[patch] Fix internal error on canonicalization of clang types [Re: why in cplus_demangle_operators operation sizeof is expected to be followed with space]

[Jan Kratochvil <jan dot kratochvil at redhat dot com>]

On Mon, 17 Sep 2012 07:49:16 +0200, Vasyl Vavrychuk wrote:
> This looks to be not a symbol but type name.

OK, reproduced [attached], patched DWARF of "int" in C++: void f(int i) {}


> But I have found it directly in the binary. Line containing this text looks like

readelf -wi
 <1><59>: Abbrev Number: 4 (DW_TAG_base_type)
    <5c>   DW_AT_name        : exact_unsigned_base_helper<sizeof(unsigned int) * 8>	


> Do you think the problem is in clang? Could you point out some
> specification, so I can reference it while contacting clang people.

Itanium C++ ABI does not say anything about spacing:
	http://www.swag.uwaterloo.ca/acd/docs/ItaniumC++ABI.htm
	(the original URL is 404 now)

I guess so far it is only agreed upon between GCC and GDB, it is true GDB
should accept any valid spacing of the language, this is also what
cp_canonicalize_string is there for.

Your patch would belons to the GCC mailing lists where libiberty/ is
maintained.

But I find more correct patch the one below.

For a testcase could you provide a minimal C++ code to produce such type?

No regressions on {x86_64,x86_64-m32,i686}-fedora18-linux-gnu.


Thanks,
Jan


gdb/
2012-09-17  Jan Kratochvil  <jan.kratochvil@redhat.com>

	Fix internal error on canonicalization of clang types.
	* cp-name-parser.y (exp): Use "sizeof ".

diff --git a/gdb/cp-name-parser.y b/gdb/cp-name-parser.y
index 6767fa5..c57ce64 100644
--- a/gdb/cp-name-parser.y
+++ b/gdb/cp-name-parser.y
@@ -1183,7 +1183,7 @@ exp	:	FLOAT
 	;
 
 exp	:	SIZEOF '(' type ')'	%prec UNARY
-		{ $$ = d_unary ("sizeof", $3); }
+		{ $$ = d_unary ("sizeof ", $3); }
 	;
 
 /* C++.  */

Attachment: type.S
Description: Text document