Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Control: fixed -1 2.4.0+dfsg-6 Control: tags -1 pending Thanks for the advice. I've included both Stevens pthread patch, and an adaptation of my thr_self patch incorporating the feedback. Updated packaging is available in the Debian GIS git. Building an updated package is complicated by the fact that osgearth fails to build with pbuilder at the moment due to the libav9 transition [1] which caused libopenscenegraph99 to become uninstallable (#722674, #720816). http://release.debian.org/transitions/html/libav9.html Kind Regards, Bas -- GnuPG: 0xE88D4AF1 (new) / 0x77A975AD (old) -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Petr Salinger:
>> Is there any lwpid_t which isn't "long" and defined by kernel headers?
>
> See upstream sources:
>
> http://svnweb.freebsd.org/base/head/sys/sys/_types.h?revision=255219&view=markup
>
>
> typedef __int32_t __lwpid_t; /* Thread ID (a.k.a. LWP) */
>
> And compare with
> http://svnweb.freebsd.org/base/head/sys/kern/syscalls.master?revision=255708&view=markup
>
>
> 432 AUE_NULL STD { int thr_self(long *id); }
Oh, I see. It seems I missunderstood what thr_self() does.
Thanks for clarifying!
--
Robert Millan
--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Is there any lwpid_t which isn't "long" and defined by kernel headers?
See upstream sources:
http://svnweb.freebsd.org/base/head/sys/sys/_types.h?revision=255219&view=markup
typedef __int32_t __lwpid_t; /* Thread ID (a.k.a. LWP) */
And compare with
http://svnweb.freebsd.org/base/head/sys/kern/syscalls.master?revision=255708&view=markup
432 AUE_NULL STD { int thr_self(long *id); }
Petr
--
To UNSUBSCRIBE, email to [email protected]
with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Petr Salinger: >> I think you can avoid this by using the primitive: >> >> lwpid_t tid; >> syscall (SYS_thr_self, &tid); > > There is a mess in kernel interfaces, > the right one is > > long tid; > syscall (SYS_thr_self, &tid); Is there any lwpid_t which isn't "long" and defined by kernel headers? -- Robert Millan -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
I think you can avoid this by using the primitive: lwpid_t tid; syscall (SYS_thr_self, &tid); There is a mess in kernel interfaces, the right one is long tid; syscall (SYS_thr_self, &tid); But it holds only for current pthread implementation, it can be changed anytime. Petr -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Steven Chamberlain: > A workaround may be to use pthread_self(), except that the exported > getCurrentThreadId function has to return 'unsigned int'. On > kfreebsd-amd64 a 64-bit pointer to a pthread_t is not absolutely > guaranteed to be unique if truncated to 32 bits, I think you can avoid this by using the primitive: lwpid_t tid; syscall (SYS_thr_self, &tid); -- Robert Millan -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
> osgearth also knows how to use Windows and Mac OS X--specific syscalls > to get a thread ID. This is not portable and there is nothing directly > equivalent on FreeBSD, see: > http://lists.freebsd.org/pipermail/freebsd-hackers/2010-June/031992.html The FreeBSD equivalent is: int pthread_getthreadid_np(void); -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Hi Steven,
On 10/04/2013 11:33 PM, Steven Chamberlain wrote:
> Hi,
>
> osgearth fails to build on kfreebsd-* because it tries to directly use a
> Linux-specific syscall:
>
> https://buildd.debian.org/status/package.php?p=osgearth&suite=sid
>> cd /«BUILDDIR»/osgearth-2.4.0+dfsg/build/src/osgEarth && /usr/bin/c++
>> -DOSGEARTH_LIBRARY -DQT_CORE_LIB -DQT_GUI_LIB -DQT_NO_DEBUG -DTIXML_USE_STL
>> -DosgEarth_EXPORTS -g -O2 -fstack-protector --param=ssp-buffer-size=4
>> -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -O3 -DNDEBUG -fPIC
>> -isystem /usr/include/qt4 -isystem /usr/include/qt4/QtGui -isystem
>> /usr/include/qt4/QtCore -I/«BUILDDIR»/osgearth-2.4.0+dfsg/src
>> -I/usr/include/gdal -I/usr/include/curl-o
>> CMakeFiles/osgEarth.dir/ThreadingUtils.cpp.o -c
>> /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp
>> /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp: In function
>> 'unsigned int osgEarth::Threading::getCurrentThreadId()':
>> /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp:42:30:
>> error: 'SYS_gettid' was not declared in this scope
>>return (unsigned)::syscall(SYS_gettid);
>> ^
>
> osgearth also knows how to use Windows and Mac OS X--specific syscalls
> to get a thread ID. This is not portable and there is nothing directly
> equivalent on FreeBSD, see:
> http://lists.freebsd.org/pipermail/freebsd-hackers/2010-June/031992.html
>
> A workaround may be to use pthread_self(), except that the exported
> getCurrentThreadId function has to return 'unsigned int'. On
> kfreebsd-amd64 a 64-bit pointer to a pthread_t is not absolutely
> guaranteed to be unique if truncated to 32 bits, but it is extremely
> likely, and certainly better than nothing...
