cppexp: map member function pointers

[wilsonk]

This is c++11 specific, but including files like require c++11 support. This moves things along a little further.

[Syniurge]

Thanks Kelly. The UO_Deref mapping looks correct, I'll check the other two this evening. DelegateFuncptrExp sounds strange tbh, but I don't know much about member function pointers. I'll wrap my head around them by checking occurrences of BO_PtrMemD and BO_PtrMemI in Clang's Parse and Sema, see when they are generated and follow what Clang does with them.

[Syniurge]

Hey Kelly I had a tiring week and still haven't looked properly at member function pointers.

Do the latest commits break anything in your tests?

[wilsonk]

Hello Elie,

No, none of the tests break…things seem to get a little further along with a couple of them that use ‘-std=c++11’ (so algorithm and array, etc.), but they don’t even compile completely even with that fix. So I can just keep working on getting things further along here with c++11 specific stuff, when needed.

It really isn’t a big deal, I can just keep it in my local repo. I understand those tiring weeks…had a few of them doing my degrees :). Just get to it whenever you can as it isn’t a core C++ item anyways.

I will pull your TypeQualifiedBuilder changes sometime today and test them here. Need to do some other work first though, :)

Thanks,

Kelly

From: Elie Morisse [mailto:[email protected]]
Sent: Friday, February 27, 2015 11:37 AM
To: Syniurge/Calypso
Cc: wilsonk
Subject: Re: [Calypso] cppexp: map member function pointers (#4)

Hey Kelly I had a tiring week and still haven't looked properly at member function pointers.

Do the latest commits break anything in your tests?

—
Reply to this email directly or view it on GitHub #4 (comment) .

[Syniurge]

Hey Kelly,

I learned today that BO_PtrMemD and BO_PtrMemI aren't for member function pointers, they're for any "member pointer type", and a value with a "member pointer type" is just an offset.

What the spec says:
http://www.open-std.org/jtc1/sc22/open/n2356/expr.html#expr.mptr.oper

And Clang has some examples:
https://github.com/llvm-mirror/clang/blob/master/test/SemaCXX/member-pointer.cpp
https://github.com/llvm-mirror/clang/blob/master/test/SemaCXX/member-pointers-2.cpp

[wilsonk]

Hello Elie,

I didn’t see the specifics below when I was looking at this. I just saw the clang error and a post on a newsgroup that seemed to indicate these were for pointers to member functions. I guess I should have looked into this further, sorry.

So now the question is…is there a D equivalent to these?

Point 6 in the spec seems to indicate that a pointer to a function could be the result of this pointer-to-member operator, does it not? With an opCall being used on the result? I guess that is what the newsgroup post was referring to. In this single case can we just translate to an opCall with the result as the operand?

Thanks,

Kelly

From: Elie Morisse [mailto:[email protected]]
Sent: Saturday, February 28, 2015 6:00 PM
To: Syniurge/Calypso
Cc: wilsonk
Subject: Re: [Calypso] cppexp: map member function pointers (#4)

Hey Kelly,

I learned today that BO_PtrMemD and BO_PtrMemI aren't for member function pointers, they're for any "member pointer type", and a value with a "member pointer type" is just an offset.

What the spec says:
http://www.open-std.org/jtc1/sc22/open/n2356/expr.html#expr.mptr.oper

And Clang has some examples:
https://github.com/llvm-mirror/clang/blob/master/test/SemaCXX/member-pointer.cpp
https://github.com/llvm-mirror/clang/blob/master/test/SemaCXX/member-pointers-2.cpp

—
Reply to this email directly or view it on GitHub #4 (comment) .

[Syniurge]

BO_PtrMemD is for « LHS.*RHS » and BO_PtrMemI is for « LHS->*RHS », where LHS is an object/pointer and RHS is a "member pointer" value, which is basically an offset. AFAIK there is no equivalent in D to "member pointer types":

http://clang.llvm.org/doxygen/classclang_1_1MemberPointerType.html

so in cpptypes.cpp they are mapped to ptrdiff_t. Since you still have access to the MemberPointerType when mapping expression the type of the result of the BO_PtrMemD/I expression is known (it's RHS.getPointeeType()). Other than that I'm not sure what would be the appropriate mapping for these expressions but you don't need opCall, it's not related.

[wilsonk]

Hello Elie,

I added a pull request for that extra build script I talked about a couple weeks back. Let me know if you want me to add some notes to the readme on how it works, if anything is unclear. I named it BUILD so that it wouldn’t be confused with the build.sh file in the root testing dir. Maybe BUILD_STL would be better? Or something else? Anyways, change as you see fit…I am just used to BUILD for my testing scripts :)

Sorry, I haven’t had much time lately to look into anything else. Hopefully I can find some more time soon.

Thanks,

Kelly

From: Elie Morisse [mailto:[email protected]]
Sent: Tuesday, March 03, 2015 8:48 AM
To: Syniurge/Calypso
Cc: wilsonk
Subject: Re: [Calypso] cppexp: map member function pointers (#4)

BO_PtrMemD is for « LHS.RHS » and BO_PtrMemI is for « LHS->RHS », where LHS is an object/pointer and RHS is a "member pointer" value, which is basically an offset. AFAIK there is no equivalent in D to "member pointer types":

http://clang.llvm.org/doxygen/classclang_1_1MemberPointerType.html

so in cpptypes.cpp they are mapped to ptrdiff_t. Since you still have access to the MemberPointerType when mapping expression the type of the result of the BO_PtrMemD/I expression is known (it's RHS.getPointeeType()). Other than that I'm not sure what would be the appropriate mapping for these expressions but you don't need opCall, it's not related.

—
Reply to this email directly or view it on GitHub #4 (comment) .

[Syniurge]

Mapping and sending back member function pointers are fully implemented for the Itanium ABI.

They join C++ vtable generation as features that need to be ported to the MSVC ABI.