Re: Apropos recent debuginfo-related cygwin failures: "Dwarf Error: Could not find,> abbrev number 1"

[Kai Tietz <ktietz70 at googlemail dot com>]

2009/6/29 Dave Korn <dave.korn.cygwin@googlemail.com>:
> Kai Tietz wrote:
>
>> I agree. This behaviour is new and seems to be a regression. I tested
>> the testcases with old gcc (4.4.1) and didn't got those message. When
>> I use gcc 4.5 I see them for cygwin and mingw targets. Dave I tried
>> also your proposed patch for binutils, but this doesn't seems to help
>> here. It is true, that we can honor in coffcode.h the case that
>> NO_LOAD is wished (I sent a patch for this already), but it doesn't
>> solves this described failure for my tests.
>
> ?Is it possible that the coffcode.h change is not needed and it was adding
> the debug_pubtypes section to the link script that is the only fix we actually
> need? ?I'm testing now with the updated script templates but coffcode.h
> reverted. ?It fixes the ld-undefined FAILs on Cygwin; it also fixed my "a.exe:
> Permission denied" error that I just got from trying to bootstrap GCC head
> using old and known good binutils; I'm now going to try a full bootstrap using
> my reverted binutils and see if it works.
>
> ? ?cheers,
> ? ? ?DaveK
>

Well, I investigated here bit more deepth. The real issue is the
setting of the flag IMAGE_SCN_CNT_INITIALIZED_DATA (SEC_DATA). If this
is removed, no dwarf2 failures are getting reported. I assume that
current code can't handle SEC_DATA | SEC_DEBUGGING at the same time.
But this is an incompatibility to pe-coff spec, but I think we have to
life by that as long as we find the reason, why setting SEC_DATA for
debugging sections confuses dwarf2 reader.

Cheer,
Kai

-- 
|  (\_/) This is Bunny. Copy and paste
| (='.'=) Bunny into your signature to help
| (")_(") him gain world domination