|
Boost : |
From: Robert Ramey (ramey_at_[hidden])
Date: 2004-11-19 23:59:25
Assuming the debug version builds and passes all tests.
In the serialization Jamfile, set the release mode optimization switches to
be the same as the debug ones.
Try to build. If it passes add optimization switches one at a time until we
find the "magic one" which is creating the problem. Then we either alter
the Jamfile or amend xml_grammar.cpp with the appropriate #pragma to
suppress that optimization.
I know its tedious but its the only way I know.
Keep me posted
Robert Ramey
"Steve Hartwell" <shspamsink_at_[hidden]> wrote in message
news:661E9EC6-39FE-11D9-BE71-003065BC30C0_at_comcast.net...
> Could other MacOS X developers confirm that the boost build 1.32
> candidate fails to build on MacOS 10.3.6 with gcc 3.3?
>
> The problem I'm seeing is that while compiling "xml_grammar.cpp",
> cc1plus slows to a crawl while its virtual memory allocation rises to
> 1.8GB until my system becomes so unresponsive I have to kill it. Jam
> then proceeds until the release build of "xml_grammar.cpp", with the
> same results (cc1plus hanging). Killing that process leads to the same
> problems in the debug compile of "xml_wgrammar.cpp", as well as the
> release compile of "xml_wgrammar.cpp".
>
> I have observed this problem both with a dual-processor G5 and a
> single-processor G4. Everything else in the build seems to be proceed
> OK after I kill the cc1plus processes that hang on "xml_grammar.cpp".
>
> The command line I'm using to build is:
> bjam -sTOOLS=darwin -sBUILD="debug release
> <runtime-link>static/dynamic"
>
> Please let me know if anyone is having better luck with this.
>
> Best regards,
>
> Steve Hartwell
>
> p.s. There is an additional difficulty with the build of boost
> 'serialization' where some directory names end with ".a" and ".dylib".
> Although these are legal directory names, this confuses the heck out of
> the Xcode IDE, and makes it impossible to include the contents of these
> named directories into an Xcode project. It would be nice if the names
> of these directories could be changed to end with "_a" and "_dylib"
> instead, or something similar.
>
> p.p.s. boring (but precise) stuff about the cc1plus hang follows:
>
> When cc1plus hangs, ps shows the following arguments:
>
> /usr/libexec/gcc/darwin/ppc/3.3/cc1plus -quiet
> -Ibin/boost/libs/serialization/build -I/usr/include
> -I/Volumes/Nod/Users/hartwell/Documents/Development/Building
> Blocks/C-C++/BOOST/boost_1_32_0 -D__GNUC__=3 -D__GNUC_MINOR__=3
> -D__GNUC_PATCHLEVEL__=0 -D__APPLE_CC__=1666 -D__DYNAMIC__ -DNDEBUG
> -DNDEBUG -DBOOST_TEST_NO_AUTO_LINK=1
> /Volumes/Nod/Users/hartwell/Documents/Development/Building
> Blocks/C-C++/BOOST/boost_1_32_0/libs/serialization/build/../src/
> xml_wgrammar.cpp -D__GNUG__=3 -fPIC -quiet -dumpbase xml_wgrammar.cpp
> -auxbase-strip
> bin/boost/libs/serialization/build/libboost_wserialization.a/darwin/
> release/runtime-link-static/xml_wgrammar o -O3 -Wno-long-double
> -Wno-inline -Wall -ftemplate-depth-256 -finline-functions
> -fcoalesce-templates -D__private_extern__=extern -o
>
> Since this is all 'ps' is willing to output for this process, here is
> the parent process, whose arg list seems more complete:
>
> c++ -c -ftemplate-depth-256 -DNDEBUG -DNDEBUG
> -DBOOST_TEST_NO_AUTO_LINK=1 -Wno-long-double -no-cpp-precomp -O3
> -finline-functions -Wno-inline -Wall -fcoalesce-templates
> -Ibin/boost/libs/serialization/build -I/usr/include
> -I/Volumes/Nod/Users/hartwell/Documents/Development/Building
> Blocks/C-C++/BOOST/boost_1_32_0 -o
> bin/boost/libs/serialization/build/libboost_wserialization.a/darwin/
> release/runtime-link-static/xml_wgrammar.o
> /Volumes/Nod/Users/hartwell/Documents/Development/Building
> Blocks/C-C++/BOOST/boost_1_32_0/libs/serialization/build/../src/
> xml_wgrammar.cpp
>
> _______________________________________________
> Unsubscribe & other changes:
http://lists.boost.org/mailman/listinfo.cgi/boost
>
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk