Discussion:
[Xcb] [PATCH pthread-stubs 0/4] Rework pthread-stubs design/implementation
Emil Velikov
2016-11-03 20:24:55 UTC
Permalink
Hi all,

A recent change (commit fa6db2f9c01 - list.m4: Add some new pthread symbols)
lead to an interesting bugreport [1] which makes me think that the current
pthread-stubs design isn't as robust (broken?) as one would expect.

In brief: if there is a pthread-stubs library, one can end up using a
mismatched API - if one uses dlopen with a binary which pulls a full pthreads
implementation. That is because the weak symbols, as provided by the
pthread-stubs DSO, get overridden.

This has gone unnoticed because a) on most platforms pthread-stubs never
creates a DSO and in the rare case where it does b) it's very uncommon to
dlopen a binary which pulls a non-lightweight pthreads symbols, from a
pthreads-stubs (linked) one.


The only solutions that I can think of are:
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
2) "kill off" pthread-stubs and let everyone
- handle the "do I need to link against pthreads or not" in their configure.ac
- link against the complete/full pthread API

Imho 2) is rather selfish/overzealous so I've went ahead and tackled 1).

Note: 1/4 a standalone fix which can be applied independently of how the
direction/discussions reached.

The remaining patches can be squashed or kept separate, based on how things go.

Regards,
Emil

Cc: Rob Clark <***@freedesktop.org>
Cc: Ben Widawsky <***@bwidawsk.net>
Cc: Uli Schlachter <***@znc.in>
Cc: Josh Triplett <***@freedesktop.org>
Cc: Jamey Sharp <***@minilop.net>

[1] https://bugs.freedesktop.org/show_bug.cgi?id=98048

Emil Velikov (4):
autogen.sh: quote path variables
Revert "list.m4: Add some new pthread symbols"
Rework the pthread-stub design
configure: drop various AC_macros

Makefile.am | 10 --
README | 33 +++-
autogen.sh | 6 +-
configure.ac | 69 ++++----
list.m4 | 3 -
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
8 files changed, 548 insertions(+), 152 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4
--
2.10.0
Emil Velikov
2016-11-03 20:24:58 UTC
Permalink
From: Emil Velikov <***@collabora.com>

The current design handles the most common use-cases, although it causes
breakage on others (when a pthreads liked library is dlopened). Refer to
the README for further details.

The new design, makes pthread-stubs a "meta" package which _never_
provides a library but a simple .pc file.

pthread-stubs checks if the run-time (libc or otherwise) expose
lightweight pthread symbols to link against and defaults to the full
blown pthread.

This way projects can reuse the Cflags/Libs without having to know the
details. Alternatively they can directly link against the pthreads
implementation, although that might bring unwarranted overhead.

Signed-off-by: Emil Velikov <***@collabora.com>
---
Makefile.am | 10 --
README | 33 +++-
configure.ac | 64 +++----
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
6 files changed, 545 insertions(+), 141 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4

diff --git a/Makefile.am b/Makefile.am
index 2329572..93123a6 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -1,12 +1,2 @@
pkgconfigdir = $(libdir)/pkgconfig
pkgconfig_DATA = pthread-stubs.pc
-
-if BUILD_LIB
-stubs.c: list.m4 stubs.c.m4
- m4 -P -I $(srcdir) stubs.c.m4 > $@
-
-lib_LTLIBRARIES = libpthread-stubs.la
-nodist_libpthread_stubs_la_SOURCES = stubs.c
-CLEANFILES = stubs.c
-endif
-EXTRA_libpthread_stubs_la_SOURCES = stubs.c.m4
diff --git a/README b/README
index 4e7e85b..44f39ee 100644
--- a/README
+++ b/README
@@ -1,9 +1,24 @@
-This library provides weak aliases for pthread functions not provided in libc
-or otherwise available by default. Libraries like libxcb rely on pthread
-stubs to use pthreads optionally, becoming thread-safe when linked to
-libpthread, while avoiding any performance hit when running single-threaded.
-libpthread-stubs supports this behavior even on platforms which do not supply
-all the necessary pthread stubs. On platforms which already supply all the
-necessary pthread stubs, this package ships only the pkg-config file
-pthread-stubs.pc, to allow libraries to unconditionally express a dependency
-on pthread-stubs and still obtain correct behavior.
+Originally this library used to provide weak aliases for pthread functions not
+provided in libc or otherwise available by default.
+
+This was found to be a fundamental issue with the design, as the following
+scenario came to light:
+
+Program/library links against pthread-stubs and none of its direct dependencies
+link against pthreads. Thus the weak pthread symbols are used. At a later stage
+a module is dlopened. A one which (directly or via some of its dependencies)
+pulls a full-fledged pthreads implementation. This leads to the weak symbols
+being overridden thus any consecutive calls to the pthreads API [by the
+pthread-stubs user] will be mismatched leading to all sorts of issues.
+
+In order to prevent that one has to traverse up the dependency chain (from the
+pthread-stubs user) and establish whether any of the dependencies will dlopen
+a module which pulls the complete pthread library.
+
+As this is not deemed practical and the alternative solution implemented here
+is as follows:
+
+The pthread-stubs package only ever provided a .pc file, which provides a set
+of Cflags and Libs.
+ - In the case of pthreads capable libc those are empty
+ - Alternatively they expand to the platform specific PTHREAD_{CFLAGS,LIBS}
diff --git a/configure.ac b/configure.ac
index ef300ad..aeacc77 100644
--- a/configure.ac
+++ b/configure.ac
@@ -2,6 +2,7 @@ AC_INIT([libpthread-stubs],
0.3,
[***@lists.freedesktop.org])
AC_CONFIG_SRCDIR([pthread-stubs.pc.in])
+AC_CONFIG_MACRO_DIR([m4])
AM_INIT_AUTOMAKE([foreign dist-bzip2])

AC_CONFIG_HEADERS([config.h])
@@ -9,39 +10,42 @@ AC_CONFIG_HEADERS([config.h])
AC_PROG_LIBTOOL
AC_PROG_CC

-
-dnl Detection code for compilers supporting the __attribute__((weak, alias))
-dnl feature. Original code present in unieject's repository
-dnl Diego Pettenò <***@gentoo.org>
-ac_save_CFLAGS="$CFLAGS"
-CFLAGS="$CFLAGS -Werror"
-AC_CACHE_CHECK([if compiler supports __attribute__((weak, alias))],
- [cc_cv_attribute_alias],
- [AC_COMPILE_IFELSE([
- void other_function(void *foo) { }
- void some_function(void *foo) __attribute__((weak, alias("other_function")));
- ],
- [cc_cv_attribute_alias=yes],
- [cc_cv_attribute_alias=no])
- ])
-CFLAGS="$ac_save_CFLAGS"
-if test "x$cc_cv_attribute_alias" = "xyes"; then
- AC_DEFINE([SUPPORT_ATTRIBUTE_ALIAS], 1, [Define this if the compiler supports the alias attribute])
+dnl Check if the following functions have stubs.
+funclist="\
+pthread_self \
+pthread_mutex_init \
+pthread_mutex_destroy \
+pthread_mutex_lock \
+pthread_mutex_unlock \
+pthread_cond_init \
+pthread_cond_destroy \
+pthread_condattr_init \
+pthread_condattr_destroy \
+pthread_cond_wait \
+pthread_cond_timedwait \
+pthread_cond_signal \
+pthread_cond_broadcast \
+pthread_equal \
+pthread_exit"
+
+AC_CHECK_FUNCS($funclist, [HAVE_STUBS=yes])
+
+
+if test "x$HAVE_STUBS" = xyes; then
+ PKG_CONFIG_CFLAGS=
+ PKG_CONFIG_LIBS=
+else
+ AX_PTHREAD
+ if test "x$ax_pthread_ok" = xno; then
+ AC_MSG_ERROR([Building pthread-stubs on this platform requires pthreads])
+ fi
+
+ PKG_CONFIG_CFLAGS=$PTHREAD_CFLAGS
+ PKG_CONFIG_LIBS=$PTHREAD_LIBS
fi

-dnl Grab the list of functions to test for from our template file.
-m4_define([funclist], [])
-m4_define([alias], [m4_if([$1],[na],[], [m4_append([funclist], [$3 ])])])
-m4_divert_push([KILL])
-m4_include([list.m4])
-m4_divert_pop([KILL])
-
-dnl Check which functions we need to provide aliases for.
-PKG_CONFIG_LIBS=
-AC_CHECK_FUNCS(funclist, [], [PKG_CONFIG_LIBS='-L${libdir} -lpthread-stubs'])
-
+AC_SUBST([PKG_CONFIG_CFLAGS])
AC_SUBST([PKG_CONFIG_LIBS])
-AM_CONDITIONAL(BUILD_LIB, test "x$PKG_CONFIG_LIBS" != x)

