Generation strategy is not associated with correct experiment when using SQL storage after JSON deserialization

[alxfed]

In particular: If you archive or store a generation strategy as json, then ingest it back and transform into an object with the help of generation_strategy_from_json() function the attached _experiment has db_id = None. Then, if you try to save it to the SQA store with the help of save_generation_strategy() (after saving the corresponding experiment first of course) the _save_generation_strategy() function is allegedly checking whether the experiment exists and already is in the database (lines 115 - 126 in save.py module), but that's not what happens. It verifies whether the _experiment attached to the recovered generation strategy object has a db_id (which is always None there in this scenario). Of course the workaround is to read the db_id of the experiment from the database, then assign this db_id directly by gs.experiment.db_id = the_existing_db_id then save the modified generation strategy object... but I have a feeling that it would be better if you did it yourself starting from the line 115 and on of the sqa_store/save.py module.

[lena-kashtelyan]

We'll investigate this and get back to you @alxfed! In the meantime, this workaround should help:

recovered_exp = experiment_from_json(...)
recovered_gs = generation_strategy_from_json(...)
recovered_gs._experiment = recovered_exp
save_generation_strategy(recovered_gs)

Let me know if that doesn't fix!

[alxfed]

As I said: I first save the recovered experiment, then take the db_id from it (this saved experiment) and 'by hand' assign it recovered_gs.experiment.db_id = saved_recovered_experiment.db_id , then save the resulting gs. And it works fine. I'm just worrying about the unsuspecting newcomers who will read the docstring and nothing else. Sorry for bothering you with this.

[lena-kashtelyan]

@alxfed, yep that makes sense, was just suggesting a workaround that doesn't manually mess with the db_id attribute : )

[lena-kashtelyan]

This is fixed now, provided generation_strategy_from_json is used correctly (with the restored experiment object passed to it as input.