[patch] Don't let virtual calls and dynamic casts call MarkVTableUsed()

Reid Kleckner rnk at google.com
Tue Jan 13 11:44:18 PST 2015


On Tue, Jan 13, 2015 at 11:24 AM, Reid Kleckner <rnk at google.com> wrote:

> On Tue, Jan 13, 2015 at 11:22 AM, Nico Weber <thakis at chromium.org> wrote:
>
>> On Tue, Jan 13, 2015 at 11:11 AM, Reid Kleckner <rnk at google.com> wrote:
>>
>>> I'm concerned that if you don't mark the vtable used enough, then
>>> codegen will crash trying to emit a vtable for a class that it assumed
>>> would be marked used because it's compiling a virtual call to a method of
>>> such a class. However, it looks like Rafael completely nuked the
>>> available_externally vtable emission optimization in r189852, which I
>>> forgot about. The vtable code still has lots of rigging to allow
>>> available_externally vtable emission, though.
>>>
>>
>> At the moment, the only thing in coding adding to the DeferredVTables
>> vector in codegen is getAddrOfVTable(), and that's only called for structor
>> body emission (and apple kext vcalls). So I think this should be fine.
>>
>
> Right, but we used to call it more, prior to PR13124 and r189852. I'm
> reading that bug to see if it's something we want to add back.
>

I'm starting to think that even if we want to bring back the
available_externally vtable optimization, it should only happen
optimistically and should not rely on Sema to mark the vtable used. Looking
at PR13124, though, that seems difficult...

Anyway, I drop my objections. The available_externally vtable optimization
doesn't exist and turns out to be very hard to implement in Clang today. =/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20150113/88deee23/attachment.html>


More information about the cfe-commits mailing list