Error executing code in Mathematica

[yzh]

I got TypeError when running the same code repeatedly on OSX 10.10.4. Each time it stops at a different value of i and returns a different error message.

m = Mathematica(logfile='mathematica-log')
for i in range(30000):
    print i
    condstr = '{x + 5*y - 1<0, x - y<0, 0<-2*x - 5*y + 1<1/100}'
    varstr = '{x, y}'
    pt = m.FindInstance(condstr, varstr)
    pt_val = (pt[1][1][2], pt[1][2][2])

Usually the error appears before i=2000, and the error message is most often of the type

TypeError: Error executing code in Mathematica
CODE:
	sage3=sage0[[2]];
Mathematica ERROR:
	sage3=sage0[[2]];

Is it caused by the same issue as in #20128? Any workaround?

I'm also curious about the following piece of code in Mathematica::set (see Line 514 of mathematica.py)

if len(out) > 8:
    raise TypeError("Error executing code in Mathematica\nCODE:\n\t%s\nMathematica ERROR:\n\t%s"%(cmd, out))

What does the constant 8 mean here? Note that the error persists when I augment the number 8 to 4000.

[mkoeppe]

I was able to reproduce this error (same Mathematica version 10.3.1.0, also on OS X.)

This error is caused by sporadic double echoing, which confuses the pexpect code. This can be seen at the end of the logfile generated by the above testcase. I'm not sure where the double echoing comes from. Perhaps some strangeness of Mac OS ptys, perhaps a race condition, perhaps math-readline is doing something wrong, perhaps a Mathematica bug.

A workaround is to disable echoing:

diff --git a/src/bin/math-readline b/src/bin/math-readline
index 5a13906..72e3795 100755
--- a/src/bin/math-readline
+++ b/src/bin/math-readline
@@ -9,6 +9,9 @@ import sys, signal, subprocess
 import readline
 from six.moves import input
 
+import os
+os.system("stty -echo")
+
 def child_exited(*args):
     global child
     status = child.poll()

[dimpase]

I don't have access to Mathematica on machines I have Sage on, so I can't test this easily. If this echo disabling does not affect the UI in any way, why don't we put this workaround in Sage?

(To me stty -echo is what one does to hide what's entered on the keyboard, but here it's not relevant?)

[mkoeppe]

Replying to dimpase:

I don't have access to Mathematica on machines I have Sage on, so I can't test this easily. If this echo disabling does not affect the UI in any way, why don't we put this workaround in Sage?

(To me stty -echo is what one does to hide what's entered on the keyboard, but here it's not relevant?)

This stty -echo affects the pseudo-terminal that Sage uses to communicate with Mathematica. It does not affect the Sage UI.

I have not tested whether this workaround has any negative effect on running Sage via SSH on a remote host.

[mkoeppe]

New commits:

​0fdafe3

Disable echo on Mathematica pty to avoid sporadic errors on Mac OS X

[vbraun]

Expect has a terminal_echo=False constructor argument, this is presumably equivalent and the right way to do it. See e.g. sage.interfaces.singular.Singular.__init__

The terminal echo has bitten us before, really all expect interfaces should run with terminal_echo=False as the echo is inherently and unfixably racy. But various interfaces have been built around the assumption of terminal echo and are hard to fix.

[mkoeppe]

Replying to vbraun:

Expect has a terminal_echo=False constructor argument, this is presumably equivalent and the right way to do it. See e.g. sage.interfaces.singular.Singular.__init__

That's what I tried first, but it didn't fix the problem.

[dimpase]

I just checked this with Mathematica 9.0 on Linux 64-bit, the test works.

I noticed that most doctests fail: I run

sage -btp --optional=database_gap,gap_packages,python2,sage,mathematica src/sage/interfaces/mathematica.py 

and get a lot of formatting errors - the output is TeX-like rather than text-like, e.g.