AC_CONFIG_FILES([Makefile pthread-stubs.pc])
AC_OUTPUT
diff --git a/m4/ax_pthread.m4 b/m4/ax_pthread.m4
new file mode 100644
index 0000000..4c4051e
--- /dev/null
+++ b/m4/ax_pthread.m4
@@ -0,0 +1,485 @@
+# ===========================================================================
+# http://www.gnu.org/software/autoconf-archive/ax_pthread.html
+# ===========================================================================
+#
+# SYNOPSIS
+#
+# AX_PTHREAD([ACTION-IF-FOUND[, ACTION-IF-NOT-FOUND]])
+#
+# DESCRIPTION
+#
+# This macro figures out how to build C programs using POSIX threads. It
+# sets the PTHREAD_LIBS output variable to the threads library and linker
+# flags, and the PTHREAD_CFLAGS output variable to any special C compiler
+# flags that are needed. (The user can also force certain compiler
+# flags/libs to be tested by setting these environment variables.)
+#
+# Also sets PTHREAD_CC to any special C compiler that is needed for
+# multi-threaded programs (defaults to the value of CC otherwise). (This
+# is necessary on AIX to use the special cc_r compiler alias.)
+#
+# NOTE: You are assumed to not only compile your program with these flags,
+# but also to link with them as well. For example, you might link with
+# $PTHREAD_CC $CFLAGS $PTHREAD_CFLAGS $LDFLAGS ... $PTHREAD_LIBS $LIBS
+#
+# If you are only building threaded programs, you may wish to use these
+# variables in your default LIBS, CFLAGS, and CC:
+#
+# LIBS="$PTHREAD_LIBS $LIBS"
+# CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+# CC="$PTHREAD_CC"
+#
+# In addition, if the PTHREAD_CREATE_JOINABLE thread-attribute constant
+# has a nonstandard name, this macro defines PTHREAD_CREATE_JOINABLE to
+# that name (e.g. PTHREAD_CREATE_UNDETACHED on AIX).
+#
+# Also HAVE_PTHREAD_PRIO_INHERIT is defined if pthread is found and the
+# PTHREAD_PRIO_INHERIT symbol is defined when compiling with
+# PTHREAD_CFLAGS.
+#
+# ACTION-IF-FOUND is a list of shell commands to run if a threads library
+# is found, and ACTION-IF-NOT-FOUND is a list of commands to run it if it
+# is not found. If ACTION-IF-FOUND is not specified, the default action
+# will define HAVE_PTHREAD.
+#
+# Please let the authors know if this macro fails on any platform, or if
+# you have any other suggestions or comments. This macro was based on work
+# by SGJ on autoconf scripts for FFTW (http://www.fftw.org/) (with help
+# from M. Frigo), as well as ac_pthread and hb_pthread macros posted by
+# Alejandro Forero Cuervo to the autoconf macro repository. We are also
+# grateful for the helpful feedback of numerous users.
+#
+# Updated for Autoconf 2.68 by Daniel Richard G.
+#
+# LICENSE
+#
+# Copyright (c) 2008 Steven G. Johnson <***@alum.mit.edu>
+# Copyright (c) 2011 Daniel Richard G. <***@iSKUNK.ORG>
+#
+# This program is free software: you can redistribute it and/or modify it
+# under the terms of the GNU General Public License as published by the
+# Free Software Foundation, either version 3 of the License, or (at your
+# option) any later version.
+#
+# This program is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
+# Public License for more details.
+#
+# You should have received a copy of the GNU General Public License along
+# with this program. If not, see <http://www.gnu.org/licenses/>.
+#
+# As a special exception, the respective Autoconf Macro's copyright owner
+# gives unlimited permission to copy, distribute and modify the configure
+# scripts that are the output of Autoconf when processing the Macro. You
+# need not follow the terms of the GNU General Public License when using
+# or distributing such scripts, even though portions of the text of the
+# Macro appear in them. The GNU General Public License (GPL) does govern
+# all other use of the material that constitutes the Autoconf Macro.
+#
+# This special exception to the GPL applies to versions of the Autoconf
+# Macro released by the Autoconf Archive. When you make and distribute a
+# modified version of the Autoconf Macro, you may extend this special
+# exception to the GPL to apply to your modified version as well.
+
+#serial 23
+
+AU_ALIAS([ACX_PTHREAD], [AX_PTHREAD])
+AC_DEFUN([AX_PTHREAD], [
+AC_REQUIRE([AC_CANONICAL_HOST])
+AC_REQUIRE([AC_PROG_CC])
+AC_REQUIRE([AC_PROG_SED])
+AC_LANG_PUSH([C])
+ax_pthread_ok=no
+
+# We used to check for pthread.h first, but this fails if pthread.h
+# requires special compiler flags (e.g. on Tru64 or Sequent).
+# It gets checked for in the link test anyway.
+
+# First of all, check if the user has set any of the PTHREAD_LIBS,
+# etcetera environment variables, and if threads linking works using
+# them:
+if test "x$PTHREAD_CFLAGS$PTHREAD_LIBS" != "x"; then
+ ax_pthread_save_CC="$CC"
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ AS_IF([test "x$PTHREAD_CC" != "x"], [CC="$PTHREAD_CC"])
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+ AC_MSG_CHECKING([for pthread_join using $CC $PTHREAD_CFLAGS $PTHREAD_LIBS])
+ AC_LINK_IFELSE([AC_LANG_CALL([], [pthread_join])], [ax_pthread_ok=yes])
+ AC_MSG_RESULT([$ax_pthread_ok])
+ if test "x$ax_pthread_ok" = "xno"; then
+ PTHREAD_LIBS=""
+ PTHREAD_CFLAGS=""
+ fi
+ CC="$ax_pthread_save_CC"
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+fi
+
+# We must check for the threads library under a number of different
+# names; the ordering is very important because some systems
+# (e.g. DEC) have both -lpthread and -lpthreads, where one of the
+# libraries is broken (non-POSIX).
+
+# Create a list of thread flags to try. Items starting with a "-" are
+# C compiler flags, and other items are library names, except for "none"
+# which indicates that we try without any flags at all, and "pthread-config"
+# which is a program returning the flags for the Pth emulation library.
+
+ax_pthread_flags="pthreads none -Kthread -pthread -pthreads -mthreads pthread --thread-safe -mt pthread-config"
+
+# The ordering *is* (sometimes) important. Some notes on the
+# individual items follow:
+
+# pthreads: AIX (must check this before -lpthread)
+# none: in case threads are in libc; should be tried before -Kthread and
+# other compiler flags to prevent continual compiler warnings
+# -Kthread: Sequent (threads in libc, but -Kthread needed for pthread.h)
+# -pthread: Linux/gcc (kernel threads), BSD/gcc (userland threads), Tru64
+# (Note: HP C rejects this with "bad form for `-t' option")
+# -pthreads: Solaris/gcc (Note: HP C also rejects)
+# -mt: Sun Workshop C (may only link SunOS threads [-lthread], but it
+# doesn't hurt to check since this sometimes defines pthreads and
+# -D_REENTRANT too), HP C (must be checked before -lpthread, which
+# is present but should not be used directly; and before -mthreads,
+# because the compiler interprets this as "-mt" + "-hreads")
+# -mthreads: Mingw32/gcc, Lynx/gcc
+# pthread: Linux, etcetera
+# --thread-safe: KAI C++
+# pthread-config: use pthread-config program (for GNU Pth library)
+
+case $host_os in
+
+ freebsd*)
+
+ # -kthread: FreeBSD kernel threads (preferred to -pthread since SMP-able)
+ # lthread: LinuxThreads port on FreeBSD (also preferred to -pthread)
+
+ ax_pthread_flags="-kthread lthread $ax_pthread_flags"
+ ;;
+
+ hpux*)
+
+ # From the cc(1) man page: "[-mt] Sets various -D flags to enable
+ # multi-threading and also sets -lpthread."
+
+ ax_pthread_flags="-mt -pthread pthread $ax_pthread_flags"
+ ;;
+
+ openedition*)
+
+ # IBM z/OS requires a feature-test macro to be defined in order to
+ # enable POSIX threads at all, so give the user a hint if this is
+ # not set. (We don't define these ourselves, as they can affect
+ # other portions of the system API in unpredictable ways.)
+
+ AC_EGREP_CPP([AX_PTHREAD_ZOS_MISSING],
+ [
+# if !defined(_OPEN_THREADS) && !defined(_UNIX03_THREADS)
+ AX_PTHREAD_ZOS_MISSING
+# endif
+ ],
+ [AC_MSG_WARN([IBM z/OS requires -D_OPEN_THREADS or -D_UNIX03_THREADS to enable pthreads support.])])
+ ;;
+
+ solaris*)
+
+ # On Solaris (at least, for some versions), libc contains stubbed
+ # (non-functional) versions of the pthreads routines, so link-based
+ # tests will erroneously succeed. (N.B.: The stubs are missing
+ # pthread_cleanup_push, or rather a function called by this macro,
+ # so we could check for that, but who knows whether they'll stub
+ # that too in a future libc.) So we'll check first for the
+ # standard Solaris way of linking pthreads (-mt -lpthread).
+
+ ax_pthread_flags="-mt,pthread pthread $ax_pthread_flags"
+ ;;
+esac
+
+# GCC generally uses -pthread, or -pthreads on some platforms (e.g. SPARC)
+
+AS_IF([test "x$GCC" = "xyes"],
+ [ax_pthread_flags="-pthread -pthreads $ax_pthread_flags"])
+
+# The presence of a feature test macro requesting re-entrant function
+# definitions is, on some systems, a strong hint that pthreads support is
+# correctly enabled
+
+case $host_os in
+ darwin* | hpux* | linux* | osf* | solaris*)
+ ax_pthread_check_macro="_REENTRANT"
+ ;;
+
+ aix*)
+ ax_pthread_check_macro="_THREAD_SAFE"
+ ;;
+
+ *)
+ ax_pthread_check_macro="--"
+ ;;
+esac
+AS_IF([test "x$ax_pthread_check_macro" = "x--"],
+ [ax_pthread_check_cond=0],
+ [ax_pthread_check_cond="!defined($ax_pthread_check_macro)"])
+
+# Are we compiling with Clang?
+
+AC_CACHE_CHECK([whether $CC is Clang],
+ [ax_cv_PTHREAD_CLANG],
+ [ax_cv_PTHREAD_CLANG=no
+ # Note that Autoconf sets GCC=yes for Clang as well as GCC
+ if test "x$GCC" = "xyes"; then
+ AC_EGREP_CPP([AX_PTHREAD_CC_IS_CLANG],
+ [/* Note: Clang 2.7 lacks __clang_[a-z]+__ */
+# if defined(__clang__) && defined(__llvm__)
+ AX_PTHREAD_CC_IS_CLANG
+# endif
+ ],
+ [ax_cv_PTHREAD_CLANG=yes])
+ fi
+ ])
+ax_pthread_clang="$ax_cv_PTHREAD_CLANG"
+
+ax_pthread_clang_warning=no
+
+# Clang needs special handling, because older versions handle the -pthread
+# option in a rather... idiosyncratic way
+
+if test "x$ax_pthread_clang" = "xyes"; then
+
+ # Clang takes -pthread; it has never supported any other flag
+
+ # (Note 1: This will need to be revisited if a system that Clang
+ # supports has POSIX threads in a separate library. This tends not
+ # to be the way of modern systems, but it's conceivable.)
+
+ # (Note 2: On some systems, notably Darwin, -pthread is not needed
+ # to get POSIX threads support; the API is always present and
+ # active. We could reasonably leave PTHREAD_CFLAGS empty. But
+ # -pthread does define _REENTRANT, and while the Darwin headers
+ # ignore this macro, third-party headers might not.)
+
+ PTHREAD_CFLAGS="-pthread"
+ PTHREAD_LIBS=
+
+ ax_pthread_ok=yes
+
+ # However, older versions of Clang make a point of warning the user
+ # that, in an invocation where only linking and no compilation is
+ # taking place, the -pthread option has no effect ("argument unused
+ # during compilation"). They expect -pthread to be passed in only
+ # when source code is being compiled.
+ #
+ # Problem is, this is at odds with the way Automake and most other
+ # C build frameworks function, which is that the same flags used in
+ # compilation (CFLAGS) are also used in linking. Many systems
+ # supported by AX_PTHREAD require exactly this for POSIX threads
+ # support, and in fact it is often not straightforward to specify a
+ # flag that is used only in the compilation phase and not in
+ # linking. Such a scenario is extremely rare in practice.
+ #
+ # Even though use of the -pthread flag in linking would only print
+ # a warning, this can be a nuisance for well-run software projects
+ # that build with -Werror. So if the active version of Clang has
+ # this misfeature, we search for an option to squash it.
+
+ AC_CACHE_CHECK([whether Clang needs flag to prevent "argument unused" warning when linking with -pthread],
+ [ax_cv_PTHREAD_CLANG_NO_WARN_FLAG],
+ [ax_cv_PTHREAD_CLANG_NO_WARN_FLAG=unknown
+ # Create an alternate version of $ac_link that compiles and
+ # links in two steps (.c -> .o, .o -> exe) instead of one
+ # (.c -> exe), because the warning occurs only in the second
+ # step
+ ax_pthread_save_ac_link="$ac_link"
+ ax_pthread_sed='s/conftest\.\$ac_ext/conftest.$ac_objext/g'
+ ax_pthread_link_step=`$as_echo "$ac_link" | sed "$ax_pthread_sed"`
+ ax_pthread_2step_ac_link="($ac_compile) && (echo ==== >&5) && ($ax_pthread_link_step)"
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ for ax_pthread_try in '' -Qunused-arguments -Wno-unused-command-line-argument unknown; do
+ AS_IF([test "x$ax_pthread_try" = "xunknown"], [break])
+ CFLAGS="-Werror -Wunknown-warning-option $ax_pthread_try -pthread $ax_pthread_save_CFLAGS"
+ ac_link="$ax_pthread_save_ac_link"
+ AC_LINK_IFELSE([AC_LANG_SOURCE([[int main(void){return 0;}]])],
+ [ac_link="$ax_pthread_2step_ac_link"
+ AC_LINK_IFELSE([AC_LANG_SOURCE([[int main(void){return 0;}]])],
+ [break])
+ ])
+ done
+ ac_link="$ax_pthread_save_ac_link"
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ AS_IF([test "x$ax_pthread_try" = "x"], [ax_pthread_try=no])
+ ax_cv_PTHREAD_CLANG_NO_WARN_FLAG="$ax_pthread_try"
+ ])
+
+ case "$ax_cv_PTHREAD_CLANG_NO_WARN_FLAG" in
+ no | unknown) ;;
+ *) PTHREAD_CFLAGS="$ax_cv_PTHREAD_CLANG_NO_WARN_FLAG $PTHREAD_CFLAGS" ;;
+ esac
+
+fi # $ax_pthread_clang = yes
+
+if test "x$ax_pthread_ok" = "xno"; then
+for ax_pthread_try_flag in $ax_pthread_flags; do
+
+ case $ax_pthread_try_flag in
+ none)
+ AC_MSG_CHECKING([whether pthreads work without any flags])
+ ;;
+
+ -mt,pthread)
+ AC_MSG_CHECKING([whether pthreads work with -mt -lpthread])
+ PTHREAD_CFLAGS="-mt"
+ PTHREAD_LIBS="-lpthread"
+ ;;
+
+ -*)
+ AC_MSG_CHECKING([whether pthreads work with $ax_pthread_try_flag])
+ PTHREAD_CFLAGS="$ax_pthread_try_flag"
+ ;;
+
+ pthread-config)
+ AC_CHECK_PROG([ax_pthread_config], [pthread-config], [yes], [no])
+ AS_IF([test "x$ax_pthread_config" = "xno"], [continue])
+ PTHREAD_CFLAGS="`pthread-config --cflags`"
+ PTHREAD_LIBS="`pthread-config --ldflags` `pthread-config --libs`"
+ ;;
+
+ *)
+ AC_MSG_CHECKING([for the pthreads library -l$ax_pthread_try_flag])
+ PTHREAD_LIBS="-l$ax_pthread_try_flag"
+ ;;
+ esac
+
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+
+ # Check for various functions. We must include pthread.h,
+ # since some functions may be macros. (On the Sequent, we
+ # need a special flag -Kthread to make this header compile.)
+ # We check for pthread_join because it is in -lpthread on IRIX
+ # while pthread_create is in libc. We check for pthread_attr_init
+ # due to DEC craziness with -lpthreads. We check for
+ # pthread_cleanup_push because it is one of the few pthread
+ # functions on Solaris that doesn't have a non-functional libc stub.
+ # We try pthread_create on general principles.
+
+ AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <pthread.h>
+# if $ax_pthread_check_cond
+# error "$ax_pthread_check_macro must be defined"
+# endif
+ static void routine(void *a) { a = 0; }
+ static void *start_routine(void *a) { return a; }],
+ [pthread_t th; pthread_attr_t attr;
+ pthread_create(&th, 0, start_routine, 0);
+ pthread_join(th, 0);
+ pthread_attr_init(&attr);
+ pthread_cleanup_push(routine, 0);
+ pthread_cleanup_pop(0) /* ; */])],
+ [ax_pthread_ok=yes],
+ [])
+
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+
+ AC_MSG_RESULT([$ax_pthread_ok])
+ AS_IF([test "x$ax_pthread_ok" = "xyes"], [break])
+
+ PTHREAD_LIBS=""
+ PTHREAD_CFLAGS=""
+done
+fi
+
+# Various other checks:
+if test "x$ax_pthread_ok" = "xyes"; then
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+
+ # Detect AIX lossage: JOINABLE attribute is called UNDETACHED.
+ AC_CACHE_CHECK([for joinable pthread attribute],
+ [ax_cv_PTHREAD_JOINABLE_ATTR],
+ [ax_cv_PTHREAD_JOINABLE_ATTR=unknown
+ for ax_pthread_attr in PTHREAD_CREATE_JOINABLE PTHREAD_CREATE_UNDETACHED; do
+ AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <pthread.h>],
+ [int attr = $ax_pthread_attr; return attr /* ; */])],
+ [ax_cv_PTHREAD_JOINABLE_ATTR=$ax_pthread_attr; break],
+ [])
+ done
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_JOINABLE_ATTR" != "xunknown" && \
+ test "x$ax_cv_PTHREAD_JOINABLE_ATTR" != "xPTHREAD_CREATE_JOINABLE" && \
+ test "x$ax_pthread_joinable_attr_defined" != "xyes"],
+ [AC_DEFINE_UNQUOTED([PTHREAD_CREATE_JOINABLE],
+ [$ax_cv_PTHREAD_JOINABLE_ATTR],
+ [Define to necessary symbol if this constant
+ uses a non-standard name on your system.])
+ ax_pthread_joinable_attr_defined=yes
+ ])
+
+ AC_CACHE_CHECK([whether more special flags are required for pthreads],
+ [ax_cv_PTHREAD_SPECIAL_FLAGS],
+ [ax_cv_PTHREAD_SPECIAL_FLAGS=no
+ case $host_os in
+ solaris*)
+ ax_cv_PTHREAD_SPECIAL_FLAGS="-D_POSIX_PTHREAD_SEMANTICS"
+ ;;
+ esac
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_SPECIAL_FLAGS" != "xno" && \
+ test "x$ax_pthread_special_flags_added" != "xyes"],
+ [PTHREAD_CFLAGS="$ax_cv_PTHREAD_SPECIAL_FLAGS $PTHREAD_CFLAGS"
+ ax_pthread_special_flags_added=yes])
+
+ AC_CACHE_CHECK([for PTHREAD_PRIO_INHERIT],
+ [ax_cv_PTHREAD_PRIO_INHERIT],
+ [AC_LINK_IFELSE([AC_LANG_PROGRAM([[#include <pthread.h>]],
+ [[int i = PTHREAD_PRIO_INHERIT;]])],
+ [ax_cv_PTHREAD_PRIO_INHERIT=yes],
+ [ax_cv_PTHREAD_PRIO_INHERIT=no])
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_PRIO_INHERIT" = "xyes" && \
+ test "x$ax_pthread_prio_inherit_defined" != "xyes"],
+ [AC_DEFINE([HAVE_PTHREAD_PRIO_INHERIT], [1], [Have PTHREAD_PRIO_INHERIT.])
+ ax_pthread_prio_inherit_defined=yes
+ ])
+
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+
+ # More AIX lossage: compile with *_r variant
+ if test "x$GCC" != "xyes"; then
+ case $host_os in
+ aix*)
+ AS_CASE(["x/$CC"],
+ [x*/c89|x*/c89_128|x*/c99|x*/c99_128|x*/cc|x*/cc128|x*/xlc|x*/xlc_v6|x*/xlc128|x*/xlc128_v6],
+ [#handle absolute path differently from PATH based program lookup
+ AS_CASE(["x$CC"],
+ [x/*],
+ [AS_IF([AS_EXECUTABLE_P([${CC}_r])],[PTHREAD_CC="${CC}_r"])],
+ [AC_CHECK_PROGS([PTHREAD_CC],[${CC}_r],[$CC])])])
+ ;;
+ esac
+ fi
+fi
+
+test -n "$PTHREAD_CC" || PTHREAD_CC="$CC"
+
+AC_SUBST([PTHREAD_LIBS])
+AC_SUBST([PTHREAD_CFLAGS])
+AC_SUBST([PTHREAD_CC])
+
+# Finally, execute ACTION-IF-FOUND/ACTION-IF-NOT-FOUND:
+if test "x$ax_pthread_ok" = "xyes"; then
+ ifelse([$1],,[AC_DEFINE([HAVE_PTHREAD],[1],[Define if you have POSIX threads libraries and header files.])],[$1])
+ :
+else
+ ax_pthread_ok=no
+ $2
+fi
+AC_LANG_POP
+])dnl AX_PTHREAD
diff --git a/pthread-stubs.pc.in b/pthread-stubs.pc.in
index 6900598..4522c04 100644
--- a/pthread-stubs.pc.in
+++ b/pthread-stubs.pc.in
@@ -3,6 +3,7 @@ exec_prefix=@exec_prefix@
libdir=@libdir@

