Re: [PATCH] win64 support for libffi (2/2)

[Dave Korn <dave dot korn dot cygwin at googlemail dot com>]

Andrew Haley wrote:
> Dave Korn wrote:
>> Andrew Haley wrote:
>>> NightStrike wrote:
>>>> When are you planning to do this?  I want to time some other changes
>>>> in my personal repo with this change.
>>> Today, if I have time, otherwise next week.
>>   I have an outstanding patch to libffi/src/x86/win32.S
>>
>> http://gcc.gnu.org/ml/java-patches/2009-q2/msg00084.html
> 
> Please commit this.

  Two things:

a)  Tests not quite finished yet, but fixed lots of FAILs.  Only other change:

-FAIL: Thread_Wait_2 -findirect-dispatch output - source compiled test
-FAIL: Thread_Wait_Interrupt output - source compiled test
+FAIL: Thread_Interrupt -findirect-dispatch output - source compiled test
+FAIL: Thread_Wait_2 output - source compiled test
+FAIL: Thread_Wait_Interrupt -O3 -findirect-dispatch output - source compiled test

... which I think is probably noise.  The Process_N tests all hang, I think
there is a bug somewhere in the underlying cygwin pthread support (there is
some ever-present noise in libstdc++ tests relating to mutexes as well, which
supports this theory).

b)  Just the libffi, or both parts?  I haven't tested the libjava
runtimeInitialized change separately.

>>   There is currently one difference between our win32.S and upstream libffi
>> win32.S, where upstream adds a new entrypoint, _ffi_closure_STDCALL.
>>
>>   Would it help if I added that function and EH annotations for it to my patch?
> 
> Yes.  Consider it pre-approved.

  Thanks, under way.  I think I have to duplicate out the function tail-end
code that the current upstream version shares with _ffi_closure_SYSV (from
.Lcls_return_result on) because as far as I've been able to figure out so far,
we can't represent shared/overlapping ranges like this in FDEs.  It's not a
huge amount of code, but if you do happen to know a way to do it, please send
me a pointer, or confirm that it's still OK with the code duplication.

    cheers,
      DaveK