Bug 103255

Comment 1 Edwin Groothuis 2006-09-14 09:32:44 UTC

State Changed
From-To: open->feedback

Awaiting maintainers feedback

Comment 2 Pav Lucistnik 2006-10-07 16:27:29 UTC

State Changed
From-To: feedback->open

Assign to maintainer 

Comment 3 Pav Lucistnik 2006-10-07 16:27:29 UTC

Responsible Changed
From-To: freebsd-ports-bugs->mi

Assign to maintainer

I get what seems to be this same problem, but running FreeBSD 5.5.

Here's gdb output from an amd64:

   ---[OK]  ---/custrtrn/Test_widestrs
cintltst in free(): error: page is already free

Program received signal SIGABRT, Aborted.
0x0000000801c8a308 in kill () from /lib/libc.so.5
(gdb) bt
#0  0x0000000801c8a308 in kill () from /lib/libc.so.5
#1  0x0000000801c7f931 in raise () from /lib/libc.so.5
#2  0x0000000801cf87f6 in abort () from /lib/libc.so.5
#3  0x0000000801c96d05 in ldexp () from /lib/libc.so.5
#4  0x0000000801c96d48 in ldexp () from /lib/libc.so.5
#5  0x0000000801c978f5 in ldexp () from /lib/libc.so.5
#6  0x0000000801c97ee5 in ldexp () from /lib/libc.so.5
#7  0x0000000801c98005 in free () from /lib/libc.so.5
#8  0x0000000800b8169b in u_strFromWCS_3_6 () from /usr/ports/devel/icu/work/icu/source/lib/libicuuc.so.36
#9  0x000000000048c737 in Test_WCHART_LongString ()
#10 0x000000080073a3f1 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#11 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#12 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#13 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#14 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#15 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#16 0x000000080073a3a3 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#17 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#18 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#19 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#20 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#21 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#22 0x000000080073a3d0 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#23 0x000000080073a3a3 in iterateTestsWithLevel () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#24 0x000000080073a5e8 in runTests () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#25 0x000000080073aa0d in processArgs () from /usr/ports/devel/icu/work/icu/source/tools/ctestfw/libicutest.so.36
#26 0x000000000042d43a in main ()
(gdb)


-- 
Rasmus Kaj <rasmus@kaj.se> -- http://www.stacken.kth.se/~kaj/
ACRONYM: A Contrived Reduction Of Nomenclature Yielding Mnemonics

According to the ICU developers, this is a gcc bug... Could you try rebuilding 
with gcc-4.x? Add the following 3 lines to the port's Makefile -- right after 
the bsd.port.pre.mk:

.if ${OSVERSION} < 601000  
USE_GCC=	4+
.endif

Thanks!

	-mi

On Wed, Oct 18, 2006 at 10:35:58AM -0400, Mikhail Teterin wrote:
> According to the ICU developers, this is a gcc bug... Could you try rebuilding 
> with gcc-4.x? Add the following 3 lines to the port's Makefile -- right after 
> the bsd.port.pre.mk:
> 
> .if ${OSVERSION} < 601000  
> USE_GCC=	4+
> .endif

# make build
"Makefile", line 39: Malformed conditional (${OSVERSION} < 601000)
"Makefile", line 41: if-less endif
make: fatal errors encountered -- cannot continue


Bye,

Gergely Czuczy
mailto: gergely.czuczy@harmless.hu

-- 
Weenies test. Geniuses solve problems that arise.

> # make build
> "Makefile", line 39: Malformed conditional (${OSVERSION} < 601000)
> "Makefile", line 41: if-less endif
> make: fatal errors encountered -- cannot continue

Please, send me the complete Makefile -- by direct e-mail. Thanks!

	-mi

ÓÅÒÅÄÁ 18 ÖÏ×ÔÅÎØ 2006 13:53, Gergely CZUCZY ÎÁÐÉÓÁ×:
> > Add the following 3 lines to the port's Makefile -- right after
> > the bsd.port.pre.mk:
> >
> > .if ${OSVERSION} < 601000 
> > USE_GCC=4+
> > .endif
>
> # make build
> "Makefile", line 39: Malformed conditional (${OSVERSION} < 601000)
> "Makefile", line 41: if-less endif
> make: fatal errors encountered -- cannot continue

Contrary to my request above, you added the three lines BEFORE, rather than 
AFTER the bsd.port.pre.mk line... Please, try again. Thanks!

