New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
easy_install of 0.9.0 fails on OS X 10.7 (Lion) (Trac #1476) #2001
Comments
@rgommers wrote on 2011-07-12 This could be due to easy_install or your setup in addition to the OS upgrade. Could you build on Snow Leopard? Can you install on Lion with standard "python setup.py install"? Can you use the standard installed gcc instead of llvm-gcc? |
trac user tacoe wrote on 2011-07-12 Thanks a ton for picking up on this. I did try python setup.py install as well. I'll try and see if there's a way to use gcc instead of llvm (in 10.7, gcc is not shipped anymore; I seem to remember it's not easy getting gcc outside of xcode). |
trac user cax wrote on 2011-07-13 I also have tried to build on Lion with no luck. In my case the problem is when trying to compile scipy/sparse/linalg/eigen/arpack/ARPACK/FWRAPPERS/veclib_cabi_c.c. Python is installed via homebrew. Log: C compiler: /usr/bin/cc -fno-strict-aliasing -O3 -march=core2 -msse4.1 -w -pipe -DNDEBUG -g -fwrapv -O3 -Wall -Wstrict-prototypes compile options: '-Iscipy/sparse/linalg/eigen/arpack/ARPACK/SRC -I/usr/local/Cellar/python/2.7.2/lib/python2.7/site-packages/numpy/core/include -c' |
@rgommers wrote on 2011-07-13 Never mind gcc then. Quite surprising that it's not shipped with OS X anymore. It seems to be complaining about this:
in particular the |
trac user cax wrote on 2011-07-14 It seems this is an "old" issue. See http://projects.scipy.org/scipy/ticket/510. It is funny because they closed the issue by saying Panther (10.3.9) was too old. By adding #include <complex.h> in scipy/sparse/linalg/eigen/arpack/ARPACK/FWRAPPERS/veclib_cabi_c.c seems to fix the issue. However, I've got another one: |
trac user deadshort wrote on 2011-07-14 Section 7.3.1 of http://www.open-std.org/jtc1/sc22/WG14/www/docs/n1256.pdf suggests that adding |
trac user dloewenherz wrote on 2011-07-17 Adding
|
@rgommers wrote on 2011-07-18 @cax/dloewenherz: that error could be related to the 10.4 SDK, did you install it? Did you install numpy without problems and does @deadshort: what about compilers without (full) C99 support? There are quite a few compilers we need to support. |
trac user dloewenherz wrote on 2011-07-18
[http://i.imgur.com/8CXQA.png] |
@rgommers wrote on 2011-07-18 17 errors, those are probably all the Fortran related tests in numpy.distutils. Right? Is your gfortran version 4.2.3 from http://r.research.att.com/tools/? Did you install numpy 1.6.0, or something else? Each OS X version has an SDK, it's on the same install dvd as XCode but in a separate folder, called optional installs or something like that. You need it if you are using the Python from python.org |
trac user dloewenherz wrote on 2011-07-18 Actually, most of the errors I see are mostly related to mathematical operations, such as sqrt. I do see some fortran-related errors, such as the following:
I'm using gfortran version 4.2.4. I'm running these tests off of Numpy 2.0.0. Apple must have stopped releasing the SDKs in an optional install folder. I only see one package (Xcode.mpkg), and "About Xcode.pdf". I have a biting suspicion that the reason for these failures is the removal of gcc from the OS X toolchain. Everything is now being compiled through LLVM. |
trac user deadshort wrote on 2011-07-18 Messy. I'm not sure there's a simple correct answer. The C99 spec recommends wrapping something or other behind I have been using +gcc44 +no_atlas, and scipy builds and installs, but then fails 6 BLAS-related tests. I feel I should test a few variations.
|
@rgommers wrote on 2011-07-19 Why no_atlas, and how? With the default build options you should pick up Accelerate, does that fix your failures? |
trac user dloewenherz wrote on 2011-07-19 Still don't have much of an idea of what's going on here... P.S. Lion is being released tomorrow. |
trac user chrisfonnesbeck wrote on 2011-07-21 I can confirm this bug. I have been trying all day to get SciPy built, but this failure gets me every time. Numpy builds without issue. Using Xcode 4.1, gfortran 4.2.3 with Python 2.7.1. |
trac user rsenk330 wrote on 2011-07-21 I also can confirm this. I got past the error in the ticket description above by adding
|
trac user cshimmin wrote on 2011-07-22 Another confirmation here. I haven't bothered messing with veclib_cabi_c.c yet... it seems from the other reports that this isn't a complete fix. Installing from pip. |
trac user chrisfonnesbeck wrote on 2011-07-22 A bit more info: a full test of numpy results in the following errors. As they have to do with umath complex, are they relevant to the reported bug?
|
trac user rsenk330 wrote on 2011-07-23 I can get this to install using gfortran from homebrew. I had to use the gfortran formula from a testing repository until it gets pushed to the master. Instructions can be found on this [https://github.com/Homebrew/legacy-homebrew/issues/6500#issuecomment-1637840 homebrew ticket]. I also had to add
|
trac user cshimmin wrote on 2011-07-26 It looks like the updated gfortran binary from r.research.att.com has made its way into homebrew. With the new binary and adding complex.h to the files indicated by @rsenk330, scipy installs successfully. However, when I run scipy.test() from ipython, I get the following sever error, causing python to abort:
I am seeing the exact numpy test failures as @chrisfonnesbeck above. They all appear to be related to complex square roots (TestCsqrt). Is there already an open ticket for this somewhere? |
@rgommers wrote on 2011-07-27 There's no ticket for the complex square root failures yet, but it should be unrelated. They're corner cases of the C99 spec that are being tested, fail on several other platforms and have little practical relevance. |
trac user mhansen wrote on 2011-07-28 I can confirm that adding
works for me when trying to build SciPy for Sage under Lion with XCode4. |
Milestone changed to |
trac user cshimmin wrote on 2011-08-10 @rgommers: I am seeing at least 4 different failure modes. I have just compiled a fresh copy of scipy 0.9.0 from PyPI (after adding patching in the #include <complex.h> lines discussed here.) Here's the (top-truncated) output from a couple of runs of scipy.test(verbose=2):
Another failure mode I saw included a warning (sorry, didn't catch this one with verbosity):
Finally, in the last failure mode I have observed when running scipy.test(), the python interpreter hangs at 100% cpu, and doesn't respond to sigint. It died silently upon sigterm though. I'm guessing at least some of this non-deterministic behavior is from the test_random_* functions, but I haven't spent time looking into this. Let me know if any additional information can help. |
trac user cshimmin wrote on 2011-08-10 Just to be clear: although the failure modes seem non-deterministic, I should mention that I have never run test() successfully; the farthest I have seen it get is to "test_definition ... Segmentation fault 11". And as a bonus, here are 2 new failures that I just observed:
Perhaps this issue needs its own ticket? |
Original ticket http://projects.scipy.org/scipy/ticket/1476 on 2011-07-11 by trac user tacoe, assigned to @cournape.
This is a GM build of Lion, meaning no changes are expected before release.
From the first error to the end of the trace:
The text was updated successfully, but these errors were encountered: