ModelicaSystem.buildModel() does not show build warnings

[bilderbuchi]

If you use the ModelicaSystem API, when buildModel() is run, content (e.g. build warnings) returned by GetErrorString gets swallowed if no compilation error occurs.
Probably this block in current master should get an elif clause that prints the error string if it's not empty.

[arun3688]

@bilderbuchi what kind of warnings is currently not displayed, can you give an example, because the buildModel() API outputs two results

successful build

>>> omc.sendExpression("buildModel(Test)")
('c:/juliatest/Test.jl/extendsbug/Test', 'Test_init.xml')

non successfull build

>>> omc.sendExpression("buildModel(Test1)")
('', '')

It should always return empty string in case of non successful builds and then we call "getErrorString()" and display the error message, Do you have some example where we miss such warnings?

[bilderbuchi]

Sure! For example, the model

model test_OMPython_145
  Real x(each start = 1.0); 
equation
  x = 1;
end test_OMPython_145;

on simulation in OMEdit outputs

[1] 19:24:23 Translation Warning
[test_OMPython_145: 2:15-2:26]: 'each' used when modifying non-array element x.

When using ModelicaSystem, I don't get any output during the build stage (when the model is instantiated). What's worse, the error string is queried, but then discarded, so I can't even get at the warning by manually querying for the error:

mod=ModelicaSystem("test_OMPython_145.mo","test_OMPython_145")
print('after building it')
print(mod.requestApi('getErrorString'))
print('done')

output:

after building it

done

Note that warnings don't fail the build, that's why they're warnings, not errors. A successful build with a useful error string is a scenario that the current code does not anticipate, hence my above suggestion.

[sjoelund]

The errors should not be printed; then you won't be able to process them. Better would be to not discard them or return them (or throw an exception!). Probably getErrorString before the API would be nicer to clear older warnings.

[bilderbuchi]

I agree. However OMPython is already full of prints for error reporting, so I thought it preferable to align with the present style, instead of doing it "better", but differently.

I think improving the situation towards more python-idiomatic code would be a separate issue.

[sjoelund]

I agree. However OMPython is already full of prints for error reporting, so I thought it preferable to align with the present style, instead of doing it "better", but differently.

I think improving the situation towards more python-idiomatic code would be a separate issue.

Well... I never use those API functions (only sendExpression) because of these problems. The API is not really possible to use :(

[bilderbuchi]

I think the ModelicaSystem API is useful (even if it could use some improvements) because it offers a ~pythonic interface that hides (needless) scripting details (like separately needing a getErrorString after ~ every command) from the user.

One thing I could not find out how to do using sendExpression alone is to (re-)simulate, optionally with overrides, without incurring a build stage that seems to always happen when using simulate() -- indeed that seems corroborated by the docstring "simulates a modelica model by generating c code, build it and run the simulation executable".

[sjoelund]

For that, I use buildModel followed by the subprocess module to start the simulation since there is no resimulate API

[bilderbuchi]

there is no resimulate API

good to know, tack saa mycket!