CVS MinGW (Crossed-)Native Compiler is Toast

Bryce McKinlay bryce@mckinlay.net.nz
Mon Sep 29 02:21:00 GMT 2003


On Saturday, Sep 27, 2003, at 20:51 Pacific/Auckland, Mohan Embar wrote:

> (GCJ people: this is the last post on this subject I'll do here. I 
> wanted to
> follow up on this so people knew the outcome.)
>
>> I think I have a faint idea of one of the things that seems to be
>> going wrong.
>>
>> Line 10300 of parse.y has this call:
>>
>>      list = lookup_method_invoke (0, identifier_wfl, type, 
>> identifier, args);
>>
>> Upon entry into this function, some prologue code seems to be
>> executed in gcc/gcc/config/i386/cygwin.asm which trashes
>> the first parameter, changing it from a 0 to garbage. It's a downward
>> spiral from there.
>
> The problem appears to be similar to this:
>
> http://www.cygwin.com/ml/cygwin/2002-06/msg00122.html
>
> When I hack the Makefile to use -mno-stack-arg-probe, the problem
> disappears.
>
> This one definitely qualified as cruel and unusual punishment.
>
> How should I follow up on this?

Presumably this is a new bug since 3.3, so see if you can figure out 
when the problem started and what patch caused it. Then you'll know who 
to blame ;-). If there's nothing obvious in the ChangeLogs, there's no 
shame in resorting to the "binary search" technique.

Regards

Bryce.




More information about the Java mailing list