cmdline.py doctest failures on fulvia (SunOS)

[jdemeyer]

On the skynet machine fulvia (SunOS 5.10 i86pc), we get some doctest failures in sage/tests/cmdline.py (added by #10300):

sage -t -long  -force_lib devel/sage/sage/tests/cmdline.py
**********************************************************************
File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/devel/sage-main/sage/tests/cmdline.py", line 158:
    sage: len(out) >= 2   # at least one character + newline
Expected:
    True
Got:
    False
**********************************************************************
File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/devel/sage-main/sage/tests/cmdline.py", line 160:
    sage: err
Expected:
    ''
Got:
    '/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/local/bin/sage-sage: line 338: readlink: command not found\n'
**********************************************************************
File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/devel/sage-main/sage/tests/cmdline.py", line 173:
    sage: (out, err, ret) = test_executable(["sage", "--startuptime"])
Exception raised:
    Traceback (most recent call last):
      File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/local/bin/ncadoctest.py", line 1231, in run_one_test
        self.run_one_example(test, example, filename, compileflags)
      File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/local/bin/sagedoctest.py", line 38, in run_one_example
        OrigDocTestRunner.run_one_example(self, test, example, filename, compileflags)
      File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/local/bin/ncadoctest.py", line 1172, in run_one_example
        compileflags, 1) in test.globs
      File "<doctest __main__.example_1[54]>", line 1, in <module>
        (out, err, ret) = test_executable(["sage", "--startuptime"])###line 173:
    sage: (out, err, ret) = test_executable(["sage", "--startuptime"])
      File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/local/lib/python/site-packages/sage/tests/cmdline.py", line 366, in test_executable
        raise RuntimeError("timeout in test_executable()")
    RuntimeError: timeout in test_executable()
**********************************************************************
File "/home/buildbot/build/sage/fulvia-1/fulvia_full/build/sage-4.6.1.alpha3/devel/sage-main/sage/tests/cmdline.py", line 174:
    sage: out.find("sage.all: ") >= 0
Expected:
    True
Got:
    False
**********************************************************************

For the first, we need something analogous to the readlink command, for the second we simply need to increase the timeout.

[jdemeyer]

The line

readlink "$SAGE_ROOT/devel/sage" | sed "s/sage-//"

from sage-sage can probably be replaced by:

cd "$SAGE_ROOT/devel/sage" && pwd -P | sed 's|.*/||; s|^sage-||'

which is a lot uglier, but at least it works on Linux, OS X 10.4, SunOS 5.10.

I will prepare a patch.

[jdemeyer]

SCRIPTS patch

[jdemeyer]

SAGELIB patch

[drkirkby]

Another option would have been to use the readline system call, which is defined by POSIX

​http://www.opengroup.org/onlinepubs/009695399/functions/readlink.html

but not the command, where POSIX defines no such command.

dave

[drkirkby]

I wont have time to look at this today. Not knowing what the command is supposed to do, I'll have to look that up first.

dave

[ddrake]

Two lines changed; the readlink replacement will definitely work and increasing the startuptime timeout will at least allow us to avoid the problem (we should, of course, just get Sage to start faster...) for now.

[drkirkby]

Part of the timeout issue on fulvia is the long running background jobs. I'm not sure if the buildbot disables them (there's a README somewhere under /usr/local). But otherwise that machine will be dog slow.

If anyone wants an account on hawk which is running a similar operating system to Solaris 10, but is a much faster machine, I don't mind giving them one.

It takes about 3 seconds to start Sage on my 3.33 GHz Sun. I can't measure the time it takes to start Mathematica - at least not without using some function to measure it with ns grandularity. (OK, not quite ns, but the Mathematica 7 command line starts effectively instantly. The GUI takes about 5 seconds, which is in fact a bit longer than Sage).

[leif]

Replying to jdemeyer:

The line

readlink "$SAGE_ROOT/devel/sage" | sed "s/sage-//"

from sage-sage can probably be replaced by:

cd "$SAGE_ROOT/devel/sage" && pwd -P | sed 's|.*/||; s|^sage-||'

which is a lot uglier, but at least it works on Linux, OS X 10.4, SunOS 5.10.

Just for the record:

SAGE_ROOT/sage also uses realpath if readlink is not available / fails.

Don't know what systems (besides Linux) support that though.

Since we rely on Bash, cd ... && pwd -P ... is of course more "portable" (i.e., generic, and actually POSIX), and likely to be a couple of femtoseconds faster (unless cd takes too long ;-) ).

Note that there is a slight difference: readlink also reads dead links (deleted current Sage branches), while cd of course fails in that case.

[leif]

Replying to leif:

Note that there is a slight difference: readlink also reads dead links (deleted current Sage branches), while cd of course fails in that case.

Perhaps keep also the old line, e.g.

    (readlink "$SAGE_ROOT/devel/sage" || 
      (cd "$SAGE_ROOT/devel/sage" && pwd -P | sed 's|.*/||') ) \
    | sed 's|^sage-||'

[leif]

And should we give an appropriate error message (perhaps switching to the main branch) in case the current was deleted?

[jdemeyer]

Replying to leif:

Perhaps keep also the old line, e.g.

    (readlink "$SAGE_ROOT/devel/sage" || 
      (cd "$SAGE_ROOT/devel/sage" && pwd -P | sed 's|.*/||') ) \
    | sed 's|^sage-||'

Why?

[jdemeyer]

Replying to leif:

Just for the record:

SAGE_ROOT/sage also uses realpath if readlink is not available / fails.

The Skynet system fulvia doesn't have readlink nor realpath.

[jdemeyer]

Replying to leif:

And should we give an appropriate error message (perhaps switching to the main branch) in case the current was deleted?

The error message sounds like a good idea, but I wouldn't automatically change branches.

[leif]

OT: Does the caret (interpreted as mark-up) in Jeroen's reply to my code example (right before "Why?") also span multiple comments(!) for others?

For me, all text (including the headlines) is typeset in superscript mode from there on.

I'll try to close the tag here:

[leif]

QED