I found a problem in execution_monitor.cpp of Boost.Test on POSIX systems.
The file uses the sigsetjmp() and siglongjmp() functions and the sigjmp_buf data type.
They all are defined by POSIX as an extention to the ANSI-C standard, i.e. the interface is defined in a header file defined by ANSI-C (<setjmp.h> in this case) and the extension has to be enabled by #defining certain symbols before #including any of the header files defined by ANSI-C or POSIX.
The glibc implementation of the POSIX API relaxes that requirement: when the symbol __STRICT_ANSI__ isn't defined and none of the POSIX symbols is defined, either, then a couple of symbols gets defined in <features.h>, thus enabling most of the POSIX stuff. Also, defining the symbol _GNU_SOURCE enables all of the POSIX stuff.
gcc defines __STRICT_ANSI__ when the -ansi flag is used. gcc versions >= 3 strangely define _GNU_SOURCE (on Linux systems. On cygwin they don't define that symbol). Hence, with gcc versions >= 3 or when -ansi isn't used with gcc, execution_monitor.cpp compiles fine. With -ansi and -U_GNU_SOURCE the compilation of execution_monitor.cpp fails.
Other compilers probably also enable POSIX by default.
However, como defines __STRICT_ANSI__ (in non-scrict and in scrict modes) and doesn't define _GNU_SOURCE or any of the POSIX enabling symbols. Therefore, compiling execution_monitor.cpp fails for como unless _GNU_SOURCE or one of the POSIX enabling symbols is defined explicitely.
In order to correctly enable the POSIX interface we need to get _POSIX_SOURCE defined for older POSIX implementations and one of _POSIX_C_SOURCE or _XOPEN_SOURCE defined to an appropriate value for newer POSIX implementations.
However, this isn't easy, because the POSIX feature flags have to be set before any of the POSIX or ANSI-C header are #included. One way to accomplish this would be on the commandline of the compiler.
When I tried to add a define to the DEFINES variable of the como tools set I ran into the problem described in a seperate posting to the Jamboost mailing list. Also, I think, the como (or any other) toolset or the Boost.Test Jamfile would be the wrong places to add the define(s). The users should have a chance to specify different POSIX feature flags.
However, for the regression tests one of the feature flags should be set. Therefore, I suggest to add -D_POSIX_SOURCE and -D_POSIX_C_SOURCE=1 to the command- lines of all compile commands for tests on POSIX systems (AFAIK, similar problems for other POSIX systems haven't been reported yet. However, the POSIX standard requires these symbols to be defined). Maybe, Boost.Thread requires a different value for POSIX_C_SOURCE.
The documentation for Boost.Test and for Boost.Thread should mention the need to enable the POSIX API on POSIX systems.
Regards, m
_______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
