Change the printing of integer literal to handle more types

http://llvm.org/bugs/show_bug.cgi?id=11179

The StmtPrinter has trouble printing IntegerLiteral if the type is short or uint128, or their unsigned variants.  This change will prevent Clang from complaining on these types.

[Richard Smith, 2011-11-02 05:37:32]

http://codereview.appspot.com/5309045/diff/2001/include/clang/AST/Expr.h
File include/clang/AST/Expr.h (right):

http://codereview.appspot.com/5309045/diff/2001/include/clang/AST/Expr.h#newc...
include/clang/AST/Expr.h:1087: "Interger type cannot be short or unsigned
short");
Typo 'Interger'

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp
File lib/Sema/SemaTemplate.cpp (right):

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4204: bool IsNegative = Value.isNegative();
APInt::isNegative just checks the top bit, which won't do the right thing for
unsigned values. You could grab the full APSInt from the TemplateArgument and
check isSigned() too.

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4206: Value = Value.abs();
I think this will still be negative if Value is SHRT_MIN etc. In such cases, you
may need to switch to the next wider integer type. For LLONG_MIN, I think you'll
need something nasty like -LLONG_MAX - 1.

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4212: }
It looks like we'll build a short literal if an enum has an underlying type of
'short' or 'unsigned short'. Can we move this check after the enum check below?

[rtrieu, 2011-11-03 01:30:38]

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp
File lib/Sema/SemaTemplate.cpp (right):

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4204: bool IsNegative = Value.isNegative();
On 2011/11/02 05:37:33, Richard Smith wrote:
> APInt::isNegative just checks the top bit, which won't do the right thing for
> unsigned values. You could grab the full APSInt from the TemplateArgument and
> check isSigned() too.

Done.

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4206: Value = Value.abs();
On 2011/11/02 05:37:33, Richard Smith wrote:
> I think this will still be negative if Value is SHRT_MIN etc. In such cases,
you
> may need to switch to the next wider integer type. For LLONG_MIN, I think
you'll
> need something nasty like -LLONG_MAX - 1.

Done.

http://codereview.appspot.com/5309045/diff/2001/lib/Sema/SemaTemplate.cpp#new...
lib/Sema/SemaTemplate.cpp:4212: }
On 2011/11/02 05:37:33, Richard Smith wrote:
> It looks like we'll build a short literal if an enum has an underlying type of
> 'short' or 'unsigned short'. Can we move this check after the enum check
below?

Done.