File "src/sage/interfaces/mathematica.py", line 322, in sage.interfaces.mathematica
Failed example:
    mlist = mathematica(slist); mlist     # optional - mathematica
Expected:
    {{1, 2}, 3., 4 + I}
Got:
    \{\{1,2\},3.,4+i\}
**********************************************************************
File "src/sage/interfaces/mathematica.py", line 326, in sage.interfaces.mathematica
Failed example:
    slist2[0]                             # optional - mathematica
Expected:
    {1, 2}
Got:
    \{1,2\}
**********************************************************************
File "src/sage/interfaces/mathematica.py", line 335, in sage.interfaces.mathematica
Failed example:
    mathematica('10.^80')     # optional - mathematica
Expected:
    1.*^80
Got:
    1.\times 10^{80}
**********************************************************************
File "src/sage/interfaces/mathematica.py", line 576, in sage.interfaces.mathematica.Mathematica.chdir
Failed example:
    mathematica('Directory[]')      # optional - mathematica
Expected:
    "/"
Got:
    /

should these get fixed?

[mkoeppe]

It is quite likely that this is specific to both the Mathematica version and the OS.

For example, the beginning of math-readline script says:

# Cleverly run Mathematica with the benefit of readline, which
# is something the usual commercial mathematica doesn't provide!

But at least Mathematica 10.3.1 does provide command-line editing (though not the full range of readline's keybindings).

[dimpase]

I think a lot of tests are outdated; e.g. the precision in N() operator should be what Mathematica does now, that is the total number of digits rather than the number of digits after '.'

I do get command-line editing on my terminal version, too.

[dimpase]

I wonder if there is a magic switch to switch to the form of output used in tests; don't know what it is exactly -- FortranForm?

[mkoeppe]

I think the failing Mathematica doctests should go on another ticket. This ticket needs review...

[dimpase]

looks good to me

[slabbe]

Replying to mkoeppe:

I think the failing Mathematica doctests should go on another ticket.

See #18888.

One problem is that the version of Mathematica may change from one user to the other...

[dimpase]

Replying to slabbe:

Replying to mkoeppe:

I think the failing Mathematica doctests should go on another ticket.

See #18888.

thanks. I completely forgot about it :-)

One problem is that the version of Mathematica may change from one user to the other...

no, if you read info on #18888 you will see it has not much to do with versions, it is a change in behaviour.

[dimpase]

Replying to dimpase:

Replying to slabbe:

Replying to mkoeppe:

I think the failing Mathematica doctests should go on another ticket.

See #18888.

thanks. I completely forgot about it :-)

One problem is that the version of Mathematica may change from one user to the other...

no, if you read info on #18888 you will see it has not much to do with versions, it is a change in behaviour.

caused by Volker's commit 7747261d ($ git show 7747261d will show it). I suspect the reason was to have a better output in a notebook (sagenb or jupyther or whatever).

I have no clue how to fix this.

[dimpase]

#18888 is now updated and ready for review.

[jdemeyer]

Replying to mkoeppe:

This stty -echo affects the pseudo-terminal that Sage uses to communicate with Mathematica. It does not affect the Sage UI.

I disagree with this statement, in fact it should be precisely the opposite: I don't think that the pexpect interface needs readline, so why would it run math-readline instead of the plain Mathematica (i.e. math) directly?

This is at least partially responsible for the breakage reported on #23112.

[mkoeppe]

Replying to jdemeyer:

Replying to mkoeppe:

This stty -echo affects the pseudo-terminal that Sage uses to communicate with Mathematica. It does not affect the Sage UI.

I disagree with this statement, in fact it should be precisely the opposite: I don't think that the pexpect interface needs readline, so why would it run math-readline instead of the plain Mathematica (i.e. math) directly?

This is at least partially responsible for the breakage reported on #23112.

The statement that you quoted is correct as stated.

Rather, what went wrong on this ticket is: It was overlooked that this script math-readline is intended to be a user script, rather than just one that is used internally by the pexpect interface. (The script is not mentioned in a manual.)

Also, note that current versions of mathematica (like "Mathematica 11.0.1 for Mac OS X x86") do provide a readline-like interface, so math-readline is not needed; so certainly the description of math-readline is outdated.

We can fix the script on #23112 (for the benefit of users of ancient versions of mathematica??), but please not by breaking the mathematica pexpect interface.

[mkoeppe]

Upon further investigation, this ticket did break echoing in mathematica_console(). So something needs fixing (in #23112).

[mkoeppe]

A working fix is now available on #23112.