| 1 | From: IainS
|
|---|
| 2 | To: Juan E Sanchez
|
|---|
| 3 |
|
|---|
| 4 |
|
|---|
| 5 | On 27 May 2010, at 16:23, Juan E Sanchez wrote:
|
|---|
| 6 |
|
|---|
| 7 | > I noticed that you submitted a patch online, that would fix the
|
|---|
| 8 | > current problem I am having with macports on Darwin. The specific
|
|---|
| 9 | > issue is that c++ exception handling is not being handled properly
|
|---|
| 10 | > on Darwin 10.6 when linked against libgcc_s_10.5.so.
|
|---|
| 11 |
|
|---|
| 12 | hmm.
|
|---|
| 13 | I don't see any real issues with the g++ testsuite on gcc-trunk (or
|
|---|
| 14 | 4.5).
|
|---|
| 15 |
|
|---|
| 16 | > Would you happen to know why I might be having this issue?
|
|---|
| 17 |
|
|---|
| 18 | In a way it should not make any difference, since libgcc_s.10.5.dylib
|
|---|
| 19 | is just a stub linked to libSystem on darwin10.
|
|---|
| 20 |
|
|---|
| 21 | I would suspect that your problem is more likely to be that you've
|
|---|
| 22 | built and installed a version of libgcc_s.10.5.dylib that is being
|
|---|
| 23 | found ahead of the system one.
|
|---|
| 24 |
|
|---|
| 25 | This would cause code to be executed from the newly-built libgcc_s
|
|---|
| 26 | instead of libSystem and the unwinder will get confused Big Time.
|
|---|
| 27 |
|
|---|
| 28 | It's essential to make sure that you *don't* build the stubs *unless*
|
|---|
| 29 | you are Apple :-)
|
|---|
| 30 | (or a darwin OS releaser - and I mean stand-alone system release).
|
|---|
| 31 |
|
|---|
| 32 | Therefore you should *not* build gcc with --slibdir= but let the
|
|---|
| 33 | mechanisms I put in place from 4.5 onwards deal with the stub libs
|
|---|
| 34 | (including the _ext which is needed for emutls etc.).
|
|---|
| 35 |
|
|---|
| 36 | I recommend something like:
|
|---|
| 37 |
|
|---|
| 38 | /GCC/source/dir/configure --prefix=/path/to/installation/ --
|
|---|
| 39 | target=x86_64-apple-darwin10 --host=x86_64-apple-darwin10 --
|
|---|
| 40 | build=x86_64-apple-darwin10 --enable-languages=c,objc,c++,obj-c+
|
|---|
| 41 | +,fortran --enable-version-specific-runtime-libs --enable-threads --
|
|---|
| 42 | enable-checking=release --with-libiconv-prefix=/usr --with-system-zlib
|
|---|
| 43 | --program-suffix=-X.Y.Z
|
|---|
| 44 |
|
|---|
| 45 | with host/target adjusted as per the requirements
|
|---|
| 46 | (or drop it so that it's found for different targets - might be easier
|
|---|
| 47 | for macports scenario).
|
|---|
| 48 |
|
|---|
| 49 | e.g:
|
|---|
| 50 | /GCC/source/dir/configure --prefix=/path/to/installation/ --enable-
|
|---|
| 51 | languages=c,objc,c++,obj-c++,fortran --enable-version-specific-runtime-
|
|---|
| 52 | libs --enable-threads --enable-checking=release --with-libiconv-
|
|---|
| 53 | prefix=/usr --with-system-zlib --program-suffix=-X.Y.Z
|
|---|
| 54 |
|
|---|
| 55 | > I notice that the XCode specs file for gcc 4.2 and your specs file
|
|---|
| 56 | > both have this correction:
|
|---|
| 57 | >
|
|---|
| 58 | > %:version-compare(>< 10.5 10.6 mmacosx-version-min= -lgcc_s.10.5)
|
|---|
| 59 |
|
|---|
| 60 | I guess you can always apply a local patch - but first make sure you
|
|---|
| 61 | aren't short-circuiting things (because it *will* upset darwin8 and 9
|
|---|
| 62 | as well as darwin10).
|
|---|
| 63 |
|
|---|
| 64 | Note that this in gcc/config/darwin10.h is also important:
|
|---|
| 65 | #undef LIB_SPEC
|
|---|
| 66 | #define LIB_SPEC "%{!static:-no_compact_unwind -lSystem}"
|
|---|
| 67 |
|
|---|
| 68 | but that's been in gcc-trunk for a while - it should be in 4.5
|
|---|
| 69 |
|
|---|
| 70 | > Has your patch been accepted?
|
|---|
| 71 |
|
|---|
| 72 | not yet,
|
|---|
| 73 | but as I say above, perhaps the real issue lies elsewhere.
|
|---|
| 74 |
|
|---|
| 75 | cheers,
|
|---|
| 76 | Iain
|
|---|
| 77 |
|
|---|
