Re: COMMITTED: Add support for STT_IFUNC

[Ian Lance Taylor <iant at google dot com>]

Tristan Gingold <gingold@adacore.com> writes:

> On Dec 5, 2008, at 6:19 AM, Ian Lance Taylor wrote:
>
>> Tristan Gingold <gingold@adacore.com> writes:
>>
>>> I haven't look at the implementation, but is a STT_IFUNC allowed to
>>> have relocations
>>> against STT_IFUNC symbols ?
>>>
>>> If not, this is must be checked at link time, otherwise the entries
>>> must be topologically ordered.
>>
>> I don't see why a topological ordering is required.  It suffices to
>> resolve the relocations lazily on demand.  Of course, a circular
>> reference will cause an error or an infinite loop.
>
> Humm, right.  But better to detect this at link time rather than
> unexpectedly at run time.

I suppose, but it's easy to construct a case using conditionals which
appears to be circular based on relocation information but actually
works in practice.  So it would be hard to detect--you would have to
do a walk through relocations and sections--and it could only ever be
a warning.

Ian