Structured load api

[pp-mo]

Basic provision of a public API for "fast" (i.e. structured) pp + ff loading.

This works via a context manager that enables fast-load for all the iris load functions.

At this point there is no public low-level interface to this functionality, equivalent to the pp or grib as_pairs functions. It would be fairly easy to add one, but it's not a priority.

[pp-mo]

@marqh @bjlittle

You might want to comment on the approach here.

I think you can already see that you get a lot “for free” (or nearly so) doing it this way.
I reckon the key ‘similarities to standard load’ which this should deliver are :
• multiple sourcefiles
• filename and uri expansion
• constraints
• callbacks
• post-load merging (especially: building from scalar coords that vary across files, e.g. realization)

I’m planning tests for all the above.
I’m also building a list of ‘known problem cases’, which I may also code as tests.

So, quite neat, but not a lot to see here.
Docs are only stubs so far.
We could easily turn the ‘switch’ into a context manager or load keyword if wanted.

The PR doesn’t show it, but I’ve already confirmed that this delivers identical results for a lot of cases
(and different but OK answers in others).

[pp-mo]

NOTE: #29 contains a neater solution to the 'load_raw' problem

[marqh]

I am happy with the approach here; I suggest you keep this branch open and merge things onto this as we work through the details

then this branch can form the PR to the Scitools/iris master

[marqh]

I like the implementation as a context manager, I think this is a neat way to provide the functionality for people to opt into

@bjlittle do you agree that htis is a useful way to present this to users?

[pp-mo]

Status change: As everything else was already agreed + merged to here, I've merged the outstanding #31 here, as that is all that remains to discuss.
So I closed this and opened a new PR to the main Iris master instead : SciTools#2241
We can discuss what remains to be discussed there.