sbcl +threads does not compile maxima (snow leopard, x86_64)

[daitakahashi]

SBCL 1.0.48 (64bit) with +thread variant failed to build maxima on Mac OS X 10.6.7 (Macbook Pro, early 2011). However, the SBCL (same version) without +threads can build maxima successfully.

The last some lines of port -d install maxima are shown below (and build process hung since ldb was invoked). Thank you very much.

;      - Loading source file
;        "/opt/local/var/macports/build/_opt_local_var_macports_sources_svn.macports.org_trunk_dports_math_maxima/work/maxima-5.24.0/src/init-cl.lisp"
; loading #P"/opt/local/var/macports/build/_opt_local_var_macports_sources_svn.macports.org_trunk_dports_math_maxima/work/maxima-5.24.0/src/init-cl.lisp"
;  - Providing system maxima
[undoing binding stack and other enclosing state... done]
[saving current Lisp image into binary-sbcl/maxima.core:
fatal error encountered in SBCL pid 79417(tid 140735075560608):
no size function for object at 0x00808f40 (widetag 0x40)


scanning space for lutexes...
writing 6416 bytes from the read-only space at 0x20000000
scanning space for lutexes...
writing 4064 bytes from the static space at 0x20100000
scanning space for lutexes...
Welcome to LDB, a low-level debugger for the Lisp runtime environment.

[daitakahashi]

Complete build log of maxima

[nikolaus@…]

I think this is more of a SBCL issue rather than macports, since SBCL thread support on OS X is (and has been for while) quite broken. For 64bit OS X it was never supported anyways as far as I know.

[easye]

Hmm. I run sbcl +threads from MacPorts on x86_64 constantly without seeing problems, although mainly I just compile and test ASDF definitions via ​QuickLisp to test compatibility, so I would not claim to be stressing SBCL +threads in my usage. But it may be more stable than Nikolaus' comment suggests…

[daitakahashi]

I am not completely sure if this is a SBCL issue or a problem comes from something different causes. But I would like to follow Nikolaus's comment and disable threads variant for reliability.

Following QuickLisp web-page, I confirmed the vecto package can be compiled without any problem (threads enabled x86_64 SBCL). So the thread enabled SBCL sometimes works fine, however, it may break down at very stressful condition (e.g., maxima build).

Thank you very much for your helpful comments.

[daitakahashi]

After some experiments, I found some relationships between the problem and a (path) length of build directory name. I set up two directories,

1. short
2. something_very_very_long_directory_name_to_test_if_the_length_of_build_directory_name_affects_maxima_build_process.

Both of them include vanilla maxima 5.24.0 source downloaded from website, and I tested cd maxima-5.24.0 && ./configure --enable-sbcl && make.

sbcl @1.0.48_0 successfully built maxima for both cases, however, sbcl @1.0.48_0+threads failed at case 2 (no problem was found at case 1). The reason of this odd behavior is still not clear, though, I think this is problematic for macports, which uses automatically generated (sometimes very long) name for build directory.

[kencu (Ken)]

The particular issue in this ticket is no longer seen with new versions of sbcl.