that are a function of configuration variables and adjust
>> expectations.  It's a non-trivial task but I'd think of that as the
>> way forward.
>
> But that is what my patch does, doesn't it?

Yes, you are correct, it does.

I'm sorry if we didn't communicate our requirements very clearly.

Please let my try again.

I'd say a new solution has the following properties:

(a) Check the presence or absence of the symbol.
(b) Check for the symbol type e.g. data, function etc.
(c) Check the size of the symbol.
(d) Allow overrides.
    (d.1) Allow the user to provide data which overrides
          the defaults.
    (d.2) Provide defaults for expected builds such that
          it just works if your build follows the normal build
          procedures.

One implementation might be to add a new --with-abilist= option
that allows the developer to define a small file with overrides
for their own ABI. This data is used as part of (d.1) and thus
allows the tests to pass.

It is a *conscious* choice on the part of the user to create
an override abilist, and they use that to ensure the test results
for the ABI are clean.

Does that make sense?

Cheers,
Carlos.


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