switching between the namelist options requires a re compile

[arezoorn]

When I added a new namelist configuration to the json files, I had to recompile. Compilation should be independent of the json files except from the compile json file, but I do not think they are. I usually recompile with every change in the json files to be cautious but sometime I forget and I get into trouble.

[rcabell]

You should be able to cd to an empty directory and re-run compose on the simulation object. That being said, a recompose option would be nice to have!

[arezoorn]

The experiment directory should always be non existing, otherwise it will complain that the directory exists. But I am not sure if you mean that by your comment.

[rcabell]

You can use `os.mkdir()` and/or `os.chdir()` to create or switch to an empty directory and then rerun compose. That should work, in theory. If it doesn't, then there should be a code change to support that.

…

On Sat, Nov 28, 2020 at 10:46 arezoorn ***@***.***> wrote: The experiment directory should always be non existing, otherwise it will complain that the directory exists. But I am not sure if you mean that by your comment. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#227 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHEYLTZJBUQ5WYB3UXHYVWDSSEZOFANCNFSM4UFRJCQQ> .

[jmccreight]

The heart of the matter is that the (run-time) model-side namelist goes into the model object and when it is restored, the model-side namelist from compile-time is restored.

I think the most elegant way to handle this is a Model.restore() method which takes an optional model-side namelist path and warns when it is not supplied.

In the mean time, one can manually edit this but it's not very elegant.

[jmccreight]

Simply recomposing wont solve the problem which is that a pickled model object has a previous model-side name list. Alternatively, pickling the model object could remove that namelist and then the namelist path i suggested previously would be required. There are pros and cons.

I'd have to look, but I'm not sure you can compose twice (without starting over). IE, I think that compose flags something in the simulation object like is_composed.

Composed is a bad term (joe and I argued over this), because the solution is to "recompose" the simulation object (start it over) and then compose to a new/fresh dir.