Name: pthread stubs
-Description: Stubs missing from libc for standard pthread functions
+Description: Meta package for pthread symbols - defaults to pthread if libc does not provide lightweight stubs.
Version: @PACKAGE_VERSION@
+Cflags: @PKG_CONFIG_CFLAGS@
Libs: @PKG_CONFIG_LIBS@
diff --git a/stubs.c.m4 b/stubs.c.m4
deleted file mode 100644
index e069b9e..0000000
--- a/stubs.c.m4
+++ /dev/null
@@ -1,91 +0,0 @@
-m4_divert(-1)m4_dnl -*- c -*-
-m4_changequote(`[',`]')
-m4_changecom()
-
-m4_define([upcase], [m4_translit([$*], [a-z], [A-Z])])
-
-m4_define([alias], [m4_dnl
-#ifndef upcase(HAVE_[$3])
-#define upcase(NEED_[$1]_STUB)
-# ifdef SUPPORT_ATTRIBUTE_ALIAS
-[$2] [$3]() __attribute__ ((weak, [alias] ("__pthread_[$1]_stub")));
-# else
-# pragma weak [$3] = __pthread_[$1]_stub
-# endif
-#endif
-])
-
-m4_divert(0)m4_dnl
-/* Copyright (C) 2006 Diego Pettenò
- * Copyright (C) 2010 M Joonas Pihlaja
- * Inspired by libX11 code copyright (c) 1995 David E. Wexelblat.
- *
- * This is an automatically created file from stubs.c.m4. Do not edit.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * Except as contained in this notice, the names of the authors or their
- * institutions shall not be used in advertising or otherwise to promote the
- * sale, use or other dealings in this Software without prior written
- * authorization from the authors.
- */
-
-#include <pthread.h>
-#include <stdlib.h>
-#include <string.h>
-#include "config.h"
-
-m4_include([list.m4])
-
-#ifdef NEED_ZERO_STUB
-static int __pthread_zero_stub()
-{
- return 0;
-}
-#endif
-
-#ifdef NEED_ABORT_STUB
-static int __pthread_abort_stub()
-{
- abort();
-}
-#endif
-
-#ifdef NEED_SELF_STUB
-static pthread_t __pthread_self_stub(void)
-{
- pthread_t th;
-
- memset(&th, 0, sizeof (th));
- return th;
-}
-#endif
-
-#ifdef NEED_EQUAL_STUB
-static int __pthread_equal_stub(pthread_t t1, pthread_t t2)
-{
- return 1;
-}
-#endif
-
-#ifdef NEED_EXIT_STUB
-static void __pthread_exit_stub(void *ret)
-{
- exit(EXIT_SUCCESS);
-}
-#endif
--
2.10.0
Emil Velikov
2016-11-03 20:42:06 UTC
Permalink
From: Emil Velikov <***@collabora.com>

