[PATCH] Add MicrosoftVFTableContext to AST

[John McCall]

On Jul 26, 2013, at 11:04 AM, Timur Iskhodzhanov <timurrrr at google.com> wrote:
> 2013/7/25 John McCall <rjmccall at gmail.com>:
>> On Jul 25, 2013, at 10:20 AM, Timur Iskhodzhanov <timurrrr at google.com>
>> wrote:
>> 
>> 2013/7/25 John McCall <rjmccall at gmail.com>:
>> 
>> On Jul 22, 2013, at 6:11 AM, Timur Iskhodzhanov <timurrrr at google.com> wrote:
>> 
>> +// The main differences are:
>> +//  1. Separate vftable and vbtable.
>> +//  2. Each non-primary base class that has new virtual methods gets its
>> +//     own vfptr and vftable, all the address points are zero.
>> 
>> This is not a difference.
>> 
>> 
>> As far as I understand, in Itanium ABI each subobject that adds new
>> virtual methods to its bases gets a new *address point* in the shared
>> vtable, but not a new vtable (at least no new sections are generated),
>> which IS different from the Microsoft ABI.
>> 
>> 
>> I’m not sure what it means for a subobject to "add virtual methods to its
>> bases”.
> 
> Rephrased to:
> //  2. Each subobject with a vfptr gets its own vftable rather than an address
> //     point in a single vtable shared between all the subobjects.
> 
> Does this make sense now?

Oh, yes, good point..  That fact that distinct vf-tables are allocated as
separate symbols is an excellent thing to mention.

>> In both ABIs, the algorithm for performing a virtual function call is to
>> adjust
>> the base pointer to a subobject which contains the method in its primary
>> v-table, load a function pointer from the v-table, and call it.
>> 
>> In your example, the difference is just that, under MSVC, B doesn’t have
>> an entry for f() in its vf-table.  (In fact, it doesn’t have a vf-table.)
>> So the
>> caller has to adjust to something that actually does have an entry for f(),
>> namely A.
>> 
>> +  // See if this class expands a vftable of the base we look at, which is
>> either
>> +  // the one defined by the vfptr base path or the primary base.
>> 
>> Still not sure why you’re not just starting from the right base and then
>> walking up the primary base chain.
>> 
>> 
>> That's not enough even in the simple case of "B: A".
>> 
>> The vfptr is in the A layout, so the "right base" is A.
>> If we don't go to the most derived class (B) from "the right base"
>> (A), we forget to add the more derived class's own new methods (and
>> probably return-adjusting thunks).
>> 
>> 
>> My point is that you should just start recursing from B instead of this
>> weird
>> combination of walking the path and then falling back on climbing the
>> primary base chain.
>> 
>> Finding B (the most-derived subobject that A is in the primary-base chain
>> of)
>> should be really easy — it’s just a depth-first search through the complete
>> object’s layout, stopping at the first thing with the same offset as A.
> 
> Hm...
> 
> Let's consider
> --------
>  struct A {
>    virtual TYPE* f();
>  };
>  struct B {
>    virtual TYPE* g();
>  };
>  struct C: A, B {  <something>  }
> --------
> We'll have two vfptrs: for A at offset 0 and for B at offset 4
> (assuming 32-bit arch).
> 
> Currently, for the B's vftable we'll do this:
>  enter AddMethods(C)
>    enter AddMethods(B)
>    allocate a vftable slot for B::g
>    leave AddMethods(B)
>  update the B::g slot with this/return adjustment if C overrides it
>  leave AddMethods(C)
> [this somewhat reflects how Itanium's VTableContext works]
> 
> Basically what you want is
>  enter AddMethods(B)
>  allocate a vftable slot for B::g,
>    if we have a (final) overrider for g() in C,
>      calculate this/return adjustment right here*.
>  leave AddMethods(B)
> 
> Ok, so at the (*) stage we can probably find the FinalOverrider to
> just write the adjustments immediately...
> This implies rewriting ComputeThisOffset to just take the final
> overrider and return the this offset in a complete object.
> There were a few minor complexities in rewriting it (e.g. API being
> not convenient) that have overloaded my brain on Friday evening
> though.
> 
> Do you think this is important for the first version? If so - I'll
> continue trying to do this next week.

I think it’s worth it, thanks.

I’ll wait to review that unless there’s something intermediate about
the current patch you’d like me to check out.

John.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-commits/attachments/20130726/cdc38da9/attachment.html>