Igraph's pagerank not working in Linux

[2460b664-5ecd-43fc-9bbe-d0f333762988]

Following error arise in linux but works fine in Mac OS.

sage: G = graphs.RandomGNP(1000, .01)
sage: I = G.igraph_graph()
sage: %timeit I.pagerank()

gives following error

        filename = 0x7ffc0bef5337 "/home/rajat/new_version/sage-8.7.beta6/src/bin/sage-ipython"
        module = 0x0
        fp = 0x1b033f0
        p = <optimized out>
        unbuffered = 0
        skipfirstline = 0
        stdin_is_interactive = 1
        help = <optimized out>
        version = <optimized out>
        saw_unbuffered_flag = <optimized out>
        cf = {cf_flags = 0}
#78 0x00007ff538f15830 in __libc_start_main (main=0x4006f0 <main>, argc=3, 
    argv=0x7ffc0bef4708, init=<optimized out>, fini=<optimized out>, 
    rtld_fini=<optimized out>, stack_end=0x7ffc0bef46f8)
    at ../csu/libc-start.c:291
        result = <optimized out>
        unwind_buf = {cancel_jmp_buf = {{jmp_buf = {0, 164082578521821197, 
                4196096, 140720508716800, 0, 0, -162438125174419443, 
                -167616841612554227}, mask_was_saved = 0}}, priv = {pad = {
              0x0, 0x0, 0x0, 0x0}, data = {prev = 0x0, cleanup = 0x0, 
              canceltype = 0}}}
        not_first_call = <optimized out>
#79 0x0000000000400729 in _start ()
No symbol table info available.

Cython backtrace
----------------

29  ../sysdeps/unix/sysv/linux/waitpid.c: No such file or directory.
Traceback (most recent call last):
  File "<string>", line 56, in <module>
  File "/usr/lib/python3/dist-packages/Cython/Debugger/libcython.py", line 689, in invoke
    for arg in string_to_argv(args):
TypeError: argument 1 must be str, not bytes
Saved trace to /home/rajat/.sage/crash_logs/crash_L9MmKW.log
------------------------------------------------------------------------
Unhandled SIGSEGV: A segmentation fault occurred.
This probably occurred because a *compiled* module has a bug
in it and is not properly wrapped with sig_on(), sig_off().
Python will now terminate.
------------------------------------------------------------------------
Segmentation fault (core dumped)

Also see https://groups.google.com/forum/#!topic/sage-devel/NtVP5AQgqNs

---

updates, tarballs: https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.tar.gz and https://github.com/dimpase/python-igraph/releases/download/0.7.1999/python_igraph-0.7.1999.tar.gz

Upstream: Fixed upstream, but not in a stable release.

CC: @dimpase @dcoudert

Component: packages: optional

Author: Dima Pasechnik

Branch/Commit: 721ccc3

Reviewer: David Coudert

Issue created by migration from https://trac.sagemath.org/ticket/27502

[sheerluck]

comment:4

Using latest code from both https://github.com/igraph/igraph.git and https://github.com/igraph/python-igraph.git I get

sage: G = graphs.RandomGNP(1000, .01)
sage: I = G.igraph_graph()
sage: %timeit I.pagerank()
1.27 ms ± 13.4 µs per loop (mean ± std. dev. of 7 runs, 1000 loops each)

So upstream have fixed some bugs already

[dcoudert]

comment:5

Thank you for reporting this. So we must update igraph.

@dimpase: I have never tried to add or update an external package. Your help is more than welcome :P

[dimpase]

comment:6

For some reason, upstream closed https://github.com/igraph/igraph/issues/636, providing a workaround (using perzonalized_pagerank() instead), rather than a fix...

[dimpase]

Upstream: Reported upstream. No feedback yet.

[dimpase]

comment:7

I'd like to wait for upstream to react - update packages using git repos rather than upstream tarballs is a bit of a hassle.

[dimpase]

comment:8

OK, upstream says it's fixed, but not in a release, and new releases won't appear any time soon. @rajat1433 - could you make github forks and create new releases there, or I can do this, whatever you prefer.

[dimpase]

Changed upstream from Reported upstream. No feedback yet. to Fixed upstream, but not in a stable release.

[2460b664-5ecd-43fc-9bbe-d0f333762988]

comment:9

Replying to @dimpase:

OK, upstream says it's fixed, but not in a release, and new releases won't appear any time soon. @rajat1433 - could you make github forks and create new releases there, or I can do this, whatever you prefer.

I don't have much of an experience in creating new releases. If you can guide me through steps I can do it. But since you have an experience you can do it faster so your help is more than welcome.

[kiwifb]

comment:10

We could just cherry-pick https://github.com/igraph/igraph/commit/923974d7520bc49a6b52d34700fb25f0cea66718 taking a snapshot is much more risky and may imply changes of behavior.

