middle-end/3544: SJLJ exceptions with dwarf2 debug broken

Richard Earnshaw rearnsha@arm.com
Tue Jul 3 07:17:00 GMT 2001

>Number:         3544
>Category:       middle-end
>Synopsis:       SJLJ exceptions with dwarf2 debug broken
>Confidential:   no
>Severity:       critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          wrong-code
>Submitter-Id:   net
>Arrival-Date:   Tue Jul 03 07:16:01 PDT 2001
>Originator:     Richard Earnshaw
>Release:        3.0
System: SunOS sun18 5.7 Generic_106541-11 sun4u sparc SUNW,Ultra-5_10
Architecture: sun4

host: sparc-sun-solaris2.7
build: sparc-sun-solaris2.7
target: arm-unknown-elf
configured with: /home/rearnsha/thirdparty/gcc/configure --prefix=/home/rearnsha/thirdparty/install --target=arm-elf --with-headers=/home/rearnsha/thirdparty/utils/newlib/libc/include --enable-languages=c,c++,objc,f77
	Building libstdc++-v3 with an arm-elf cross compiler generates object 
	files that reference non-existent internal labels from the unwind 
	information tables.  This seems to be occuring because dwarf2out.c
	and except.c maintain distinct counters for the funcdef number and
	that these are getting out of step.

	In dwarf2out (dwarf2out_begin_prologue) we have:
		fde->funcdef_number = current_funcdef_number;

	and (output_call_frame_info) has:

	but in except.c (output_function_exception_table) we have

		if (! cfun->uses_eh_lsda)

		funcdef_number = (USING_SJLJ_EXCEPTIONS
		    ? sjlj_funcdef_number
		    : current_funcdef_number);

	hence the generated label and the fde->funcdef_number can get out
	of step if not all functions use lsda information.

	Build libstdc++-v3/libsupc++/eh_aux_runtime.cc for arm-unknown-elf
	with the gcc-3.0 release.  Note that the resulting object file 
	defines the labels LSDA0 & LSDA1, but that the exception unwind
	table references LSDA1 and LSDA2.  The resulting object file will
	fail to link.
	No idea.

More information about the Gcc-bugs mailing list