Is this a new option? Is it set anywhere?

It is a new option, and no it's not set by default anywhere. The expectation is it'll be set in make option or board profile if someone wants it... or if it's a desieable enough feature it could be made the default

Hmm, doesn't that make it almost impossible to discover? Someone has to go look at the code to see "hey, look at that! I can enable MICROPY_PY_UJSON_ORDERED_DICT!". Just wondering...

Yeah it's certainly not ideal... At the very least I should enable it on the coverage build and add a unit test - this initial PR was thrown up very quickly in the middle of a project rebase/test/release cycle on my end and I didn't want it to be forgotten.

Just put in in mpconfig.h, like all other options.

Just to check, this assumes the CPython used to run the test will also preserve dict order. Do all CPython versions do this, and if not: from which version on? I.e. we whould be sure this always works.

Ah yes, that is the assumption yes. This has been the case since Python 3.6, though this aspect was written in as official in 3.7
https://docs.python.org/3.6/whatsnew/3.6.html#whatsnew36-compactdict

Arguably we should look at bringing in this same dict implementation as an option in micropython, it was shown to be significantly more efficient than the previous cpython dict and would likely be better than our current ordered dict implementation at least (certainly for very large dicts)

... and yes this is why the new unit test is failing, cpython 3.5 is used on travis :-(

Perhaps rather than making this a #define feature I should have just added the object_pairs_hook argument like in cpython:
https://docs.python.org/3.4/library/json.html#json.load

It would be more work to add this level of generic support rather than just added an ordered flag however

It would be more work to add this level of generic support rather than just added an ordered flag however

Would also be larger in code size.

Perhaps change the check so that when coverage is enabled, it just checks whether the returned dict is the same as sorted(dict).

Checking loaded dict == sorted dict results the same as the comparison I've already got in place - it works on micropython with my patch and cpython 3.6+ but does not work on cpython3.5. I can put an extra try block in there to loads with object_pairs_hook=OrderedDict for cpython though.

Ah but the idea is that CPython never runs any code for which it's behavior changes between versions, i.e. code guarded by is_coverage only gets executed by MicroPython, not CPython, and the else clause is where the expected result gets printed. Pseudocode:

if is_coverage:
  print(is_dict_with_same_content_and_sorted(a,b))
else:
  print(True)

Ah I didn't realise that's how it would work, that's brilliant, thanks. I've updated as such.

Same comment as for other commit (plus this isn't about locals here)