>
> Please refer to attached patch which fixes package build at least. Thanks!
Thanks for the patch.
Another option I've looked at is to use the undocumented thr_self(2)
syscall, this was recommended over pthread_self() on freebsd-hackers@:
http://lists.freebsd.org/pipermail/freebsd-hackers/2010-June/031991.html
The attached patch is what I was working on, but it breaks the build
even worse than without it.
I'll have another go at making my patch work, otherwise I'll apply your
patch to at least get the package to build. I had changed the package to
linux-any since upstream doesn't support FreeBSD, but that's not really
a fix for the problem.
Kind Regards,
Bas
--
GnuPG: 0xE88D4AF1 (new) / 0x77A975AD (old)
Index: osgearth/src/osgEarth/ThreadingUtils.cpp
===
--- osgearth.orig/src/osgEarth/ThreadingUtils.cpp 2013-10-05 00:21:18.0 +0200
+++ osgearth/src/osgEarth/ThreadingUtils.cpp 2013-10-05 01:02:46.0 +0200
@@ -23,6 +23,9 @@
#else
# include
# include
+# if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
+# include
+# endif
#endif
using namespace osgEarth::Threading;
@@ -39,6 +42,12 @@
#elif __APPLE__
return ::syscall(SYS_thread_selfid);
#else
- return (unsigned)::syscall(SYS_gettid);
+ if defined(__FreeBSD__) || defined(__FreeBSD_kernel__)
+long lwpid;
+thr_self( &lwpid );
+return (unsigned) lwpid;
+ else
+return (unsigned)::syscall(SYS_gettid);
+ endif
#endif
}
Index: osgearth/CMakeModules/FindFREEBSD_THR.cmake
===
--- /dev/null 1970-01-01 00:00:00.0 +
+++ osgearth/CMakeModules/FindFREEBSD_THR.cmake 2013-10-05 00:21:18.0 +0200
@@ -0,0 +1,14 @@
+execute_process(COMMAND uname -r OUTPUT_VARIABLE FREEBSD_KERNEL_RELEASE OUTPUT_STRIP_TRAILING_WHITESPACE)
+
+set(FREEBSD_KERNEL_HEADERS_DIR "/usr/src/kfreebsd-headers-${FREEBSD_KERNEL_RELEASE}/")
+
+FIND_PATH(FREEBSD_THR_INCLUDE_DIR
+ NAMES "sys/thr.h"
+ PATH_SUFFIXES sys
+ PATHS ${FREEBSD_KERNEL_HEADERS_DIR}
+)
+
+set(FREEBSD_THR_INCLUDE_DIRS ${FREEBSD_THR_INCLUDE_DIR})
+include(FindPackageHandleStandardArgs)
+find_package_handle_standard_args(FREEBSD_THR DEFAULT_MSG FREEBSD_THR_INCLUDE_DIR)
+mark_as_advanced(FREEBSD_THR_INCLUDE_DIR)
Index: osgearth/CMakeLists.txt
===
--- osgearth.orig/CMakeLists.txt 2013-10-05 00:21:18.0 +0200
+++ osgearth/CMakeLists.txt 2013-10-05 00:22:50.0 +0200
@@ -196,6 +196,12 @@
#FIND_PACKAGE(GDAL)
+FIND_PACKAGE(FREEBSD_THR)
+
+if(NOT FREEBSD_THR_FOUND)
+message(FATAL_ERROR "FreeBSD kernel header sys/thr.h could not be found and is a mandatory dependency")
+endif(NOT FREEBSD_THR_FOUND)
+
# Create bin and lib directories if required
Index: osgearth/src/osgEarth/CMakeLists.txt
===
--- osgearth.orig/src/osgEarth/CMakeLists.txt 2013-10-05 00:21:18.0 +0200
+++ osgearth/src/osgEarth/CMakeLists.txt 2013-10-
Bug#725383: FTBFS on kfreebsd-*: SYS_gettid is Linux-specific