The current design handles the most common use-cases, although it causes
breakage on others (when a pthreads liked library is dlopened). Refer to
the README for further details.

The new design, makes pthread-stubs a "meta" package which _never_
provides a library but a simple .pc file.

pthread-stubs checks if the run-time (libc or otherwise) expose
lightweight pthread symbols to link against and defaults to the full
blown pthread.

This way projects can reuse the Cflags/Libs without having to know the
details. Alternatively they can directly link against the pthreads
implementation, although that might bring unwarranted overhead.

v2: Use correct conditional in configure.ac

Signed-off-by: Emil Velikov <***@collabora.com>
---
Makefile.am | 10 --
README | 33 +++-
configure.ac | 64 +++----
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
6 files changed, 545 insertions(+), 141 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4

diff --git a/Makefile.am b/Makefile.am
index 2329572..93123a6 100644
--- a/Makefile.am
+++ b/Makefile.am
@@ -1,12 +1,2 @@
pkgconfigdir = $(libdir)/pkgconfig
pkgconfig_DATA = pthread-stubs.pc
-
-if BUILD_LIB
-stubs.c: list.m4 stubs.c.m4
- m4 -P -I $(srcdir) stubs.c.m4 > $@
-
-lib_LTLIBRARIES = libpthread-stubs.la
-nodist_libpthread_stubs_la_SOURCES = stubs.c
-CLEANFILES = stubs.c
-endif
-EXTRA_libpthread_stubs_la_SOURCES = stubs.c.m4
diff --git a/README b/README
index 4e7e85b..44f39ee 100644
--- a/README
+++ b/README
@@ -1,9 +1,24 @@
-This library provides weak aliases for pthread functions not provided in libc
-or otherwise available by default. Libraries like libxcb rely on pthread
-stubs to use pthreads optionally, becoming thread-safe when linked to
-libpthread, while avoiding any performance hit when running single-threaded.
-libpthread-stubs supports this behavior even on platforms which do not supply
-all the necessary pthread stubs. On platforms which already supply all the
-necessary pthread stubs, this package ships only the pkg-config file
-pthread-stubs.pc, to allow libraries to unconditionally express a dependency
-on pthread-stubs and still obtain correct behavior.
+Originally this library used to provide weak aliases for pthread functions not
+provided in libc or otherwise available by default.
+
+This was found to be a fundamental issue with the design, as the following
+scenario came to light:
+
+Program/library links against pthread-stubs and none of its direct dependencies
+link against pthreads. Thus the weak pthread symbols are used. At a later stage
+a module is dlopened. A one which (directly or via some of its dependencies)
+pulls a full-fledged pthreads implementation. This leads to the weak symbols
+being overridden thus any consecutive calls to the pthreads API [by the
+pthread-stubs user] will be mismatched leading to all sorts of issues.
+
+In order to prevent that one has to traverse up the dependency chain (from the
+pthread-stubs user) and establish whether any of the dependencies will dlopen
+a module which pulls the complete pthread library.
+
+As this is not deemed practical and the alternative solution implemented here
+is as follows:
+
+The pthread-stubs package only ever provided a .pc file, which provides a set
+of Cflags and Libs.
+ - In the case of pthreads capable libc those are empty
+ - Alternatively they expand to the platform specific PTHREAD_{CFLAGS,LIBS}
diff --git a/configure.ac b/configure.ac
index ef300ad..8962156 100644
--- a/configure.ac
+++ b/configure.ac
@@ -2,6 +2,7 @@ AC_INIT([libpthread-stubs],
0.3,
[***@lists.freedesktop.org])
AC_CONFIG_SRCDIR([pthread-stubs.pc.in])
+AC_CONFIG_MACRO_DIR([m4])
AM_INIT_AUTOMAKE([foreign dist-bzip2])

AC_CONFIG_HEADERS([config.h])
@@ -9,39 +10,42 @@ AC_CONFIG_HEADERS([config.h])
AC_PROG_LIBTOOL
AC_PROG_CC

-
-dnl Detection code for compilers supporting the __attribute__((weak, alias))
-dnl feature. Original code present in unieject's repository
-dnl Diego Pettenò <***@gentoo.org>
-ac_save_CFLAGS="$CFLAGS"
-CFLAGS="$CFLAGS -Werror"
-AC_CACHE_CHECK([if compiler supports __attribute__((weak, alias))],
- [cc_cv_attribute_alias],
- [AC_COMPILE_IFELSE([
- void other_function(void *foo) { }
- void some_function(void *foo) __attribute__((weak, alias("other_function")));
- ],
- [cc_cv_attribute_alias=yes],
- [cc_cv_attribute_alias=no])
- ])
-CFLAGS="$ac_save_CFLAGS"
-if test "x$cc_cv_attribute_alias" = "xyes"; then
- AC_DEFINE([SUPPORT_ATTRIBUTE_ALIAS], 1, [Define this if the compiler supports the alias attribute])
+dnl Check if the following functions have stubs.
+funclist="\
+pthread_self \
+pthread_mutex_init \
+pthread_mutex_destroy \
+pthread_mutex_lock \
+pthread_mutex_unlock \
+pthread_cond_init \
+pthread_cond_destroy \
+pthread_condattr_init \
+pthread_condattr_destroy \
+pthread_cond_wait \
+pthread_cond_timedwait \
+pthread_cond_signal \
+pthread_cond_broadcast \
+pthread_equal \
+pthread_exit"
+
+AC_CHECK_FUNCS($funclist, [], [HAVE_STUBS=no])
+
+
+if test "x$HAVE_STUBS" != xno; then
+ PKG_CONFIG_CFLAGS=
+ PKG_CONFIG_LIBS=
+else
+ AX_PTHREAD
+ if test "x$ax_pthread_ok" = xno; then
+ AC_MSG_ERROR([Building pthread-stubs on this platform requires pthreads])
+ fi
+
+ PKG_CONFIG_CFLAGS=$PTHREAD_CFLAGS
+ PKG_CONFIG_LIBS=$PTHREAD_LIBS
fi

-dnl Grab the list of functions to test for from our template file.
-m4_define([funclist], [])
-m4_define([alias], [m4_if([$1],[na],[], [m4_append([funclist], [$3 ])])])
-m4_divert_push([KILL])
-m4_include([list.m4])
-m4_divert_pop([KILL])
-
-dnl Check which functions we need to provide aliases for.
-PKG_CONFIG_LIBS=
-AC_CHECK_FUNCS(funclist, [], [PKG_CONFIG_LIBS='-L${libdir} -lpthread-stubs'])
-
+AC_SUBST([PKG_CONFIG_CFLAGS])
AC_SUBST([PKG_CONFIG_LIBS])
-AM_CONDITIONAL(BUILD_LIB, test "x$PKG_CONFIG_LIBS" != x)

