metoffice / cmip6-dreq-interface Goto Github PK

djq fails sanity with any DREQ post 01.beta.32: what has broken?

There need to be two scripts to do this:

• a script to create a request which includes every experiment of every mip;
• a script which will take a single big response to such a request and spit it out into flat files per single-request

Currently there's just one big check tree for JSONifiers, which means that you can't write simple ones which don't pass the checks.

This should be fixed by making checkers be per-implementation

Currently you can only load a tag, but latest may not be the same as the trunk.

• Sort out with MO people if I can / should.
• Decide where it should live (my repo? probably not):
  • Rehost it if need be.
  • Sort out MO membership & groups
• License and copyright:
  • suitable copyright headers;
  • decide on a license;
  • add it to the top-level with suitable pointers.

This is everything that needs to be done so that other people could (perhaps) understand it.

• Document the sorting behaviour of JSONifiers (this is #2).
• Provide provenance for the implementations.
• Documentation (at least some) for command line tools.
• Python sample documentation
• Read it reasonably carefully.

Additionally any other cleanups that are needed.

They need to be updated and checked.

This is related to #3.

• Get someone to read it
• Elaborate this

Something is creating the wrong query?

The program makes quite a lot of use of things it calls 'fluids', which implement dynamically-scoped bindings. While they have all the advantages of dynamically-scoped bindings over globals, they also have all the disadvantages of them over lexically-scoped bindings. I need to understand which ones are needed and simplify / remove the ones that aren't. I suspect that the root / tag / path (new) fluids should go.

This is quite a big change.

Things that need to be done

• Clean up branches at MO: dqi-development needs to go or be recreated at least.
• Release notes
• Merge to master and push master to GH and hence home.
• Decide on a release tagging scheme: it it's not YYYMMDD then this needs to be justified.
• Tag it.
• Push.

Some of these mean that the release can't happen until I can fix things on the MO repo: so it will need to be Monday, not Friday.

Currently the output of a JSONifier gets sorted based on a 'label' in the dicts it returns. But alternative JSONifiers might not return suitable dicts. So one of the following should be done:

• it should be documented that they must do so;
• the thing should check and not sort if it can't find a label field;
• there should be a way of specifying what the key is and whether to sort.

I think one of the first or second options is probably best: the third would be a DTRT choice but would make the implementation stupidly hairy.

This was removed in 7461702 but I'd not noticed since the compiled file was still there.

This is why you must run make clean