[BUG] A reconstruct() call alters the sys.stdout

[patquem]

Hello,

I am working on a pysap test with unitest.
But my unitest can't be correctly executed due to the following AttributeError:

File "/home/300.86-METRO_CARAC/PYTHON/miniconda3/lib/python3.7/unittest/case.py", line 647, in run
result.stopTest(self)
File "/home/300.86-METRO_CARAC/PYTHON/pycharm-community-2021.1.2/plugins/python-ce/helpers/pycharm/teamcity/unittestpy.py", line 263, in stopTest
output = sys.stdout.getvalue()
AttributeError: '_io.TextIOWrapper' object has no attribute 'getvalue'

This error is related to a mri.reconstructors.base.ReconstructorBase.reconstruct() function call.

When this function is not called, the sys.stdout is :

and the unitest is correctly executed.

But when this function is called, the sys.stdout becomes :

and the unitest goes wrong (AttributeError).

Could you please investigate and fix this ?
(this could be related to the buffer or a file not correctly closed when passing in the reconstruct() function ?)

Note : The following workaround which consists in :

import sys
from importlib import reload
reload(sys) 

after calling the function didn't work for me.

Thank you,
Patrick

[chaithyagr]

Thank you for pointing this issue. However I am afraid this isn't related to pysap-mri.
If I am right, you are using pycharm and this seems to be a known issue in running tests on pycharm and I have indeed faced this in past.

Here is some more details: https://intellij-support.jetbrains.com/hc/en-us/community/posts/360000019990-Unittest-Error-AttributeError-file-object-has-no-attribute-getvalue-
I would recommend to switch to pytest for testing within pycharm as I don't think I ever faced this issue there.

Can you please run the test from terminal and not use pycharm to ensure that pysap-mri isn't causing it? From what I see, the same tests pass at CI, hence I am closing this issue here. Feel free to ping here if you still have the issues.

[chaithyagr]

Also, which version of pycharm are you using?
I would recommend updating it as these parts of pycharm have been unstable and constantly undergo changes

[patquem]

Hello @chaithyagr ,

The problem is not related to PyCharm.
When I execute my unitest in console mode (out of PyCharm) I am faced to the same problem.

File "init.py", line 72, in run_tests
xmlrunner.XMLTestRunner(output=rslt_path, verbosity=3).run(test_suite)
File "/home/300.86-METRO_CARAC/PYTHON/envs/env_aar114/lib/python3.7/site-packages/xmlrunner/xmlrunner.py", line 468, in run
self._restore_standard_output()
File "/home/300.86-METRO_CARAC/PYTHON/envs/env_aar114/lib/python3.7/site-packages/xmlrunner/xmlrunner.py", line 397, in _restore_standard_output
sys.stdout = sys.stdout.delegate
AttributeError: 'io.TextIOWrapper' object has no attribute 'delegate'

the reconstruct() function introduces something wrong for the sys.stdout interpretation by xmlrunner afterwards.

Patrick

[chaithyagr]

Reopening this here, will have a look to be sure

[chaithyagr]

Can you please post how did you run the test, also which particular test is failing. The stacktrace given above indicates the issue is in xmlrunner more than pysap-mri. Further, I could run all the tests locally and also we have successful test at travis CI here

[patquem]

I am sorry, I can't do it (reasonably).
The test is not a one from the pysap release but a test that belongs to our test-suite and based on specific developments we realise here in CEA-Grenoble.
We didn't modify mri.reconstruct(). We just observe that the sys.stdout is altered after passing in this function and makes our xmlrunner test-suite report down.
As previously said, this could be related to a File that is not correctly closed in mri.reconstruct() (or something wrong with the Buffer ?).

[chaithyagr]

I understand that. However, I am sorry but without a minimum reproducible example, I cant be able to reproduce and hence fix the bug, that too if any. I am quite confident the codes at pysap-mri is fine as it is constantly used by many users and also the codes are constantly tested with rigid CI pipelines.
In fact, the reconstruct function is a simple wrapper to the Modopt functions to optimize. (see here) . We dont have any buffers or file opening here.
Please make sure you have latest version of modopt.

If possible, please do try to setup a minimum reproducible example which isnt directly related to your internal codes. Or you can send in an email with the minimum reproducible codes if you are comfortable with it.

[patquem]

Hello,
We have finally found what was the cause of the disturbances. This was the same as for modopt I met a while.
The progress bar.
Could you please report the following changes to make the progress bar a 'user' argument ?
in the mri context, we have done in primal_dual.py :
line 30 : progress=True added in the condatvu arguments
line 181 : progress=progress added in the Condat arguments

Maybe it could be relevant to do the same for the other optimisers, if not :)
Thanks,
Patrick

[chaithyagr]

Sure, I can take some time and open a PR for this. While I still don't understand how this can cause that issue, it surely can be a good feature to add.

[chaithyagr]

Thank you for helping with this.