[cfe-commits] [Windows] Use thiscall as the default calling convention for class methods

[Timur Iskhodzhanov]

r160121, thanks!

On Thu, Jul 12, 2012 at 1:04 PM, Timur Iskhodzhanov <timurrrr at google.com> wrote:
> On Thu, Jul 12, 2012 at 12:33 AM, John McCall <rjmccall at apple.com> wrote:
>> On Jul 11, 2012, at 1:29 PM, Timur Iskhodzhanov wrote:
>>> On Wed, Jul 11, 2012 at 10:00 PM, John McCall <rjmccall at apple.com> wrote:
>>>> On Jul 11, 2012, at 4:53 AM, Timur Iskhodzhanov wrote:
>>>>> On Mon, Jul 9, 2012 at 9:46 PM, John McCall <rjmccall at apple.com> wrote:
>>>>>> On Jul 9, 2012, at 6:52 AM, Timur Iskhodzhanov wrote:
>>>>>>> On Sat, Jul 7, 2012 at 10:44 AM, John McCall <rjmccall at apple.com> wrote:
>>>>>>>> On Jun 28, 2012, at 1:39 PM, Timur Iskhodzhanov wrote:
>>>>>>>>> Part one is attached - it propagates the "isInsance" boolean to the
>>>>>>>>> two places where it should be read afterwards [in part 2].
>>>>>>>>
>>>>>>>> I didn't really like this approach, so I committed a different one as r159893.
>>>>>>>> Let me know if this helps.
>>>>>>> You've meant r159894.
>>>>>>> Yeah, I like your approach much more than mine :)
>>>>>>>
>>>>>>>>> The only logic change is around lib/CodeGen/CGCall.cpp:150.
>>>>>>>>> Without this change, the definition of `void __cdecl cdecl_method() {}`
>>>>>>>>> ends up using "thiscall" (and the call remains "cdecl" - ouch!).
>>>>>>>>> Probably it was a bug in the first place.
>>>>>>>>
>>>>>>>> Let me know if this is still a problem arising in your second patch.
>>>>>>> It still is!
>>>>>>> See the FIXME in the attached patch.
>>>>>>> Other than that, the tests pass OK and it was easy to achieve this.
>>>>>>
>>>>>> Aha, that's interesting.  Okay, I think the isVariadic bit should probably be
>>>>>> passed down to getDefaultMethodCallConv().  While you're at it, could
>>>>>> you rename that to getDefaultCXXMethodCallConv()?
>>>>> Done.
>>>>> Actually, I've found and fixed a mangler bug when adding this argument :)
>>>>> See the attached patch
>>>>
>>>> Looks good, thanks!
>>> bug_12785p2_2.patch is OK
>>
>> Yes.
>>
>>> On Wed, Jul 11, 2012 at 9:57 PM, John McCall <rjmccall at apple.com> wrote:
>>>> On Jul 11, 2012, at 5:34 AM, Timur Iskhodzhanov wrote:
>>>>> With the attached patch the cdecl test passes now.
>>>>>
>>>>> However, I'm not very sure what I'm doing here and why it doesn't work
>>>>> in the first place - can you please take a look?
>>>>
>>>> Hmm.  The problem here is that getCanonicalCallConv is turning CC_C
>>>> into CC_Default.  That needs to be disabled in Microsoft mode, just like
>>>> it apparently is in MRTD mode.
>>>>
>>>> ...also, the way that MRTD mode is implemented is wrong, and we should
>>>> be handling this in IR-generation the same that we're handling the
>>>> thiscall rules.
>>> a) bug_12785p3.patch is OK as a short-term solution and
>>>    getCanonicalCallConv can be fixed later
>>> OR
>>> b) I should just fix getCanonicalCallConv as a part of my patch?
>>>
>>> If (b) - should I fix it completely (with fixing "is implemented wrong")
>>> or just alter one condition in the Microsoft mode?
>>> I'm not sure it's clear to me how to fix it "completely" at this point
>>> (to be honest, I haven't looked at its code yet - just checking e-mail)
>>
>> (b).  Just alter the one condition in Microsoft mode;  this will make it
>> so that function types that are explicitly cdecl are canonically different
>> from function types that do not have an explicit CC.  This should make
>> your proposed patch unnecessary.
> Got it.
> Attached is bug_12785p3_2.patch which does pretty much what you've
> asked for instead of bug_12785p3.patch
> For some reason, LangOpts.MicrosoftMode is false with "-cxx-abi
> microsoft", so I'm just checking for the ABI.
> I can change that post-commit if that doesn't sound good for you.
>
>> If you want to fix the existing MRTD stuff, I'd be happy to provide guidance,
>> but that's not really "on you", so to speak.
> I can do this if you believe that'd benefit me (like knowing the
> codebase better).
> Otherwise, I have a lot of "-cxx-abi microsoft" work to do :)
>
>> John.