Double scalars not handled by Octave

[blink1073]

Simple test case yields 1 instead of 1.5:

o = Octave()
o.start()
o.run_code('a=1.5')
o.get_variable('a')

This appears to be a limitation when calling javaObject('org.json.JSONObject').toString() in Octave. I am using 3.8.1 in Ubuntu 14.04. I traced json_dump and calling toString() on the javaObject('java.lang.Double', value) object works as expected, but not when it is rolled into the full JSON object.

[blink1073]

This causes 6 test failures when USE_OCTAVE is True.

[isbadawi]

Yes, I ran into this earlier when trying to make the Octave tests run on travis -- it works fine on my Mac laptop, but not in my Ubuntu VM. I asked a question on stackoverflow but no replies. Not sure what's going on!

[blink1073]

As you pointed out, it works if we use the raw value. Is there a case where doing so is a bad idea?

[blink1073]

The test suite passes with both Matlab and Octave if I use raw doubles here.

[isbadawi]

IIRC, I tried making that change before, and ran into errors where other parts of the code expected an object. Can you try with more complicated data structures, e.g. structs with double fields, etc?

[blink1073]

In [2]: o.run_code('a.b=4332.22')
Out[2]: 
{'content': {'code': 'a.b=4332.22',
  'datadir': '/tmp/MatlabData/',
  'stdout': 'a =\n\n  scalar structure containing the fields:\n\n    b =  4332.2\n\n',
  'figures': []},
 'success': 'true'}

In [3]: o.get_variable('a')
Out[3]: {'b': 4332.22}

[blink1073]

In [5]: m.run_code('a.b=4332.22')
Out[5]: 
{'content': {'code': 'a.b=4332.22',
  'datadir': '/tmp/MatlabData/',
  'stdout': 'a =\n\n  scalar structure containing the fields:\n\n    b =  4332.2\n\n',
  'figures': []},
 'success': 'true'}

In [6]: m.get_variable('a')
Out[6]: {'b': 4332.22}

[blink1073]

In [9]: m.run_code('a.b=4332.22;a.c="apple"')
Out[9]: 
{'content': {'code': 'a.b=4332.22;a.c="apple"',
  'datadir': '/tmp/MatlabData/',
  'stdout': 'a =\n\n  scalar structure containing the fields:\n\n    b =  4332.2\n    c = apple\n\n',
  'figures': []},
 'success': 'true'}

In [10]: m.get_variable('a')
Out[10]: {'b': 4332.22, 'c': 'apple'}

In [11]: m.run_code('a = {4332.22,"apple"}')
Out[11]: 
{'content': {'code': 'a = {4332.22,"apple"}',
  'datadir': '/tmp/MatlabData/',
  'stdout': 'a = \n{\n  [1,1] =  4332.2\n  [1,2] = apple\n}\n',
  'figures': []},
 'success': 'true'}

In [12]: m.get_variable('a')
Out[12]: [4332.22, 'apple']

[isbadawi]

Okay, here's the error I ran into, but it was while playing with json_dump at the octave prompt. Without the change:

+octave:1> javaaddpath('java/json.jar')
+octave:2> json_dump(5)
ans = 5.0

With the change (replacing the line you mention with obj = value;):

+octave:1> javaaddpath('java/json.jar')
+octave:2> json_dump(5)
error: scalar cannot be indexed with .
error: evaluating argument list element number 1
error: called from:
error:   /Users/isbadawi/code/py/python-matlab-bridge/pymatbridge/matlab/util/json_v0.2.2/json/json_dump.m at line 70, column 11

Line 70 tries to do an obj.toString(). Maybe we just never trigger the error case from our code?

[blink1073]

Yeah, I don't think we'd ever get into that case because we are always returning structures.