Add date expressions for first and last day of NHS financial year

[rebkwok]

Fixes #719

[iaindillingham]

When you tested against the EMIS backend, you tested the expressions with the context of population and event_date; the former because it's special and the latter because it's not, I assume. Why not do the same here?

I'm sure there's a good reason 🙂

[rebkwok]

I did the tpp tests first, and it made sense to just test against the patient ids that were returned. population on its own just gets you a set of patient_id in the results. I tried to do the same thing in the emis tests, but the emis backend generates patient ids (I confess I didn't look too closely at exactly what it does), so the patient ids in the results were always 0 and 1. The emis one is only that way to have some way to checking that the 2 patients it returns are the expected ones.

[iaindillingham]

Ah, of course! Yes, the EMIS backend handles patient IDs differently, making is much harder to test against just population. More here:

cohort-extractor/cohortextractor/emis_backend.py

Lines 1585 to 1596 in 84dddec

	def truncate_patient_id(patient_id):
	"""
	EMIS patient IDs are 128 byte strings which significantly bloat our output
	files. However, because they are actually hex-encoded SHA-512 hashes we can
	get away with throwing away all but the first 63 bits and storing them as
	int64s (63 bits so we can safely store them as signed int64s without
	worrying about negative patient IDs, which might upset expectations
	elsewhere). 63 bits still gives us plenty of entropy as sqrt(2^63) is about
	2 billion so well above the threshold where we'd need to worry about
	collisions with a population of a few tens of millions.
	"""
	return int(patient_id[:16], 16) >> 1

[iaindillingham]

Very clear! I feel I've learnt a lot more about cohort-extractor, just from looking at this PR.