AC_CONFIG_FILES([Makefile pthread-stubs.pc])
AC_OUTPUT
diff --git a/m4/ax_pthread.m4 b/m4/ax_pthread.m4
new file mode 100644
index 0000000..4c4051e
--- /dev/null
+++ b/m4/ax_pthread.m4
@@ -0,0 +1,485 @@
+# ===========================================================================
+# http://www.gnu.org/software/autoconf-archive/ax_pthread.html
+# ===========================================================================
+#
+# SYNOPSIS
+#
+# AX_PTHREAD([ACTION-IF-FOUND[, ACTION-IF-NOT-FOUND]])
+#
+# DESCRIPTION
+#
+# This macro figures out how to build C programs using POSIX threads. It
+# sets the PTHREAD_LIBS output variable to the threads library and linker
+# flags, and the PTHREAD_CFLAGS output variable to any special C compiler
+# flags that are needed. (The user can also force certain compiler
+# flags/libs to be tested by setting these environment variables.)
+#
+# Also sets PTHREAD_CC to any special C compiler that is needed for
+# multi-threaded programs (defaults to the value of CC otherwise). (This
+# is necessary on AIX to use the special cc_r compiler alias.)
+#
+# NOTE: You are assumed to not only compile your program with these flags,
+# but also to link with them as well. For example, you might link with
+# $PTHREAD_CC $CFLAGS $PTHREAD_CFLAGS $LDFLAGS ... $PTHREAD_LIBS $LIBS
+#
+# If you are only building threaded programs, you may wish to use these
+# variables in your default LIBS, CFLAGS, and CC:
+#
+# LIBS="$PTHREAD_LIBS $LIBS"
+# CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+# CC="$PTHREAD_CC"
+#
+# In addition, if the PTHREAD_CREATE_JOINABLE thread-attribute constant
+# has a nonstandard name, this macro defines PTHREAD_CREATE_JOINABLE to
+# that name (e.g. PTHREAD_CREATE_UNDETACHED on AIX).
+#
+# Also HAVE_PTHREAD_PRIO_INHERIT is defined if pthread is found and the
+# PTHREAD_PRIO_INHERIT symbol is defined when compiling with
+# PTHREAD_CFLAGS.
+#
+# ACTION-IF-FOUND is a list of shell commands to run if a threads library
+# is found, and ACTION-IF-NOT-FOUND is a list of commands to run it if it
+# is not found. If ACTION-IF-FOUND is not specified, the default action
+# will define HAVE_PTHREAD.
+#
+# Please let the authors know if this macro fails on any platform, or if
+# you have any other suggestions or comments. This macro was based on work
+# by SGJ on autoconf scripts for FFTW (http://www.fftw.org/) (with help
+# from M. Frigo), as well as ac_pthread and hb_pthread macros posted by
+# Alejandro Forero Cuervo to the autoconf macro repository. We are also
+# grateful for the helpful feedback of numerous users.
+#
+# Updated for Autoconf 2.68 by Daniel Richard G.
+#
+# LICENSE
+#
+# Copyright (c) 2008 Steven G. Johnson <***@alum.mit.edu>
+# Copyright (c) 2011 Daniel Richard G. <***@iSKUNK.ORG>
+#
+# This program is free software: you can redistribute it and/or modify it
+# under the terms of the GNU General Public License as published by the
+# Free Software Foundation, either version 3 of the License, or (at your
+# option) any later version.
+#
+# This program is distributed in the hope that it will be useful, but
+# WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
+# Public License for more details.
+#
+# You should have received a copy of the GNU General Public License along
+# with this program. If not, see <http://www.gnu.org/licenses/>.
+#
+# As a special exception, the respective Autoconf Macro's copyright owner
+# gives unlimited permission to copy, distribute and modify the configure
+# scripts that are the output of Autoconf when processing the Macro. You
+# need not follow the terms of the GNU General Public License when using
+# or distributing such scripts, even though portions of the text of the
+# Macro appear in them. The GNU General Public License (GPL) does govern
+# all other use of the material that constitutes the Autoconf Macro.
+#
+# This special exception to the GPL applies to versions of the Autoconf
+# Macro released by the Autoconf Archive. When you make and distribute a
+# modified version of the Autoconf Macro, you may extend this special
+# exception to the GPL to apply to your modified version as well.
+
+#serial 23
+
+AU_ALIAS([ACX_PTHREAD], [AX_PTHREAD])
+AC_DEFUN([AX_PTHREAD], [
+AC_REQUIRE([AC_CANONICAL_HOST])
+AC_REQUIRE([AC_PROG_CC])
+AC_REQUIRE([AC_PROG_SED])
+AC_LANG_PUSH([C])
+ax_pthread_ok=no
+
+# We used to check for pthread.h first, but this fails if pthread.h
+# requires special compiler flags (e.g. on Tru64 or Sequent).
+# It gets checked for in the link test anyway.
+
+# First of all, check if the user has set any of the PTHREAD_LIBS,
+# etcetera environment variables, and if threads linking works using
+# them:
+if test "x$PTHREAD_CFLAGS$PTHREAD_LIBS" != "x"; then
+ ax_pthread_save_CC="$CC"
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ AS_IF([test "x$PTHREAD_CC" != "x"], [CC="$PTHREAD_CC"])
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+ AC_MSG_CHECKING([for pthread_join using $CC $PTHREAD_CFLAGS $PTHREAD_LIBS])
+ AC_LINK_IFELSE([AC_LANG_CALL([], [pthread_join])], [ax_pthread_ok=yes])
+ AC_MSG_RESULT([$ax_pthread_ok])
+ if test "x$ax_pthread_ok" = "xno"; then
+ PTHREAD_LIBS=""
+ PTHREAD_CFLAGS=""
+ fi
+ CC="$ax_pthread_save_CC"
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+fi
+
+# We must check for the threads library under a number of different
+# names; the ordering is very important because some systems
+# (e.g. DEC) have both -lpthread and -lpthreads, where one of the
+# libraries is broken (non-POSIX).
+
+# Create a list of thread flags to try. Items starting with a "-" are
+# C compiler flags, and other items are library names, except for "none"
+# which indicates that we try without any flags at all, and "pthread-config"
+# which is a program returning the flags for the Pth emulation library.
+
+ax_pthread_flags="pthreads none -Kthread -pthread -pthreads -mthreads pthread --thread-safe -mt pthread-config"
+
+# The ordering *is* (sometimes) important. Some notes on the
+# individual items follow:
+
+# pthreads: AIX (must check this before -lpthread)
+# none: in case threads are in libc; should be tried before -Kthread and
+# other compiler flags to prevent continual compiler warnings
+# -Kthread: Sequent (threads in libc, but -Kthread needed for pthread.h)
+# -pthread: Linux/gcc (kernel threads), BSD/gcc (userland threads), Tru64
+# (Note: HP C rejects this with "bad form for `-t' option")
+# -pthreads: Solaris/gcc (Note: HP C also rejects)
+# -mt: Sun Workshop C (may only link SunOS threads [-lthread], but it
+# doesn't hurt to check since this sometimes defines pthreads and
+# -D_REENTRANT too), HP C (must be checked before -lpthread, which
+# is present but should not be used directly; and before -mthreads,
+# because the compiler interprets this as "-mt" + "-hreads")
+# -mthreads: Mingw32/gcc, Lynx/gcc
+# pthread: Linux, etcetera
+# --thread-safe: KAI C++
+# pthread-config: use pthread-config program (for GNU Pth library)
+
+case $host_os in
+
+ freebsd*)
+
+ # -kthread: FreeBSD kernel threads (preferred to -pthread since SMP-able)
+ # lthread: LinuxThreads port on FreeBSD (also preferred to -pthread)
+
+ ax_pthread_flags="-kthread lthread $ax_pthread_flags"
+ ;;
+
+ hpux*)
+
+ # From the cc(1) man page: "[-mt] Sets various -D flags to enable
+ # multi-threading and also sets -lpthread."
+
+ ax_pthread_flags="-mt -pthread pthread $ax_pthread_flags"
+ ;;
+
+ openedition*)
+
+ # IBM z/OS requires a feature-test macro to be defined in order to
+ # enable POSIX threads at all, so give the user a hint if this is
+ # not set. (We don't define these ourselves, as they can affect
+ # other portions of the system API in unpredictable ways.)
+
+ AC_EGREP_CPP([AX_PTHREAD_ZOS_MISSING],
+ [
+# if !defined(_OPEN_THREADS) && !defined(_UNIX03_THREADS)
+ AX_PTHREAD_ZOS_MISSING
+# endif
+ ],
+ [AC_MSG_WARN([IBM z/OS requires -D_OPEN_THREADS or -D_UNIX03_THREADS to enable pthreads support.])])
+ ;;
+
+ solaris*)
+
+ # On Solaris (at least, for some versions), libc contains stubbed
+ # (non-functional) versions of the pthreads routines, so link-based
+ # tests will erroneously succeed. (N.B.: The stubs are missing
+ # pthread_cleanup_push, or rather a function called by this macro,
+ # so we could check for that, but who knows whether they'll stub
+ # that too in a future libc.) So we'll check first for the
+ # standard Solaris way of linking pthreads (-mt -lpthread).
+
+ ax_pthread_flags="-mt,pthread pthread $ax_pthread_flags"
+ ;;
+esac
+
+# GCC generally uses -pthread, or -pthreads on some platforms (e.g. SPARC)
+
+AS_IF([test "x$GCC" = "xyes"],
+ [ax_pthread_flags="-pthread -pthreads $ax_pthread_flags"])
+
+# The presence of a feature test macro requesting re-entrant function
+# definitions is, on some systems, a strong hint that pthreads support is
+# correctly enabled
+
+case $host_os in
+ darwin* | hpux* | linux* | osf* | solaris*)
+ ax_pthread_check_macro="_REENTRANT"
+ ;;
+
+ aix*)
+ ax_pthread_check_macro="_THREAD_SAFE"
+ ;;
+
+ *)
+ ax_pthread_check_macro="--"
+ ;;
+esac
+AS_IF([test "x$ax_pthread_check_macro" = "x--"],
+ [ax_pthread_check_cond=0],
+ [ax_pthread_check_cond="!defined($ax_pthread_check_macro)"])
+
+# Are we compiling with Clang?
+
+AC_CACHE_CHECK([whether $CC is Clang],
+ [ax_cv_PTHREAD_CLANG],
+ [ax_cv_PTHREAD_CLANG=no
+ # Note that Autoconf sets GCC=yes for Clang as well as GCC
+ if test "x$GCC" = "xyes"; then
+ AC_EGREP_CPP([AX_PTHREAD_CC_IS_CLANG],
+ [/* Note: Clang 2.7 lacks __clang_[a-z]+__ */
+# if defined(__clang__) && defined(__llvm__)
+ AX_PTHREAD_CC_IS_CLANG
+# endif
+ ],
+ [ax_cv_PTHREAD_CLANG=yes])
+ fi
+ ])
+ax_pthread_clang="$ax_cv_PTHREAD_CLANG"
+
+ax_pthread_clang_warning=no
+
+# Clang needs special handling, because older versions handle the -pthread
+# option in a rather... idiosyncratic way
+
+if test "x$ax_pthread_clang" = "xyes"; then
+
+ # Clang takes -pthread; it has never supported any other flag
+
+ # (Note 1: This will need to be revisited if a system that Clang
+ # supports has POSIX threads in a separate library. This tends not
+ # to be the way of modern systems, but it's conceivable.)
+
+ # (Note 2: On some systems, notably Darwin, -pthread is not needed
+ # to get POSIX threads support; the API is always present and
+ # active. We could reasonably leave PTHREAD_CFLAGS empty. But
+ # -pthread does define _REENTRANT, and while the Darwin headers
+ # ignore this macro, third-party headers might not.)
+
+ PTHREAD_CFLAGS="-pthread"
+ PTHREAD_LIBS=
+
+ ax_pthread_ok=yes
+
+ # However, older versions of Clang make a point of warning the user
+ # that, in an invocation where only linking and no compilation is
+ # taking place, the -pthread option has no effect ("argument unused
+ # during compilation"). They expect -pthread to be passed in only
+ # when source code is being compiled.
+ #
+ # Problem is, this is at odds with the way Automake and most other
+ # C build frameworks function, which is that the same flags used in
+ # compilation (CFLAGS) are also used in linking. Many systems
+ # supported by AX_PTHREAD require exactly this for POSIX threads
+ # support, and in fact it is often not straightforward to specify a
+ # flag that is used only in the compilation phase and not in
+ # linking. Such a scenario is extremely rare in practice.
+ #
+ # Even though use of the -pthread flag in linking would only print
+ # a warning, this can be a nuisance for well-run software projects
+ # that build with -Werror. So if the active version of Clang has
+ # this misfeature, we search for an option to squash it.
+
+ AC_CACHE_CHECK([whether Clang needs flag to prevent "argument unused" warning when linking with -pthread],
+ [ax_cv_PTHREAD_CLANG_NO_WARN_FLAG],
+ [ax_cv_PTHREAD_CLANG_NO_WARN_FLAG=unknown
+ # Create an alternate version of $ac_link that compiles and
+ # links in two steps (.c -> .o, .o -> exe) instead of one
+ # (.c -> exe), because the warning occurs only in the second
+ # step
+ ax_pthread_save_ac_link="$ac_link"
+ ax_pthread_sed='s/conftest\.\$ac_ext/conftest.$ac_objext/g'
+ ax_pthread_link_step=`$as_echo "$ac_link" | sed "$ax_pthread_sed"`
+ ax_pthread_2step_ac_link="($ac_compile) && (echo ==== >&5) && ($ax_pthread_link_step)"
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ for ax_pthread_try in '' -Qunused-arguments -Wno-unused-command-line-argument unknown; do
+ AS_IF([test "x$ax_pthread_try" = "xunknown"], [break])
+ CFLAGS="-Werror -Wunknown-warning-option $ax_pthread_try -pthread $ax_pthread_save_CFLAGS"
+ ac_link="$ax_pthread_save_ac_link"
+ AC_LINK_IFELSE([AC_LANG_SOURCE([[int main(void){return 0;}]])],
+ [ac_link="$ax_pthread_2step_ac_link"
+ AC_LINK_IFELSE([AC_LANG_SOURCE([[int main(void){return 0;}]])],
+ [break])
+ ])
+ done
+ ac_link="$ax_pthread_save_ac_link"
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ AS_IF([test "x$ax_pthread_try" = "x"], [ax_pthread_try=no])
+ ax_cv_PTHREAD_CLANG_NO_WARN_FLAG="$ax_pthread_try"
+ ])
+
+ case "$ax_cv_PTHREAD_CLANG_NO_WARN_FLAG" in
+ no | unknown) ;;
+ *) PTHREAD_CFLAGS="$ax_cv_PTHREAD_CLANG_NO_WARN_FLAG $PTHREAD_CFLAGS" ;;
+ esac
+
+fi # $ax_pthread_clang = yes
+
+if test "x$ax_pthread_ok" = "xno"; then
+for ax_pthread_try_flag in $ax_pthread_flags; do
+
+ case $ax_pthread_try_flag in
+ none)
+ AC_MSG_CHECKING([whether pthreads work without any flags])
+ ;;
+
+ -mt,pthread)
+ AC_MSG_CHECKING([whether pthreads work with -mt -lpthread])
+ PTHREAD_CFLAGS="-mt"
+ PTHREAD_LIBS="-lpthread"
+ ;;
+
+ -*)
+ AC_MSG_CHECKING([whether pthreads work with $ax_pthread_try_flag])
+ PTHREAD_CFLAGS="$ax_pthread_try_flag"
+ ;;
+
+ pthread-config)
+ AC_CHECK_PROG([ax_pthread_config], [pthread-config], [yes], [no])
+ AS_IF([test "x$ax_pthread_config" = "xno"], [continue])
+ PTHREAD_CFLAGS="`pthread-config --cflags`"
+ PTHREAD_LIBS="`pthread-config --ldflags` `pthread-config --libs`"
+ ;;
+
+ *)
+ AC_MSG_CHECKING([for the pthreads library -l$ax_pthread_try_flag])
+ PTHREAD_LIBS="-l$ax_pthread_try_flag"
+ ;;
+ esac
+
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+
+ # Check for various functions. We must include pthread.h,
+ # since some functions may be macros. (On the Sequent, we
+ # need a special flag -Kthread to make this header compile.)
+ # We check for pthread_join because it is in -lpthread on IRIX
+ # while pthread_create is in libc. We check for pthread_attr_init
+ # due to DEC craziness with -lpthreads. We check for
+ # pthread_cleanup_push because it is one of the few pthread
+ # functions on Solaris that doesn't have a non-functional libc stub.
+ # We try pthread_create on general principles.
+
+ AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <pthread.h>
+# if $ax_pthread_check_cond
+# error "$ax_pthread_check_macro must be defined"
+# endif
+ static void routine(void *a) { a = 0; }
+ static void *start_routine(void *a) { return a; }],
+ [pthread_t th; pthread_attr_t attr;
+ pthread_create(&th, 0, start_routine, 0);
+ pthread_join(th, 0);
+ pthread_attr_init(&attr);
+ pthread_cleanup_push(routine, 0);
+ pthread_cleanup_pop(0) /* ; */])],
+ [ax_pthread_ok=yes],
+ [])
+
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+
+ AC_MSG_RESULT([$ax_pthread_ok])
+ AS_IF([test "x$ax_pthread_ok" = "xyes"], [break])
+
+ PTHREAD_LIBS=""
+ PTHREAD_CFLAGS=""
+done
+fi
+
+# Various other checks:
+if test "x$ax_pthread_ok" = "xyes"; then
+ ax_pthread_save_CFLAGS="$CFLAGS"
+ ax_pthread_save_LIBS="$LIBS"
+ CFLAGS="$CFLAGS $PTHREAD_CFLAGS"
+ LIBS="$PTHREAD_LIBS $LIBS"
+
+ # Detect AIX lossage: JOINABLE attribute is called UNDETACHED.
+ AC_CACHE_CHECK([for joinable pthread attribute],
+ [ax_cv_PTHREAD_JOINABLE_ATTR],
+ [ax_cv_PTHREAD_JOINABLE_ATTR=unknown
+ for ax_pthread_attr in PTHREAD_CREATE_JOINABLE PTHREAD_CREATE_UNDETACHED; do
+ AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <pthread.h>],
+ [int attr = $ax_pthread_attr; return attr /* ; */])],
+ [ax_cv_PTHREAD_JOINABLE_ATTR=$ax_pthread_attr; break],
+ [])
+ done
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_JOINABLE_ATTR" != "xunknown" && \
+ test "x$ax_cv_PTHREAD_JOINABLE_ATTR" != "xPTHREAD_CREATE_JOINABLE" && \
+ test "x$ax_pthread_joinable_attr_defined" != "xyes"],
+ [AC_DEFINE_UNQUOTED([PTHREAD_CREATE_JOINABLE],
+ [$ax_cv_PTHREAD_JOINABLE_ATTR],
+ [Define to necessary symbol if this constant
+ uses a non-standard name on your system.])
+ ax_pthread_joinable_attr_defined=yes
+ ])
+
+ AC_CACHE_CHECK([whether more special flags are required for pthreads],
+ [ax_cv_PTHREAD_SPECIAL_FLAGS],
+ [ax_cv_PTHREAD_SPECIAL_FLAGS=no
+ case $host_os in
+ solaris*)
+ ax_cv_PTHREAD_SPECIAL_FLAGS="-D_POSIX_PTHREAD_SEMANTICS"
+ ;;
+ esac
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_SPECIAL_FLAGS" != "xno" && \
+ test "x$ax_pthread_special_flags_added" != "xyes"],
+ [PTHREAD_CFLAGS="$ax_cv_PTHREAD_SPECIAL_FLAGS $PTHREAD_CFLAGS"
+ ax_pthread_special_flags_added=yes])
+
+ AC_CACHE_CHECK([for PTHREAD_PRIO_INHERIT],
+ [ax_cv_PTHREAD_PRIO_INHERIT],
+ [AC_LINK_IFELSE([AC_LANG_PROGRAM([[#include <pthread.h>]],
+ [[int i = PTHREAD_PRIO_INHERIT;]])],
+ [ax_cv_PTHREAD_PRIO_INHERIT=yes],
+ [ax_cv_PTHREAD_PRIO_INHERIT=no])
+ ])
+ AS_IF([test "x$ax_cv_PTHREAD_PRIO_INHERIT" = "xyes" && \
+ test "x$ax_pthread_prio_inherit_defined" != "xyes"],
+ [AC_DEFINE([HAVE_PTHREAD_PRIO_INHERIT], [1], [Have PTHREAD_PRIO_INHERIT.])
+ ax_pthread_prio_inherit_defined=yes
+ ])
+
+ CFLAGS="$ax_pthread_save_CFLAGS"
+ LIBS="$ax_pthread_save_LIBS"
+
+ # More AIX lossage: compile with *_r variant
+ if test "x$GCC" != "xyes"; then
+ case $host_os in
+ aix*)
+ AS_CASE(["x/$CC"],
+ [x*/c89|x*/c89_128|x*/c99|x*/c99_128|x*/cc|x*/cc128|x*/xlc|x*/xlc_v6|x*/xlc128|x*/xlc128_v6],
+ [#handle absolute path differently from PATH based program lookup
+ AS_CASE(["x$CC"],
+ [x/*],
+ [AS_IF([AS_EXECUTABLE_P([${CC}_r])],[PTHREAD_CC="${CC}_r"])],
+ [AC_CHECK_PROGS([PTHREAD_CC],[${CC}_r],[$CC])])])
+ ;;
+ esac
+ fi
+fi
+
+test -n "$PTHREAD_CC" || PTHREAD_CC="$CC"
+
+AC_SUBST([PTHREAD_LIBS])
+AC_SUBST([PTHREAD_CFLAGS])
+AC_SUBST([PTHREAD_CC])
+
+# Finally, execute ACTION-IF-FOUND/ACTION-IF-NOT-FOUND:
+if test "x$ax_pthread_ok" = "xyes"; then
+ ifelse([$1],,[AC_DEFINE([HAVE_PTHREAD],[1],[Define if you have POSIX threads libraries and header files.])],[$1])
+ :
+else
+ ax_pthread_ok=no
+ $2
+fi
+AC_LANG_POP
+])dnl AX_PTHREAD
diff --git a/pthread-stubs.pc.in b/pthread-stubs.pc.in
index 6900598..4522c04 100644
--- a/pthread-stubs.pc.in
+++ b/pthread-stubs.pc.in
@@ -3,6 +3,7 @@ exec_prefix=@exec_prefix@
libdir=@libdir@

