problem building a GCJ-friendly GNU Crypto

Fri Sep 19 15:47:00 GMT 2003

On 2003年9月19日, Mark Wielaard wrote:
> Finally found the patch that broke GNU Crypto:
> http://gcc.gnu.org/ml/gcc-patches/2003-09/msg00848.html
> But I don't have a simple testcase yet. With this patch the GNU crypto
> test-suite gives 857 of 984 tests failed. Reverting it makes all tests
> pass.
>> I am hoping that the patch author or someone on the gcj mailinglist
> looks at this patch and immediately sees something obviously wrong for
> java. If not, we must try to come up with a simple test case that shows
> what goes wrong.

Hi Mark,
I've looked over the patch again and it all looks reasonable. The major
difference is that fold is now more aggressive recursively calling itself
after each transformation. So the GNU crypto problem may be one of the
other transforms in "fold" that is now getting invoked where it wasn't
previously. [This also might explain which MIN_EXPR and MAX_EXPR started
appearing recently!]
A reduced test case would be great. There were no new regressions with
the libjava testsuite. Does anyone know if there have been any recent
regression in Mauve or Jacks that could be related? They may be easier
to analyse.
I'm working hard to make fold java-safe, so I'd rather fix the errant
transformation in fold rather than remove calls to it from the java
front-end.
My apologies for any inconvenience.
Roger
--
Roger Sayle, E-mail: roger@eyesopen.com
OpenEye Scientific Software, WWW: http://www.eyesopen.com/
Suite 1107, 3600 Cerrillos Road, Tel: (+1) 505-473-7385
Santa Fe, New Mexico, 87507. Fax: (+1) 505-473-0833