	-mi

Hello,

> According to the ICU developers, this is a gcc bug... Could you try
> rebuilding with gcc-4.x?

For this bug :
---
 cintltst in free(): error: page is already free
 Abort trap (core dumped)
---
It is ok with GCC 4 (gcc version 4.0.4 20061005 (prerelease) [FreeBSD]) and 
CFLAGS = -O2 or with the default gcc (gcc version 3.4.6) but with 
CFLAGS= -O. I run FreeBSD 6.2

But i've got another problem because of my locale fr_FR :
# locale
LANG=fr_FR.ISO8859-15
LC_CTYPE="fr_FR.ISO8859-15"
LC_COLLATE="fr_FR.ISO8859-15"
LC_TIME="fr_FR.ISO8859-15"
LC_NUMERIC="fr_FR.ISO8859-15"
LC_MONETARY="fr_FR.ISO8859-15"
LC_MESSAGES="fr_FR.ISO8859-15"
LC_ALL=fr_FR.ISO8859-15

The test fails with :
   ---OK:   CollationMonkeyTest
         ERROR : The junk is the same as the French collator.

      ---ERRORS (1) in TestProperty

--------------------------------------
Errors in total: 1.
            TestProperty
         CollationAPITest
      collate

Looking the source, i think it is normal, the test gets a default locator 
and compare it to a french locator : there are the same on my system.

file test/intltest/apicoll.cpp  
void
CollationAPITest::TestProperty(/* char* par */)
[...]

   junk = Collator::createInstance(abcd, success);

    if (U_FAILURE(success))
    {
        errln("Junk collation creation failed, should at least return 
default.");
        delete col;
        return;
    }

    delete col;
    col = Collator::createInstance(success);
    if (U_FAILURE(success))
    {
        errln("Creating default collator failed.");
        delete junk;
        return;
    }

    doAssert(((RuleBasedCollator *)col)->getRules() == ((RuleBasedCollator 
*)junk)->getRules(),
               "The default collation should be returned.");
    Collator *frCol = Collator::createInstance(Locale::getFrance(), 
success);
    if (U_FAILURE(success))
    {
        errln("Creating French collator failed.");
        delete col; delete junk;
        return;
    }

    doAssert((*frCol != *junk), "The junk is the same as the French 
collator.");

------------

So if i revert my locale to "C" (setenv LC_ALL C), all the tests run ok

Regards.

On Sunday 22 October 2006 07:13, Patrick Lamaizière wrote:
= But i've got another problem because of my locale fr_FR :

This looks like a problem with the ICU itself (rather than the FreeBSD 
port) -- could you, please, report it to the ICU maintainers?

	http://bugs.icu-project.org/cgi-bin/icu-bugs

Thanks!

	-mi

Le dimanche 22 octobre 2006 17:16, Mikhail Teterin a écrit :
> = But i've got another problem because of my locale fr_FR :
>
> This looks like a problem with the ICU itself (rather than the FreeBSD
> port) -- could you, please, report it to the ICU maintainers?
>
> 	http://bugs.icu-project.org/cgi-bin/icu-bugs

I did it:
http://source.icu-project.org/cgi-bin/icu-bugs/incoming?id=5473;user=guest

For the FreeBSD port, is-it possible to set the locale to "C" in the 
Makefile as a work around ?

Regards.

On Sunday 22 October 2006 13:01, Patrick Lamaizière wrote:
= For the FreeBSD port, is-it possible to set the locale to "C" in the 
= Makefile as a work around ?

It is, of course... However, this would hide the bugs like this and thus 
partially defeat the purpose of the testing.

If you know, this particular test is wrong -- I can add a patch for the test 
to the port, but if we run the tests in "C" only, we will not catch a non-C 
locale bug in the future.