Name: pthread stubs
-Description: Stubs missing from libc for standard pthread functions
+Description: Meta package for pthread symbols - defaults to pthread if libc does not provide lightweight stubs.
Version: @PACKAGE_VERSION@
+Cflags: @PKG_CONFIG_CFLAGS@
Libs: @PKG_CONFIG_LIBS@
diff --git a/stubs.c.m4 b/stubs.c.m4
deleted file mode 100644
index e069b9e..0000000
--- a/stubs.c.m4
+++ /dev/null
@@ -1,91 +0,0 @@
-m4_divert(-1)m4_dnl -*- c -*-
-m4_changequote(`[',`]')
-m4_changecom()
-
-m4_define([upcase], [m4_translit([$*], [a-z], [A-Z])])
-
-m4_define([alias], [m4_dnl
-#ifndef upcase(HAVE_[$3])
-#define upcase(NEED_[$1]_STUB)
-# ifdef SUPPORT_ATTRIBUTE_ALIAS
-[$2] [$3]() __attribute__ ((weak, [alias] ("__pthread_[$1]_stub")));
-# else
-# pragma weak [$3] = __pthread_[$1]_stub
-# endif
-#endif
-])
-
-m4_divert(0)m4_dnl
-/* Copyright (C) 2006 Diego Pettenò
- * Copyright (C) 2010 M Joonas Pihlaja
- * Inspired by libX11 code copyright (c) 1995 David E. Wexelblat.
- *
- * This is an automatically created file from stubs.c.m4. Do not edit.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included in
- * all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
- * AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * Except as contained in this notice, the names of the authors or their
- * institutions shall not be used in advertising or otherwise to promote the
- * sale, use or other dealings in this Software without prior written
- * authorization from the authors.
- */
-
-#include <pthread.h>
-#include <stdlib.h>
-#include <string.h>
-#include "config.h"
-
-m4_include([list.m4])
-
-#ifdef NEED_ZERO_STUB
-static int __pthread_zero_stub()
-{
- return 0;
-}
-#endif
-
-#ifdef NEED_ABORT_STUB
-static int __pthread_abort_stub()
-{
- abort();
-}
-#endif
-
-#ifdef NEED_SELF_STUB
-static pthread_t __pthread_self_stub(void)
-{
- pthread_t th;
-
- memset(&th, 0, sizeof (th));
- return th;
-}
-#endif
-
-#ifdef NEED_EQUAL_STUB
-static int __pthread_equal_stub(pthread_t t1, pthread_t t2)
-{
- return 1;
-}
-#endif
-
-#ifdef NEED_EXIT_STUB
-static void __pthread_exit_stub(void *ret)
-{
- exit(EXIT_SUCCESS);
-}
-#endif
--
2.10.0
Uli Schlachter
2016-11-04 15:49:18 UTC
Permalink
Post by Emil Velikov
The current design handles the most common use-cases, although it causes
breakage on others (when a pthreads liked library is dlopened). Refer to
the README for further details.
The new design, makes pthread-stubs a "meta" package which _never_
provides a library but a simple .pc file.
pthread-stubs checks if the run-time (libc or otherwise) expose
lightweight pthread symbols to link against and defaults to the full
blown pthread.
This way projects can reuse the Cflags/Libs without having to know the
details. Alternatively they can directly link against the pthreads
implementation, although that might bring unwarranted overhead.
v2: Use correct conditional in configure.ac
---
Makefile.am | 10 --
README | 33 +++-
configure.ac | 64 +++----
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
6 files changed, 545 insertions(+), 141 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4
[...]
Post by Emil Velikov
diff --git a/README b/README
index 4e7e85b..44f39ee 100644
--- a/README
+++ b/README
@@ -1,9 +1,24 @@
-This library provides weak aliases for pthread functions not provided in libc
-or otherwise available by default. Libraries like libxcb rely on pthread
-stubs to use pthreads optionally, becoming thread-safe when linked to
-libpthread, while avoiding any performance hit when running single-threaded.
-libpthread-stubs supports this behavior even on platforms which do not supply
-all the necessary pthread stubs. On platforms which already supply all the
-necessary pthread stubs, this package ships only the pkg-config file
-pthread-stubs.pc, to allow libraries to unconditionally express a dependency
-on pthread-stubs and still obtain correct behavior.
+Originally this library used to provide weak aliases for pthread functions not
+provided in libc or otherwise available by default.
+
+This was found to be a fundamental issue with the design, as the following
+
+Program/library links against pthread-stubs and none of its direct dependencies
+link against pthreads. Thus the weak pthread symbols are used. At a later stage
+a module is dlopened. A one which (directly or via some of its dependencies)
I fail to parse "A one" here.
Post by Emil Velikov
+pulls a full-fledged pthreads implementation. This leads to the weak symbols
+being overridden thus any consecutive calls to the pthreads API [by the
+pthread-stubs user] will be mismatched leading to all sorts of issues.
+
+In order to prevent that one has to traverse up the dependency chain (from the
I'd propose to add a comma after "that", since I first read this as "In
order to prevent that one, has".
Post by Emil Velikov
+pthread-stubs user) and establish whether any of the dependencies will dlopen
+a module which pulls the complete pthread library.
+
+As this is not deemed practical and the alternative solution implemented here
The "and" seems wrong, perhaps replace with ","?
Post by Emil Velikov
+
+The pthread-stubs package only ever provided a .pc file, which provides a set
provided -> provides
Post by Emil Velikov
+of Cflags and Libs.
+ - In the case of pthreads capable libc those are empty
+ - Alternatively they expand to the platform specific PTHREAD_{CFLAGS,LIBS}
Perhaps the README should start with an explanation of the current state
and the history only follows later? Most users would likely want to know
which symbols and definitions are guaranteed by this library.

