std::use_facet<std::num_put>() not thread safe

• Key: STDCXX-450
• Type: Bug
• Status: Resolved
• Resolution: Cannot Reproduce
• Priority: Blocker
• Assignee: Travis Vitek
• Reporter: Martin Sebor
• Component/s: 22. Localization, Thread Safety
• Affects Version/s: 4.1.2, 4.1.3, 4.1.4
• Fix Version/s: 4.2.0

Description

The 22.locale.num.put.mt thread safety tests aborts at runtime indicating that there is a thread safety issue in the locale implementation. Unfortunately, running the test through the Intel Thread Checker hasn't revealed the cause of the problem since the thread checker exits prematurely, most likely due to a bug in the tool itself (see below).

$ icc --version && tcheck_cl -v && make 22.locale.num.put.mt CXXOPTS=-tcheck LDOPTS=-tcheck && ./22.locale.num.put.mt --nthreads=4 --nloops=10000 || tcheck_cl ./22.locale.num.put.mt --nthreads=4 --nloops=1000000
icc (ICC) 9.1 20070320
Copyright (C) 1985-2007 Intel Corporation. All rights reserved.

Intel(R) Thread Checker 3.1 command line instrumentation driver (24400)
Copyright (c) 2007 Intel Corporation. All rights reserved.
icc -cxxlib-nostd -g -w1 -tcheck -I/amd/devco/sebor/stdcxx/include/ansi -D_RWSTDDEBUG -D_REENTRANT -I/amd/devco/sebor/stdcxx/include -I/build/sebor/stdcxx-icc-9.1_049-15S/include -I/amd/devco/sebor/stdcxx/tests/include -L/build/sebor/stdcxx-icc-9.1_049-15S/rwtest -lrwtest15S -cxxlib-nostd -lpthread -L/build/sebor/stdcxx-icc-9.1_049-15S/lib -tcheck /amd/devco/sebor/stdcxx/tests/localization/22.locale.num.put.mt.cpp /build/sebor/stdcxx-icc-9.1_049-15S/lib/libstd15S.a /build/sebor/stdcxx-icc-9.1_049-15S/rwtest/librwtest15S.a -lstd15S -lcxaguard -lsupc++ -lm -o 22.locale.num.put.mt
icc: warning: thread checking compilation disables optimization

1. INFO (S1) (10 lines):
2. TEXT:
3. COMPILER: Intel C++, _INTEL_COMPILER = 910, __INTEL_COMPILER_BUILD_DATE = 20070320, __EDG_VERSION_ = 306
4. ENVIRONMENT: x86_64/LP64 running linux-elf 2.4.20 with glibc 2.3
5. FILE: 22.locale.num.put.mt.cpp
6. COMPILED: Jun 13 2007, 14:05:31
7. COMMENT: thread safety
   ############################################################

1. CLAUSE: lib.locale.num.put

1. NOTE (S2) (5 lines):
2. TEXT: executing "locale -a > /tmp/tmpfile-qNKfDC"
3. CLAUSE: lib.locale.num.put
4. FILE: process.cpp
5. LINE: 274

1. INFO (S1) (3 lines):
2. TEXT: testing std::num_put<charT> with 4 threads, 10000 iterations each, in locales { "aa_DJ" "aa_DJ.iso88591" "aa_DJ.utf8" "aa_ER" "aa_ER@saaho" "aa_ER.utf8" "aa_ER.utf8@saaho" "aa_ET" "aa_ET.utf8" "af_ZA" "af_ZA.iso88591" "af_ZA.utf8" "am_ET" "am_ET.utf8" "an_ES" "an_ES.iso885915" "an_ES.utf8" "ar_AE" "ar_AE.iso88596" "ar_AE.utf8" "ar_BH" "ar_BH.iso88596" "ar_BH.utf8" "ar_DZ" "ar_DZ.iso88596" "ar_DZ.utf8" "ar_EG" "ar_EG.iso88596" "ar_EG.utf8" "ar_IN" "ar_IN.utf8" "ar_IQ" }
   # CLAUSE: lib.locale.num.put
   
