IPython 0.13: merge user configuration with Sage configuration

[jhpalmieri]

After #12719, user configuration files (in .sage/ipython-0.12/profile_default/ and .../profile_sage/) seem to be effectively ignored: I think they are read after the Sage terminal app has already been initialized, at which point it is too late.

The attached patch remedies this, reading the configuration from .sage/ipython-0.12/profile_default/ipython_config.py during the initialization. Question: should it read from this directory or from profile_sage?

Apply trac_14188_vb.patch​ and trac_14188_docfix.patch​

[vbraun]

It should read from /profile_sage/ipython_config.py for sure.

[jhpalmieri]

Here's a new version which reads from profile_sage.

[vbraun]

Reworked patch

[vbraun]

I changed your code to override load_config_file() instead of stuffing it in the ctor. Now the configuration is only loaded once, as one can easily check by putting a print statement into ipython_config.py.

Also, after discussion with Jason and William we decided to default to colors in the terminal. For simplicity I just added this into this patch.

[vbraun]

And I agree with your changes, of course. Feel free to set it to positive review if you are happy with it.

[klee]

Apply trac_14188_vb.patch

[vbraun]

And IPython is of course not smart enough to turn off colors when the output is a pipe, thanks a bunch. The second patch is the one-line fix to do it by hand.

[jdemeyer]

Second patch needs review.

[jhpalmieri]

If someone customizes their IPython configuration, this can cause doctests to fail (for example, if you turn on an output prompt like "Out[1]:", interfaces/sage0.py fails). I think this is okay, but I'm going to post at #12415 about it.

[jhpalmieri]

All tests passed.

[jdemeyer]

I see doctest failures on OS X 10.6: ​http://build.sagemath.org/sage/builders/UW%20bsd%20%28OSX%2010.6%20x86_64%29/builds/214/steps/shell_8/logs/stdio

[jdemeyer]

Same problem on lena (Fedora 16 x86_64): ​http://build.sagemath.org/sage/builders/Skynet%20lena%20%28Fedora%2016%20x86_64%29/builds/113/steps/shell_7/logs/stdio

[jhpalmieri]

I see this on bsd.math.washington.edu (OS X 10.6), but I have no problems on lena.

[jhpalmieri]

A typical failure on OS X 10.6:

    sage: subsage = Sage()
    sage: s = ZZ(subsage('initial_seed()'))
    <boom>

It looks like when running Sage as a subprocess on this platform, sys.stdout.isatty() returns True, even while doctesting. How else can we detect this situation?

[vbraun]

Pexpect runs in a pty, so on any platform isatty() will return true. This is why we explicitly disable colors when starting a sage-in-sage session. My current theory is that the pty gets messed up somehow by the ansi codes since we disable colors by sending %colors NoColor, so at least one colored prompt will be displayed. I'm currently trying if disabling color via command line switch works...

[jdemeyer]

Replying to jhpalmieri:

How else can we detect this situation?

pexpect children have the environment variable TERM unset (#12221, similar: #12263). We did this precisely to solve problems like this where the terminal did strange things.

sage: "TERM" in os.environ
True
sage: sage0('"TERM" in os.environ')
False

[jhpalmieri]

A change like this fixes the failures on bsd.math.washington.edu. I haven't run a full test suite yet.

sage/misc/interpreter.py

@@ -660 +660 @@
         verbose_crash = True),
     TerminalInteractiveShell = Config(
         ast_node_interactivity = 'all',
-        colors = 'LightBG' if sys.stdout.isatty() else 'NoColor',
+        colors = 'LightBG' if (sys.stdout.isatty() and 'TERM' in os.environ) else 'NoColor',
         confirm_exit = False,
         separate_in = ''),
     # The extension is *always* loaded for SageTerminalApp

[vbraun]

Improved patch

[vbraun]

I'd rather always turn off colors in the sage<->sage interface than rely on the TERM variable. The updated patch fixes sage0 on OSX (bsd.math) for me.

[jhpalmieri]

This passes tests for me on bsd.