Perhaps the list from configure.ac should also be mentioned her? After
all, people would want to know which parts of pthreads are
provided/guaranteed by pthread-stubs.
Post by Emil Velikov
diff --git a/configure.ac b/configure.ac
index ef300ad..8962156 100644
--- a/configure.ac
+++ b/configure.ac
[...]

Uli
--
“I’m Olaf and I like warm hugs.”
Emil Velikov
2016-11-04 17:34:33 UTC
Permalink
Post by Uli Schlachter
Post by Emil Velikov
The current design handles the most common use-cases, although it causes
breakage on others (when a pthreads liked library is dlopened). Refer to
the README for further details.
The new design, makes pthread-stubs a "meta" package which _never_
provides a library but a simple .pc file.
pthread-stubs checks if the run-time (libc or otherwise) expose
lightweight pthread symbols to link against and defaults to the full
blown pthread.
This way projects can reuse the Cflags/Libs without having to know the
details. Alternatively they can directly link against the pthreads
implementation, although that might bring unwarranted overhead.
v2: Use correct conditional in configure.ac
---
Makefile.am | 10 --
README | 33 +++-
configure.ac | 64 +++----
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
6 files changed, 545 insertions(+), 141 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4
[...]
Post by Emil Velikov
diff --git a/README b/README
index 4e7e85b..44f39ee 100644
--- a/README
+++ b/README
@@ -1,9 +1,24 @@
-This library provides weak aliases for pthread functions not provided in libc
-or otherwise available by default. Libraries like libxcb rely on pthread
-stubs to use pthreads optionally, becoming thread-safe when linked to
-libpthread, while avoiding any performance hit when running single-threaded.
-libpthread-stubs supports this behavior even on platforms which do not supply
-all the necessary pthread stubs. On platforms which already supply all the
-necessary pthread stubs, this package ships only the pkg-config file
-pthread-stubs.pc, to allow libraries to unconditionally express a dependency
-on pthread-stubs and still obtain correct behavior.
+Originally this library used to provide weak aliases for pthread functions not
+provided in libc or otherwise available by default.
+
+This was found to be a fundamental issue with the design, as the following
+
+Program/library links against pthread-stubs and none of its direct dependencies
+link against pthreads. Thus the weak pthread symbols are used. At a later stage
+a module is dlopened. A one which (directly or via some of its dependencies)
I fail to parse "A one" here.
"A module"
Post by Uli Schlachter
Post by Emil Velikov
+pulls a full-fledged pthreads implementation. This leads to the weak symbols
+being overridden thus any consecutive calls to the pthreads API [by the
+pthread-stubs user] will be mismatched leading to all sorts of issues.
+
+In order to prevent that one has to traverse up the dependency chain (from the
I'd propose to add a comma after "that", since I first read this as "In
order to prevent that one, has".
Post by Emil Velikov
+pthread-stubs user) and establish whether any of the dependencies will dlopen
+a module which pulls the complete pthread library.
+
+As this is not deemed practical and the alternative solution implemented here
The "and" seems wrong, perhaps replace with ","?
Post by Emil Velikov
+
+The pthread-stubs package only ever provided a .pc file, which provides a set
provided -> provides
Post by Emil Velikov
+of Cflags and Libs.
+ - In the case of pthreads capable libc those are empty
+ - Alternatively they expand to the platform specific PTHREAD_{CFLAGS,LIBS}
Perhaps the README should start with an explanation of the current state
and the history only follows later? Most users would likely want to know
which symbols and definitions are guaranteed by this library.
Perhaps the list from configure.ac should also be mentioned her? After
all, people would want to know which parts of pthreads are
provided/guaranteed by pthread-stubs.
Sure thing, can reshuffle things and apply some polish.

Thanks
Emil

Emil Velikov
2016-11-03 20:24:56 UTC
Permalink
From: Emil Velikov <***@collabora.com>

If the pathname has space things will explode in various ways. Avoid
that by adding quotation marks where applicable.

Signed-off-by: Emil Velikov <***@collabora.com>
---
autogen.sh | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/autogen.sh b/autogen.sh
index fc34bd5..f9b61bb 100755
--- a/autogen.sh
+++ b/autogen.sh
@@ -1,13 +1,13 @@
#! /bin/sh

-srcdir=`dirname $0`
+srcdir=`dirname "$0"`
test -z "$srcdir" && srcdir=.

ORIGDIR=`pwd`
-cd $srcdir
+cd "$srcdir"

autoreconf -v --install || exit 1
-cd $ORIGDIR || exit $?
+cd "$ORIGDIR" || exit $?

if test -z "$NOCONFIGURE"; then
$srcdir/configure "$@"
--
2.10.0
Emil Velikov
2016-11-03 20:24:57 UTC
Permalink
This reverts commit fa6db2f9c018c54a47e94c0175450303d700aa92.

This commit causes pthread-stubs to provide a library (with stubs) for
most platforms. This by itself is the goal of the project, although
unveiled severe design issue with the project as a whole.

That will be covered and addressed in a later commit.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98048
---
list.m4 | 3 ---
1 file changed, 3 deletions(-)

diff --git a/list.m4 b/list.m4
index 10c1965..79b5b59 100644
--- a/list.m4
+++ b/list.m4
@@ -13,6 +13,3 @@ alias(zero, int, pthread_cond_signal)
alias(zero, int, pthread_cond_broadcast)
alias(equal, int, pthread_equal)
alias(exit, void, pthread_exit)
-alias(zero, int, pthread_mutexattr_init)
-alias(zero, int, pthread_mutexattr_settype)
-alias(zero, int, pthread_mutexattr_destroy)
--
2.10.0
Uli Schlachter
2016-11-04 15:51:23 UTC
Permalink
Post by Emil Velikov
This reverts commit fa6db2f9c018c54a47e94c0175450303d700aa92.
This commit causes pthread-stubs to provide a library (with stubs) for
most platforms. This by itself is the goal of the project, although
unveiled severe design issue with the project as a whole.
That will be covered and addressed in a later commit.
I'd add some text here.

pthread-stubs won't provide/guarantee these symbols any more. That's "a
good thing", because otherwise it would cause linking with libpthreads
on all(?) platforms.
Post by Emil Velikov
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98048
---
list.m4 | 3 ---
1 file changed, 3 deletions(-)
diff --git a/list.m4 b/list.m4
index 10c1965..79b5b59 100644
--- a/list.m4
+++ b/list.m4
@@ -13,6 +13,3 @@ alias(zero, int, pthread_cond_signal)
alias(zero, int, pthread_cond_broadcast)
alias(equal, int, pthread_equal)
alias(exit, void, pthread_exit)
-alias(zero, int, pthread_mutexattr_init)
-alias(zero, int, pthread_mutexattr_settype)
-alias(zero, int, pthread_mutexattr_destroy)
--
“I’m Olaf and I like warm hugs.”
Emil Velikov
2016-11-03 20:24:59 UTC
Permalink
From: Emil Velikov <***@collabora.com>

As of last commit we no longer create any binaries, thus config.h, CC
and libtool are no longer required.

