Re: [patch]: Replace stryoul call to fetch address

[Pedro Alves <palves@redhat.com>]

On 02/27/2013 06:38 PM, Corinna Vinschen wrote:

>> IIRC, the matching Cygwin code that that special
>> Cygwin signals handling was never implemented, or it was disabled
>> on Cygwin, or some such, and that gdb bits is actually causing
>> trouble -- see http://sourceware.org/ml/gdb-patches/2013-02/msg00122.html.
>> We should just zap it all.
> 
> I don't know exactly.  Probably cgf would be able to answer this
> better than me.  

I think I recall some email from him long about about the
cygwin side of that, but he wasn't against removing these
bits of on the gdb end in that url above, so...

> In fact I have a SEGV right now which I still have
> to investigate, but I'm working on GDB only as a side job.

Sure.  I just meant to point out that that's exactly the code
you're touching.  ;-)

>> One thing that comes to mind is I think we'll need to have separate
>> mingw64/cygwin64 osabis.  Currently, mingw 32/64 use
>> GDB_OSABI_CYGWIN, and that limps along, but with LP vs LLP, that
>> won't work.
> 
> Why not?  The only difference between the x86_64 Cygwin and Mingw ABI
> is the sizeof long.  And that's noted in the dwarf debug info.

Not every use use of the target's "long" goes through the debug info.  I
see uses of  gdbarch_long_bit and builtin_long in the expression
machinery, for example in c-exp.y, for handling integer constants,
or in eval.c, for type promotion.  x64's long is fixed to 32-bit
in amd64_windows_init_abi.

> Apart
> from a strange crash when trying to load stripped executables, I'm
> using a x86_64 Mingw GDB to debug x86_64 Cygwin DLL and binaries.
> I'm not sure this single difference justifies distinct OSABIs.

Sure, we'll limp along.  But there are cases that bypass debug info.
A distinct OSABI seems like the proper mechanism to me.  Time
will tell.  ;-)

> 
>> We'll need a way to distinguish Cygwin vs native Windows
>> binaries.  Probably by checking for cygwin.dll in the dll import list?
> 
> A check for cygwin1.dll is already done in windows_make_so.

Yeah, though that's only after the dll is loaded.  I was thinking
of looking at import list, when the executable is loaded in gdb,
hence before the program is run.

> 
>> Another point where that'd be good is in step-over-longjmp support
>> (gdbarch_get_longjmp_target).  IIRC, the offset of PC within
>> the jmpbuf of msvcrt.dll is not the same as Cygwin's (32-bit;
>> dunno about 64-bit).
> 
> Cygwin 64 bit uses the same jmpbuf layout as native Windows.  Only
> two members are used differently, but those should not be used by
> GDB anyway.

OK.

> Ok, I apply it then with only the int changed to unsigned.

Thanks.

-- 
Pedro Alves