   # INFO (S1) (3 lines):
   # TEXT: exercising std::num_put<char>
   # CLAUSE: lib.locale.num.put
   
   /amd/devco/sebor/stdcxx/src/locale_body.h:242: static _rw::rw_facet::_C_facet_type __rw::rw_locale::_C_get_facet_type(const __rw::_rw_facet &): Assertion '0 != __facet._C_pid' failed.
   ./22.locale.num.put.mt[0x40eb27]
   ./22.locale.num.put.mt[0x40ebff]
   ./22.locale.num.put.mt[0x413760]
   ./22.locale.num.put.mt[0x419cbd]
   ./22.locale.num.put.mt[0x41d0e7]
   ./22.locale.num.put.mt[0x41d263]
   ./22.locale.num.put.mt[0x440ea3]
   ./22.locale.num.put.mt[0x440f51]
   ./22.locale.num.put.mt[0x4088d7]
   ./22.locale.num.put.mt[0x408cad]
   ./22.locale.num.put.mt[0x40d787]
   /lib64/tls/libpthread.so.0[0x353a00610a]
   /lib64/tls/libc.so.6(__clone+0x73)[0x35395c68c3]
   Aborted
   Intel(R) Thread Checker 3.1 command line instrumentation driver (24400)
   Copyright (c) 2007 Intel Corporation. All rights reserved.
   Building project
   Instrumenting
   11% 22.locale.num.put.mt ( All Functions ):.......................................................
   .
   33% libc-2.3.4.so ( Minimal ):....
   44% libcxaguard.so.5 ( Minimal ):..
   55% libdl-2.3.4.so ( Minimal ):..
   66% libgcc_s-3.4.6-20060404.so.1 ( Minimal ):..
   77% libimf.so ( Minimal ):..
   88% libm-2.3.4.so ( Minimal ):..
   100% libpthread-2.3.4.so ( Minimal ):..
   
   Running: /build/sebor/stdcxx-icc-9.1_049-15S/tests/22.locale.num.put.mt --nthreads=4 --nloops=1000000
   
   # INFO (S1) (10 lines):
   # TEXT:
   # COMPILER: Intel C++, _INTEL_COMPILER = 910, __INTEL_COMPILER_BUILD_DATE = 20070320, __EDG_VERSION_ = 306
   # ENVIRONMENT: x86_64/LP64 running linux-elf 2.4.20 with glibc 2.3
   # FILE: 22.locale.num.put.mt.cpp
   # COMPILED: Jun 13 2007, 14:05:31
   # COMMENT: thread safety
   ############################################################
   
   # CLAUSE: lib.locale.num.put
   
   # NOTE (S2) (5 lines):
   # TEXT: executing "locale -a > /tmp/tmpfile-j8uQY8"
   # CLAUSE: lib.locale.num.put
   # FILE: process.cpp
   # LINE: 274
   
   # INFO (S1) (3 lines):
   # TEXT: testing std::num_put<charT> with 4 threads, 1000000 iterations each, in locales { "aa_DJ" "aa_DJ.iso88591" "aa_DJ.utf8" "aa_ER" "aa_ER@saaho" "aa_ER.utf8" "aa_ER.utf8@saaho" "aa_ET" "aa_ET.utf8" "af_ZA" "af_ZA.iso88591" "af_ZA.utf8" "am_ET" "am_ET.utf8" "an_ES" "an_ES.iso885915" "an_ES.utf8" "ar_AE" "ar_AE.iso88596" "ar_AE.utf8" "ar_BH" "ar_BH.iso88596" "ar_BH.utf8" "ar_DZ" "ar_DZ.iso88596" "ar_DZ.utf8" "ar_EG" "ar_EG.iso88596" "ar_EG.utf8" "ar_IN" "ar_IN.utf8" "ar_IQ" }
3. CLAUSE: lib.locale.num.put

1. INFO (S1) (3 lines):
2. TEXT: exercising std::num_put<char>
3. CLAUSE: lib.locale.num.put

Application finished

_______________________________________________________________________________

ID	Short	Severity	Coun	Context[B	Descript	1st Access[	2nd
	Description	Name	t	est]	ion	Best]	Access[Best]

_______________________________________________________________________________

Comments

[Martin Sebor]

Specifying the -o option to the Thread Checker made the tool produce the attached file after a successful run of the test.

[Martin Sebor]

Added Thread Safety to the list of components affected by this issue.

This most liklely affects all previous versions as far back as 3.0, and possibly prior. Scheduled for 4.2.

[Farid Zaripov]

I can't to reproduce this bug on MSCV 8.0 and gcc 4.2.0/Linux. Perhaps the bug accidentally fixed while fixing the another mt bugs (i.e. numpunct<>).

[Travis Vitek]

Appears to have been fixed by recent multithreading fixes.

[Martin Sebor]

Is it possible that the problem was not in num_put but rather in numpunct or in the facet or locale cache?

[Martin Sebor]

Actually, I looked through our nightly test results and the test still fails on a number of platforms, including these two:
http://people.apache.org/~sebor/stdcxx/results/hpux-11.11-pa-acc-3.73-12D-580086-log.gz.txt
http://people.apache.org/~sebor/stdcxx/results/linux_redhat_el-4.4-em64t-icc-10.0-12s-580086-log.gz.txt

The test also fails with SIGHUP (due to a timeout) on some platforms, e.g.,
http://people.apache.org/~sebor/stdcxx/results/hpux-11.11-pa-acc-3.73-12d-580086-log.gz.txt

[Farid Zaripov]

The r580086 doesn't includes the latest mt fixes.