Source: osgearth Version: 2.4.0+dfsg-5 Tags: patch User: [email protected] Usertags: kfreebsd X-Debbugs-Cc: [email protected] Severity: serious Justification: fails to build from source (but built successfully in the past) Hi, osgearth fails to build on kfreebsd-* because it tries to directly use a Linux-specific syscall: https://buildd.debian.org/status/package.php?p=osgearth&suite=sid > cd /«BUILDDIR»/osgearth-2.4.0+dfsg/build/src/osgEarth && /usr/bin/c++ > -DOSGEARTH_LIBRARY -DQT_CORE_LIB -DQT_GUI_LIB -DQT_NO_DEBUG -DTIXML_USE_STL > -DosgEarth_EXPORTS -g -O2 -fstack-protector --param=ssp-buffer-size=4 > -Wformat -Werror=format-security -D_FORTIFY_SOURCE=2 -O3 -DNDEBUG -fPIC > -isystem /usr/include/qt4 -isystem /usr/include/qt4/QtGui -isystem > /usr/include/qt4/QtCore -I/«BUILDDIR»/osgearth-2.4.0+dfsg/src > -I/usr/include/gdal -I/usr/include/curl-o > CMakeFiles/osgEarth.dir/ThreadingUtils.cpp.o -c > /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp > /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp: In function > 'unsigned int osgEarth::Threading::getCurrentThreadId()': > /«BUILDDIR»/osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp:42:30: error: > 'SYS_gettid' was not declared in this scope >return (unsigned)::syscall(SYS_gettid); > ^ osgearth also knows how to use Windows and Mac OS X--specific syscalls to get a thread ID. This is not portable and there is nothing directly equivalent on FreeBSD, see: http://lists.freebsd.org/pipermail/freebsd-hackers/2010-June/031992.html A workaround may be to use pthread_self(), except that the exported getCurrentThreadId function has to return 'unsigned int'. On kfreebsd-amd64 a 64-bit pointer to a pthread_t is not absolutely guaranteed to be unique if truncated to 32 bits, but it is extremely likely, and certainly better than nothing... Please refer to attached patch which fixes package build at least. Thanks! -- System Information: Debian Release: jessie/sid APT prefers unstable APT policy: (500, 'unstable'), (1, 'experimental') Architecture: kfreebsd-amd64 (x86_64) Kernel: kFreeBSD 9.0-2-amd64-xenhvm Locale: LANG=en_GB.UTF-8, LC_CTYPE=en_GB.UTF-8 (charmap=UTF-8) Shell: /bin/sh linked to /bin/dash Index: osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp === --- osgearth-2.4.0+dfsg.orig/src/osgEarth/ThreadingUtils.cpp 2013-04-11 19:07:39.0 +0100 +++ osgearth-2.4.0+dfsg/src/osgEarth/ThreadingUtils.cpp 2013-10-04 22:04:38.898332000 +0100 @@ -20,9 +20,11 @@ #ifdef _WIN32 extern "C" unsigned long __stdcall GetCurrentThreadId(); -#else +#elif defined(__APPLE__) || defined(__LINUX__) # include # include +#else +# include #endif using namespace osgEarth::Threading; @@ -38,7 +40,10 @@ return (unsigned)::GetCurrentThreadId(); #elif __APPLE__ return ::syscall(SYS_thread_selfid); -#else +#elif __LINUX__ return (unsigned)::syscall(SYS_gettid); +#else + /* :XXX: this truncates to 32 bits, but better than nothing */ + return (unsigned)pthread_self(); #endif }
