Closed nwieters closed 5 months ago
Need to do more testing since --contained-run is now/still breaking.
I reactivated the lines that take care of the use_venv
setting from the runscript.
Hi, I rerun the failing tests and they are now passing. Please feel free to review and also merge this into release.
Hi @mandresm , as I mentioned already, these changes let the setting in the runscript always win or give an error if runscript setting and argument are not campatible. You may want to change this behaviour. When I started to fix this bug I had a different approach. Maybe you can have a look at that again, if this is useful for you. It is in the commit a269640a0292127e4193467958c7511fc3b26606
Just so I understand this correctly: my current mental model of how this works is as follows:
${EXP_BASE}/scripts
and not /some/user/location/scripts/my_run.yaml
${EXP_BASE/scripts/my_run.yaml
will change what happens, but if and only if the options are not overridden by a command line flagIs that more or less correct? Or did I misunderstand the overwrite order?
Just so I understand this correctly: my current mental model of how this works is as follows:
* Command line arguments win, always. They are recycled run to run (e.g. the next run resubmits itself with the same flags as were given the first time) * User specifications in the runscript are next. Changing the submission script between runs (assuming it is stored in a different location) has no influence on the run, since it will recycle the on in `${EXP_BASE}/scripts` and not `/some/user/location/scripts/my_run.yaml` * Changing `${EXP_BASE/scripts/my_run.yaml` **will** change what happens, but if _and only if_ the options are not overridden by a command line flag * Internal (e.g. component/setup/etc) configuration files come last.
Is that more or less correct? Or did I misunderstand the overwrite order?
That is correct. Somehow we had it the other way around only for use_venv
, and I don't remember why, but to me this is nonsense (even if I implemented that approach myself) so I'm bringing things back to "normal"
All tests passing so... #bump #patch
Just so I understand this correctly: my current mental model of how this works is as follows:
* Command line arguments win, always. They are recycled run to run (e.g. the next run resubmits itself with the same flags as were given the first time) * User specifications in the runscript are next. Changing the submission script between runs (assuming it is stored in a different location) has no influence on the run, since it will recycle the on in `${EXP_BASE}/scripts` and not `/some/user/location/scripts/my_run.yaml` * Changing `${EXP_BASE/scripts/my_run.yaml` **will** change what happens, but if _and only if_ the options are not overridden by a command line flag * Internal (e.g. component/setup/etc) configuration files come last.
Is that more or less correct? Or did I misunderstand the overwrite order?
That is correct. Somehow we had it the other way around only for
use_venv
, and I don't remember why, but to me this is nonsense (even if I implemented that approach myself) so I'm bringing things back to "normal"
There was probably a good reason at the time, but I can't remember it either ;-) This is cleaner anyway. So, all green!
Closes #1150