[cfe-dev] Generated AST for implicit CXXConstructorDecl

[Richard Smith via cfe-dev]

On Tue, May 10, 2016 at 6:50 AM, Mikhail Ramalho via cfe-dev <
cfe-dev at lists.llvm.org> wrote:

> I almost forgot:
>
> I know that we can get the VarDecl from the DeclRef (by using getDecl() or
> getFoundDecl()). What I meant by "if I can generate the AST with the
> varDecls" was actually an AST with the explicit VarDecls of __i0, __i1,
> etc.
>

The VarDecls are in the AST, you can access them by calling
getArrayIndices() on the CXXCtorInitializer.


> Thank you,
>
>
> 2016-05-10 14:45 GMT+01:00 Mikhail Ramalho <mikhail.ramalho at gmail.com>:
>
>> Hello all,
>>
>> I'm using libTooling to generate the AST of a given program (that will
>> later be translated to our own internal AST), and I just found this odd
>> CXXConstructorDecl for the following program:
>>
>> class X
>> {
>>   public:
>>     X() {};
>>
>>     char str[20][20][230];
>> };
>>
>> int main(void)
>> {
>>   X x;
>>   X x1 = x;
>>   return 0;
>> }
>>
>>
>> The implicit constructor for X is:
>>
>> CXXConstructorDecl 0x3959fd0 <line:3:7> col:7 implicit used X 'void
>> (const class X &) throw()' inline
>> |   |-ParmVarDecl 0x395a110 <col:7> col:7 used 'const class X &'
>> |   |-CXXCtorInitializer Field 0x3908c48 'str' 'char [20][20][230]'
>> |   | `-ImplicitCastExpr 0x395a6e8 <col:7> 'char' <LValueToRValue>
>> |   |   `-ArraySubscriptExpr 0x395a6c0 <col:7> 'const char' lvalue
>> |   |     |-ImplicitCastExpr 0x395a6a8 <col:7> 'const char *'
>> <ArrayToPointerDecay>
>> |   |     | `-ArraySubscriptExpr 0x395a560 <col:7> 'char const[230]'
>> lvalue
>> |   |     |   |-ImplicitCastExpr 0x395a548 <col:7> 'char const (*)[230]'
>> <ArrayToPointerDecay>
>> |   |     |   | `-ArraySubscriptExpr 0x395a400 <col:7> 'char
>> const[20][230]' lvalue
>> |   |     |   |   |-ImplicitCastExpr 0x395a3e8 <col:7> 'char const
>> (*)[20][230]' <ArrayToPointerDecay>
>> |   |     |   |   | `-MemberExpr 0x395a298 <col:7> 'char
>> const[20][20][230]' lvalue .str 0x3908c48
>> |   |     |   |   |   `-DeclRefExpr 0x395a270 <col:7> 'const class X'
>> lvalue ParmVar 0x395a110 '' 'const class X &'
>> |   |     |   |   `-ImplicitCastExpr 0x395a368 <col:7> 'unsigned int'
>> <LValueToRValue>
>> |   |     |   |     `-DeclRefExpr 0x395a340 <col:7> 'unsigned int' lvalue
>> Var 0x395a2e0 '__i0' 'unsigned int'
>> |   |     |   `-ImplicitCastExpr 0x395a4c0 <col:7> 'unsigned int'
>> <LValueToRValue>
>> |   |     |     `-DeclRefExpr 0x395a498 <col:7> 'unsigned int' lvalue Var
>> 0x395a438 '__i1' 'unsigned int'
>> |   |     `-ImplicitCastExpr 0x395a620 <col:7> 'unsigned int'
>> <LValueToRValue>
>> |   |       `-DeclRefExpr 0x395a5f8 <col:7> 'unsigned int' lvalue Var
>> 0x395a598 '__i2' 'unsigned int'
>> |   `-CompoundStmt 0x395a748 <col:7>
>>
>> Which should look this (I think):
>>
>> X(const X& ref) : str(ref.str[__i0][__i1][__i2]) {}
>>
>> Or maybe:
>>
>> X(const X& ref) { str(ref.str[__i0][__i1][__i2]); }
>>
>> (None of the above is compiled)
>> (If we call emit-llvm, it's translated into a memcpy)
>>
>> Anyway, there are a few things missing on this AST, in my opinion:
>>
>> 1. No VarDecl for __i0, __i1, __i2.
>> 2. No loop to iterate and copy the array elements (a call to memcpy would
>> be good as well, but I don't expect that clang will generate that).
>>
>> Does anyone knows if I can generate the AST with the varDecls and the
>> loop? I found a method on SemaDeclCXX that seems to build a loop, it's
>> called buildSingleCopyAssignRecursively. It's called
>> by buildSingleCopyAssign, that even tries to build a memcpy.
>>
>> Am I out of luck and will have to go through all the implicit
>> generated CXXConstructorDecls and build a loop myself or is there an
>> alternative?
>>
>> Thank you,
>>
>> --
>>
>> Mikhail Ramalho.
>>
>
>
>
> --
>
> Mikhail Ramalho.
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20160510/2b41c3ba/attachment.html>