Signed-off-by: Emil Velikov <***@collabora.com>
---
configure.ac | 5 -----
1 file changed, 5 deletions(-)

diff --git a/configure.ac b/configure.ac
index aeacc77..8206174 100644
--- a/configure.ac
+++ b/configure.ac
@@ -5,11 +5,6 @@ AC_CONFIG_SRCDIR([pthread-stubs.pc.in])
AC_CONFIG_MACRO_DIR([m4])
AM_INIT_AUTOMAKE([foreign dist-bzip2])

-AC_CONFIG_HEADERS([config.h])
-
-AC_PROG_LIBTOOL
-AC_PROG_CC
-
dnl Check if the following functions have stubs.
funclist="\
pthread_self \
--
2.10.0
Emil Velikov
2016-11-03 20:39:30 UTC
Permalink
Post by Emil Velikov
Hi all,
A recent change (commit fa6db2f9c01 - list.m4: Add some new pthread symbols)
lead to an interesting bugreport [1] which makes me think that the current
pthread-stubs design isn't as robust (broken?) as one would expect.
In brief: if there is a pthread-stubs library, one can end up using a
mismatched API - if one uses dlopen with a binary which pulls a full pthreads
implementation. That is because the weak symbols, as provided by the
pthread-stubs DSO, get overridden.
This has gone unnoticed because a) on most platforms pthread-stubs never
creates a DSO and in the rare case where it does b) it's very uncommon to
dlopen a binary which pulls a non-lightweight pthreads symbols, from a
pthreads-stubs (linked) one.
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
2) "kill off" pthread-stubs and let everyone
- handle the "do I need to link against pthreads or not" in their configure.ac
- link against the complete/full pthread API
Forgot to mention:

If people are uneasy with my proposal, we'll be forced to go for 2 +
full link for mesa (and libdrm even). Alternatively please roll a
release and let me know so that we can update mesa/libdrm accordingly.

Thanks
Emil
Jamey Sharp
2016-11-03 20:59:15 UTC
Permalink
I haven't thought about this very hard, or reviewed the actual
patches, or contributed anything to XCB for years... but for what it's
worth:

I have no objection to plan #1. It was probably hubris to think we
could match the system's thread implementation in any meaningful way.
It has worked better than it had any real right to, but as you say,
that's mostly because on most installations pthread-stubs didn't have
to do anything at all.

On the other hand, I think plan #2 would be pretty unfortunate. I
think it would impose a measurable performance penalty for the
majority of users, since nobody's going to want to do the
stubs-detection magic in their configure scripts, so they'll just link
against full pthreads even on systems where they don't need to.

So there's my two cents.
Jamey
Post by Emil Velikov
Hi all,
A recent change (commit fa6db2f9c01 - list.m4: Add some new pthread symbols)
lead to an interesting bugreport [1] which makes me think that the current
pthread-stubs design isn't as robust (broken?) as one would expect.
In brief: if there is a pthread-stubs library, one can end up using a
mismatched API - if one uses dlopen with a binary which pulls a full pthreads
implementation. That is because the weak symbols, as provided by the
pthread-stubs DSO, get overridden.
This has gone unnoticed because a) on most platforms pthread-stubs never
creates a DSO and in the rare case where it does b) it's very uncommon to
dlopen a binary which pulls a non-lightweight pthreads symbols, from a
pthreads-stubs (linked) one.
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
2) "kill off" pthread-stubs and let everyone
- handle the "do I need to link against pthreads or not" in their configure.ac
- link against the complete/full pthread API
Imho 2) is rather selfish/overzealous so I've went ahead and tackled 1).
Note: 1/4 a standalone fix which can be applied independently of how the
direction/discussions reached.
The remaining patches can be squashed or kept separate, based on how things go.
Regards,
Emil
[1] https://bugs.freedesktop.org/show_bug.cgi?id=98048
autogen.sh: quote path variables
Revert "list.m4: Add some new pthread symbols"
Rework the pthread-stub design
configure: drop various AC_macros
Makefile.am | 10 --
README | 33 +++-
autogen.sh | 6 +-
configure.ac | 69 ++++----
list.m4 | 3 -
m4/ax_pthread.m4 | 485 ++++++++++++++++++++++++++++++++++++++++++++++++++++
pthread-stubs.pc.in | 3 +-
stubs.c.m4 | 91 ----------
8 files changed, 548 insertions(+), 152 deletions(-)
create mode 100644 m4/ax_pthread.m4
delete mode 100644 stubs.c.m4
--
2.10.0
Josh Triplett
2016-11-03 21:13:49 UTC
Permalink
Post by Emil Velikov
Hi all,
A recent change (commit fa6db2f9c01 - list.m4: Add some new pthread symbols)
lead to an interesting bugreport [1] which makes me think that the current
pthread-stubs design isn't as robust (broken?) as one would expect.
In brief: if there is a pthread-stubs library, one can end up using a
mismatched API - if one uses dlopen with a binary which pulls a full pthreads
implementation. That is because the weak symbols, as provided by the
pthread-stubs DSO, get overridden.
This has gone unnoticed because a) on most platforms pthread-stubs never
creates a DSO and in the rare case where it does b) it's very uncommon to
dlopen a binary which pulls a non-lightweight pthreads symbols, from a
pthreads-stubs (linked) one.
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
2) "kill off" pthread-stubs and let everyone
- handle the "do I need to link against pthreads or not" in their configure.ac
- link against the complete/full pthread API
Imho 2) is rather selfish/overzealous so I've went ahead and tackled 1).
The pthread stubs on most platforms exist to make the single-threaded
case faster, by not actually locking. Ideally, platform C libraries
would provide those stubs; however, pthread-stubs exists in large part
because of platforms that don't have those stubs. Having stubs even on
more common platforms is a more recent development, and potentially
problematic.

I don't have any fundamental objections here, but Linux isn't the
platform that really needs pthread-stubs in the first place. Could you
please seek out some feedback from people working on platforms that need
pthread-stubs for the core stubs, to find out how this might impact
them? I've CCed a couple of lists of potentially interested folks.

If a few non-Linux platform developers ack this, that would help; I
don't think a change like this should go in with the review coming
exclusively from Linux developers. :)
Samuel Thibault
2016-11-03 22:41:29 UTC
Permalink
Post by Josh Triplett
Post by Emil Velikov
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
I don't have any fundamental objections here, but Linux isn't the
platform that really needs pthread-stubs in the first place. Could you
please seek out some feedback from people working on platforms that need
pthread-stubs for the core stubs, to find out how this might impact
them? I've CCed a couple of lists of potentially interested folks.
Concerning GNU/Hurd and GNU/kFreeBSD, we simply use glibc's forwarding
mechanism, so pthread-stubs is empty there, and thus 1) is fine there.
Note: I'm only talking about GNU/kFreeBSD here, not FreeBSD which has
its own libc which I don't know the status of concerning pthread stubs.

Samuel
Josh Triplett
2016-11-04 00:11:56 UTC
Permalink
Post by Samuel Thibault
Post by Josh Triplett
Post by Emil Velikov
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
I don't have any fundamental objections here, but Linux isn't the
platform that really needs pthread-stubs in the first place. Could you
please seek out some feedback from people working on platforms that need
pthread-stubs for the core stubs, to find out how this might impact
them? I've CCed a couple of lists of potentially interested folks.
Concerning GNU/Hurd and GNU/kFreeBSD, we simply use glibc's forwarding
mechanism, so pthread-stubs is empty there, and thus 1) is fine there.
Note: I'm only talking about GNU/kFreeBSD here, not FreeBSD which has
its own libc which I don't know the status of concerning pthread stubs.
That was part of my hope, namely that there'd be sufficient overlap
between those communities that there might be folks who can speak to
both.
Emil Velikov
2016-11-04 14:01:29 UTC
Permalink
Post by Josh Triplett
Post by Emil Velikov
Hi all,
A recent change (commit fa6db2f9c01 - list.m4: Add some new pthread symbols)
lead to an interesting bugreport [1] which makes me think that the current
pthread-stubs design isn't as robust (broken?) as one would expect.
In brief: if there is a pthread-stubs library, one can end up using a
mismatched API - if one uses dlopen with a binary which pulls a full pthreads
implementation. That is because the weak symbols, as provided by the
pthread-stubs DSO, get overridden.
This has gone unnoticed because a) on most platforms pthread-stubs never
creates a DSO and in the rare case where it does b) it's very uncommon to
dlopen a binary which pulls a non-lightweight pthreads symbols, from a
pthreads-stubs (linked) one.
1) make pthread-stubs "meta package" who's Cflags/Libs expand to the platform
specific PTHREAD_{CFLAGS,LIBS} if the runtime does not provide lightweight
pthread symbols, or
2) "kill off" pthread-stubs and let everyone
- handle the "do I need to link against pthreads or not" in their configure.ac
- link against the complete/full pthread API
Imho 2) is rather selfish/overzealous so I've went ahead and tackled 1).
The pthread stubs on most platforms exist to make the single-threaded
case faster, by not actually locking. Ideally, platform C libraries
would provide those stubs; however, pthread-stubs exists in large part
because of platforms that don't have those stubs. Having stubs even on
more common platforms is a more recent development, and potentially
problematic.
I don't have any fundamental objections here, but Linux isn't the
platform that really needs pthread-stubs in the first place. Could you
please seek out some feedback from people working on platforms that need
pthread-stubs for the core stubs, to find out how this might impact
them? I've CCed a couple of lists of potentially interested folks.
If a few non-Linux platform developers ack this, that would help; I
don't think a change like this should go in with the review coming
exclusively from Linux developers. :)
Indeed and thank you.

So I take it that my analysis does sound about right ? Either way
it'll be great if more people give the scenario mentioned a test [on
their platform of choice].

Cheers,
Emil
P.S. From a quick look FreeBSD still has the DSO [1], while OpenBSD
have grown a completely different pthread-stubs :-\

[1] https://github.com/freebsd/freebsd-ports/blob/d7cb557b47478b4047ff701030e9309f957ab6b2/devel/libpthread-stubs/pkg-plist
Loading...