[Segfault] Graph.pagerank()

[ntfrgl]

Arch Linux 3.15.1
Python 2.7.7
igraph 0.7.1
python-igraph 0.7.0

import igraph
igraph.Graph(n=2, edges=[[0, 1], [1, 0]]).pagerank()

Consistently crashes with an error message like the following:

[:] *** Process received signal ***
[:] Signal: Segmentation fault (11)
[:] Signal code: Address not mapped (1)
[:] Failing at address: 0x7fc23f506255
[:] [ 0] /usr/lib/libpthread.so.0(+0xf4b0)[0x7fc21d53f4b0]
[:] [ 1] /usr/lib/libpython2.7.so.1.0(PyObject_IsTrue+0x1f)[0x7fc21d7d508f]
[:] [ 2] /usr/lib/python2.7/site-packages/igraph/_igraph.so(igraphmodule_Graph_personalized_pagerank+0x45b)[0x7fc2136f001b]
[:] [ 3] /usr/lib/libpython2.7.so.1.0(PyEval_EvalFrameEx+0x56c9)[0x7fc21d82f669]
[...]
[:] [17] /usr/lib/libpython2.7.so.1.0(PyEval_EvalCodeEx+0x830)[0x7fc21d830970]
[:] [18] /usr/lib/libpython2.7.so.1.0(PyEval_EvalCode+0x19)[0x7fc21d830a69]
[:] [19] /usr/lib/libpython2.7.so.1.0(+0xfbeaf)[0x7fc21d849eaf]
[:] [20] /usr/lib/libpython2.7.so.1.0(PyRun_FileExFlags+0x72)[0x7fc21d84b0d2]
[:] [21] /usr/lib/libpython2.7.so.1.0(PyRun_SimpleFileExFlags+0xe7)[0x7fc21d84c2d7]
[:] [22] /usr/lib/libpython2.7.so.1.0(Py_Main+0xc7a)[0x7fc21d85d66a]
[:] [23] /usr/lib/libc.so.6(__libc_start_main+0xf0)[0x7fc21d1a2000]
[:] [24] python2[0x40073e]
[:] *** End of error message ***
zsh: segmentation fault (core dumped)

Everything else works. The same code runs fine on OS X. Perhaps something is being interpreted differently between gcc and clang?

[ntamas]

Also works on Ubuntu Linux where igraph is compiled with gcc so I'm pretty sure it's not a gcc vs clang issue. I'll try to install Arch in a VM soon and reproduce it.

[dacapo1142]

I've got this problem too on Ubuntu 14.04 (python3, python-igraph0.7)

[ntfrgl]

Is there any further useful information I could provide?

[ntamas]

Finally managed to reproduce the issue in an Ubuntu VM. I'll keep this issue updated as I make progress.

[ntamas]

Found a workaround (although the cause is still not clear): just call personalized_pagerank() instead of pagerank().

[ntfrgl]

Thank you! That works for me.

[crsqq]

Got the same symptoms using two different environments: Ubuntu 14.04 and OS X 10.9, both with python (3.4.0 / 3.4.1), python-igraph==0.7 and igraph 0.7.1.

[ntamas]

Fun fact: this does not happen when using the debug build of Python on Ubuntu, only with the release build.

[ntamas]

Okay, this is fixed in the development branch now; see the patch above.

[ntamas]

For the record: the problem was that the eps argument was parsed as a double on the C side even though the variable that was supposed to store the parsed value was declared as a float. The workaround works because calling personalized_pagerank directly does not pass any value to the eps argument so the default does not get overwritten. An alternative workaround is to update the Graph.pagerank() method in igraph/__init__.py so it does not forward the eps argument to personalized_pagerank.

[dimpase]

Why was this closed? There is still a segfault on pagerank() in the latest release, although not in the development version (as we see on https://trac.sagemath.org/ticket/27502).

[ntamas]

I tend to close issues after the fix has been committed and not after releases. However, development has stalled on python-igraph as I don't have time for it any more so there was no new release since the fix was committed. Your best bet right now is to compile the Python interface from source instead.

[dimpase]

For igraph, from the current master I am able to make a release tarball (after some changes in src/Makefile.am, which I should offer as a PR), and it passes the testsuite on Linux (I didn't try anything else, yet). Would you be willing to make at least a pre-release based on this? After all it'd be just 5 minutes or so at the keyboard for you...

[ntamas]

Sorry, it's not that easy, but I wanted to let you know that it seems like igraph's development will hopefully gain some momentum; we aim for a proper 0.8.0 release next month.