	-mi

Hi

I had the same problem until I added -fno-strict-aliasing to my CFLAGS in
/etc/make.conf

I have experimented with settings and the following worked for me

CFLAGS= -O2 -pipe -funroll-loops -fomit-frame-pointers -fno-strict-aliasing

This one worked but I have not tested omitting the others and rebuilding

:-)
Torben

hi,

-fno-strict-aliasing does the job. you can omit all of the other flags.
to be honest i am not completely sure that this is the best way to go. 
anyway. in case i will face any postgres faults i will check the status 
of this bug(?) again. :]

andras

Hi all

I have done some cleanup of my make.conf (from a previous post) so I don't
get too many strange build errors because of my ICU hack.

The following is a snip from my new make.conf and should give you all an
idea of how to make different options depending of the build.

   ...

   #  Build settings for ports collection
   .if ${.CURDIR:M*/usr/ports/*}
   CFLAGS= -O2 -pipe -funroll-loops
   CXXFLAGS= -O2 -pipe -funroll-loops
   .endif

   # Build settings for base system
   .if ${.CURDIR:M*/usr/src/*} || ${.CURDIR:M*/usr/obj/*}
   CFLAGS+= -O2 -pipe -funroll-loops -ffast-math
   CXXFLAGS+= -O2 -pipe -funroll-loops -ffast-math
   COPTFLAGS= -O2 -pipe -funroll-loops -ffast-math
   .endif

   ...

   # Hack to build ICU
   .if ${.CURDIR:M*/usr/ports/devel/icu}
   CFLAGS+= -fno-strict-aliasing
   .endif

   ...

This snippet gives you different options for making the base system and the
ports collection. At the end there is a hack where I add the -fno-strict
aliasing to the ICU port only -- the hack as I originally posted it will
most likely break other builds.

I have no idea why it works -- it was just frustrated trial and error until
I could make ICU. Afterwards I was so happy I had to share it with you guys
and gals.



Med venlig hilsen / Kind Regards


Torben Jakobsen
Consultant


____________________________________________


(Embedded image moved to file: pic05758.gif)
International Business Machines A/S
Cell phone: +45 4120 1195
E-mail: torben.jakobsen@dk.ibm.com
Web: www.ibm.com

On Monday 06 November 2006 05:00, Torben Jakobsen wrote:
=	# Build settings for base system
=	.if ${.CURDIR:M*/usr/src/*} || ${.CURDIR:M*/usr/obj/*}
=	CFLAGS+= -O2 -pipe -funroll-loops -ffast-math
=	CXXFLAGS+= -O2 -pipe -funroll-loops -ffast-math
=	COPTFLAGS= -O2 -pipe -funroll-loops -ffast-math
=	.endif

Ok, using fast-math in the base is, probably, wrong. But that's not
related to ICU.

=	# Hack to build ICU
=	.if ${.CURDIR:M*/usr/ports/devel/icu}
=	CFLAGS+= -fno-strict-aliasing
=	.endif

AFAIK, -fno-strict-aliasing is part of CFLAGS by default... Anyway,
can you tell me, what the final CFLAGS are in icu?

	make -C /usr/ports/devel/icu -V CFLAGS -V CXXFLAGS

Thanks!

	-mi

Comment 17 dfilter service 2006-11-07 19:15:55 UTC

mi          2006-11-07 19:14:25 UTC

  FreeBSD ports repository

  Modified files:
    devel/icu            Makefile 
  Added files:
    devel/icu/files      patch-intltest 
  Log:
  Add a patch to fix test-failures in some "exotic" locales (obtained from
  ICU):
  
  PR:     102629
  
  Make sure to use not only "-O2", but also "-fno-strict-aliasing"
  -- but only when using the "stock" cc/c++ as the newer gcc-4.x does not
  seem to exhibit a problem. This seems to help the test-exposed
  miscompilations.
  
  PR:     103255
  
  Revision  Changes    Path
  1.23      +8 -2      ports/devel/icu/Makefile
  1.1       +64 -0     ports/devel/icu/files/patch-intltest (new)
_______________________________________________
cvs-all@freebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/cvs-all
To unsubscribe, send any mail to "cvs-all-unsubscribe@freebsd.org"

Comment 18 Mikhail Teterin 2006-11-07 19:16:53 UTC

State Changed
From-To: open->closed

Using -fno-strict-aliasing is essential. The gcc-3.x is 
known to be buggy without this. Port updated to add the 
flag automatically, when using the "stock" cc and c++. 

Thanks to all for investigating.