Hi Sheri,
thank you for testing JIT and providing feedback for us. (and thanks for everyone else as well who are spent their time on testing JIT). Since the JIT is more dependent on the platfrom than a generic C code there is a much higher probability of various platform dependent errors, and I can test only a limited number of systems. To make JIT work everywhere I really depend on you, guys!
You are right the RunTest.bat was not updated. The strange thing is that its current state shouldn't call JIT at all since its support is totally missing from RunTest.bat and the basic tests does not force JIT compilation. This is true for forcing pcre_study. Could you send me the differences as well?
pcre_jit_test isa different story. The failed entry should be { 0, 0, "ab$", "ab\r\n" } in the regression_test_cases array. Could you send me a static, debug, standalone binary (in privately) which contains only this failing test case? If there is no external library dependency, I could debug it on my machine.
Regards,
Zoltan
PS: the library increase is not surprising. The JIT is a heavyweight optimization.
Sheri <silvermoonwoman@???> írta:
>I rebuilt RC1 without JIT support and when I run the RunTest script from >
the Msys console for that build, the script has no issues. It is only >
when built with JIT support I have the issue described. The attachment >
must have been too big for the list, so I sent it separately to Philip. >
Also I played with test3 and found that in order for it to run under >
Msys, the French locale would need to be specified with "fr" (not >
"fr_FR" or "french").>
>
Regards,>
Sheri>
>
> I built 8.20-RC1 on Windows XP in Msys/MINGW32 (simulated Unix environment)>
> configured with Cmake 2.8 using my usual options plus JIT support.>
>>
> pcretest -C reports:>
>>
> PCRE version 8.20-RC1 2011-09-12>
> Compiled with>
> UTF-8 support>
> Unicode properties support>
> Just-in-time compiler support>
> Newline sequence is ANYCRLF>
> \R matches CR, LF, or CRLF only>
> Internal link size = 2>
> POSIX malloc threshold = 10>
> Default match limit = 10000000>
> Default recursion depth limit = 10000000>
> Match recursion uses heap>
>>
> The size of my pcre shared library went from 289KB to 453KB>
>>
> In previous versions I was able to use the included RunTest scripts>
> from the Msys console without exception (aside from test3 which never>
> runs at all from the Runtest script due I think to an erroneous grep>
> or sed error for "french"). Also RunTest.bat executed in the Windows>
> command environment always worked fine (including test3). Running>
> RunTest.bat in the Windows cmd environment still works fine and>
> reports no discrepancies for 8.20-RC1. The bat file should be>
> updated as it only runs tests 1 thru 12.>
>>
> However there are numerous differences for 8.20-RC1 if I run the>
> RunTest sh script from my Msys console (in tests 1, 2, 4 and 11).>
> I manually renamed the interim testtry files to capture them and have>
> attached a zip file of the output from those tests. The same tests>
> operate successfully via RunTest.bat using Windows cmd. As I said>
> previously, the RunTest sh scripts of previous versions didn't have>
> these issues.>
>>
> Also, running the pcre_jit_test program apparently reports a discrepancy.>
> Among its output series of dots in the output is this:>
>>
> Return value differs(-1:1): 'ab$' @ 'ab>
> '>
> Successful test ratio: 99%>
>>
>
>
-- >
## List details at
https://lists.exim.org/mailman/listinfo/pcre-dev >
