Process: raise when exposed_outputs gets non-existing namespace

[sphuber]

Fixes #5261

Up till now, a call to exposed_outputs providing a namespace as
argument that doesn't exist, i.e., no outputs were ever exposed in that
namespace, would simply be a no-op. Given that this would often be
because the namespace was accidentally misspelled, the error would go
unnoticed and the process would finish without error message.

To prevent this situation that is difficult to debug, the method will
now raise a KeyError if the top-level namespace does not exist.

[codecov]

Codecov Report

Merging #5265 (24fc5c0) into develop (d9f8c85) will increase coverage by 0.01%.
The diff coverage is 100.00%.

@@             Coverage Diff             @@
##           develop    #5265      +/-   ##
===========================================
+ Coverage    82.02%   82.02%   +0.01%     
===========================================
  Files          533      533              
  Lines        38255    38257       +2     
===========================================
+ Hits         31373    31376       +3     
+ Misses        6882     6881       -1     

Flag	Coverage Δ
django	77.07% <100.00%> (+0.01%)	⬆️
sqlalchemy	76.36% <100.00%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
aiida/engine/processes/process.py	91.77% <100.00%> (+0.04%)	⬆️
aiida/transports/plugins/local.py	81.71% <0.00%> (+0.26%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d9f8c85...24fc5c0. Read the comment docs.

[sphuber]

@janssenhenning I think this should fix the problem as now it will throw an exception. Let us know what you think of that solution.

[janssenhenning]

@sphuber Thank you for the quick fix. Judging from the test this is the behaviour I would like 👍.
I have one question: Does this also prevent the case, where the namespace argument is missing from the exposed_outputs call but the output was exposed with a namespace? If not, would it also be possible to catch this case?

[sphuber]

Does this also prevent the case, where the namespace argument is missing from the exposed_outputs call but the output was exposed with a namespace? If not, would it also be possible to catch this case?

No it does not. This would also be more tricky to do. For example, imagine a process that exposes a process without namespace and a process with a namespace. If the parent process then calls exposed_outputs without namespace on accident, there is no way to know for the code whether this was actually an accident or intended. I would say that this case is therefore not really possible to solve in a robust and general way.

[janssenhenning]

Okay I see. The current solution seems fine to me

[sphuber]

@ramirezfranciscof could you give this a look? It is not a big one and would be good to get merged for v2.0

[sphuber]

@chrisjsewell @ramirezfranciscof let me know if you want to still give this a look. Otherwise I am merging tomorrow