[dimpase]

comment:11

I've managed to make a tarball from the master that also passes the testsuite (SAGE_CHECK=yes ./sage -f igraph) it's here https://github.com/dimpase/igraph/releases/tag/0.7.1999

[dimpase]

Commit: 721ccc3

[dimpase]

Branch: u/dimpase/packages/igraph_etc

[dimpase]

comment:12

please review these, and test on OSX. On Linux this passes both SAGE_CHECK=yes ./sage -f igraph python_igraph and the relevant doctests in src/sage/graphs/

---

New commits:

721ccc3

updated igraph and python_igraph to master

[dimpase]

Description changed:

--- 
+++ 
@@ -56,3 +56,10 @@

Also see https://groups.google.com/forum/#!topic/sage-devel/NtVP5AQgqNs + +--- + +updates, tarballs: +https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.1.tar.gz +and +https://github.com/dimpase/python-igraph/releases/download/0.7.1999/python_igraph-0.7.1999.tar.gz

[dimpase]

Author: Dima Pasechnik

[dcoudert]

comment:14

What's the right set of instructions to test this patch (I have never done that yet, or I don't remember). Thanks.

[dimpase]

comment:15

You need to download the linked tarballs to upstream/ And the usual git thing, naturally.

[dimpase]

comment:16

And ./sage -f python_igraph

[dcoudert]

comment:17

Thank you.

I add to rename file https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.1.tar.gz to https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.tar.gz (so .1.tar.gz -> .tar.gz)

Then everything is working well and I can do

sage: G = graphs.PetersenGraph()
sage: I = G.igraph_graph()
sage: I.pagerank()
[0.09999999999999999,
 0.09999999999999999,
 0.10000000000000002,
 0.10000000000000002,
 0.1,
 0.1,
 0.1,
 0.09999999999999999,
 0.1,
 0.09999999999999999]

Someone should double check on linux.

[dimpase]

comment:18

Yep, a typo in file name, sorry. Fixed.

[dimpase]

Description changed:

--- 
+++ 
@@ -60,6 +60,6 @@
 ---

 updates, tarballs:
-https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.1.tar.gz
+https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.tar.gz
 and
 https://github.com/dimpase/python-igraph/releases/download/0.7.1999/python_igraph-0.7.1999.tar.gz

[dcoudert]

comment:19

I'm no able to access https://github.com/dimpase/igraph/releases/download/0.7.1999/igraph-0.7.1999.tar.gz

[dimpase]

comment:20

Sorry, my fault. Please try now (and check that it is identical to the one you renamed).

[dcoudert]

comment:21

For me the files are the same.

I tried on a linux computer and it's working well. I can use the pagerank method of igraph.

For me this patch is good to go. Should we update the ticket title and description before setting it to positive review?

[dimpase]

comment:22

How about we add a doctest with pagenumber computation?

[kiwifb]

comment:23

Should we really? We had those conversations in the last few weeks. Do we really want to add tests to sage about bugs in upstream packages? No, if anything the test belongs to the upstream package. This is a case in point and adding a test to sage is scope creep.

[dcoudert]

comment:24

We plan to add method pagerank to generic graphs. It will call either methods of networkx (pure Python, or using numpy or scipy), or if installed igraph. So we will certainly have a test to check that it's working well, as we have with other optional packages (e.g., bliss, tdlib, etc.).

[kiwifb]

comment:25

That is fine, you are just checking that sage works, not that an upstream package as been fixed through sage.

[jhpalmieri]

comment:26

For what it's worth, with this branch I get two failures when I do sage -f -c igraph on OS X Mojave:

...
124: Merging layouts 2 (igraph_i_layout_merge):      FAILED (layout.at:62)
...
241: SCG of a graph, stochastic matrix (igraph_scg) : FAILED (scg.at:73)
...
ERROR: 249 tests were run,
2 failed unexpectedly.
2 tests were skipped.
## -------------------------- ##
## testsuite.log was created. ##
## -------------------------- ##

The previous version of igraph failed two other (?) tests, so I don't think the new failures should necessarily be an obstacle for a positive review. Failures from old version:

...
 25: Sparse matrix, solvers (igraph_sparsemat_t):    FAILED (types.at:148)
...
 32: Another sparse matrix (igraph_spmatrix_t):      FAILED (types.at:183)
...
ERROR: All 234 tests were run,
2 failed unexpectedly.
## -------------------------- ##
## testsuite.log was created. ##
## -------------------------- ##

[dcoudert]

Reviewer: David Coudert

[dcoudert]

comment:27

I also tried sage -f -c igraph with

• 0 error a linux computer (fedora 28) equipped with 2 4-cores Intel(R) Xeon(R) CPU W5580 @ 3.20GHz (10 years old, but still operational)
• 2 errors an OSX laptop, same as in #comment:26

I set this ticket as positive review. It fixes some issues, and I don't know where pre-existing issues come from.

[embray]

comment:28

Ticket retargeted after milestone closed (if you don't believe this ticket is appropriate for the Sage 8.8 release please retarget manually)

[vbraun]

Changed branch from u/dimpase/packages/igraph_etc to 721ccc3