Re: [PATCH] Enable visibility annotations

On 04/20/2016 12:22 PM, Pedro Alves wrote:
> On 04/20/2016 07:02 AM, Yury Gribov wrote:
>
>> Pedro,
>>
>> Do you think the above is doable for gdb?
>
> I don't know -- I'm already juggling too many balls in the
> air, and I'm afraid that if that depends on me, it'll take a
> while before I'll manage to try.
>
> Since gdb's use of private symbols is not an isolated incident,
> I don't think we should try to clean up all packages that make
> use of private symbols, in a rush.  Instead, I think readline
> needs to take a staged approach:
>
> - Export all the private symbols that programs are using today.
>    Simply accept today's reality.  The dynamic symbol table of
>    libreadline.so still shrinks, precedent for visibility
>    annotations is still set, and private symbol leakage
>    is contained, because future software will no longer be
>    able to abuse all the other private symbols that are
>    not exported.

Agreed, that's my plan for now.

> - Claim that the symbols may no longer be available in a
>    future release.

You mean just email respective package maintainers?

> - Give time for packages to clean themselves up, and propose
>    any necessary new replacement APIs.

This would require significant expertise in readline though...

> - Optionally, in the release after the next, mark the symbols
>    as deprecated with __attribute__((deprecated)), so packages
>    that abuse private symbols get a build-time warning.

That won't help as these symbols are not present in headers anyway.  All 
users have their own private declarations.

> - In some future release, stop exporting the symbols.
>
>> That would of course leave the
>> problem of linking older gdb with new readline which will need to be
>> resolved by distros.
>
> Right.
>
> Thanks,
> Pedro Alves