[TheNetherWatcher]

Fixes #617.

ann2rr's start_time and stop_time were compared with the length of each interval, not its position in the record. They now select intervals by when they end in the record, matching the WFDB ann2rr program's -f and -t options:

• annotations from start_time to stop_time (both inclusive, as in ann2rr) are used
• the first interval is measured from start_time (from 0 when it isn't given, as before)

With record 100 (360 Hz):

Call	Before	After
start_time=0.9	30 intervals, all longer than 0.9 s	2272 intervals, first one 46 samples (0.9 s → first beat at 370)
stop_time=0.5	2 intervals, both shorter than 0.5 s	1 interval, the beat at sample 77

A parametrized test covers start only, stop only, both, a beat exactly at the start time (a zero-length first interval, as in ann2rr) and a range with no beats. Four of the cases fail without the change.

[tompollard]

Thanks, this looks great! I compared it with the WFDB ann2rr program (-f and -t, with and without -A) on records 100, 1003, 201, 203 and 232, and the intervals match exactly for every normal time range.

Two optional suggestions:

1. With start_time after stop_time, the result is silently empty (the C program swaps them). Raising a ValueError might be clearer:

   if start_time is not None and stop_time is not None and start_time > stop_time:
       raise ValueError("start_time must not be after stop_time")

2. round() rounds halves to even, while the WFDB library rounds them up ((long)(x*f + 0.5)), so a time exactly half a sample in can give a different sample (for example start_time=0.002 at 250 Hz gives 0 here and 1 in ann2rr). int(start_time * ann.fs + 0.5) would match exactly. This is a very small edge case, so feel free to leave it.

(stop_time=0 gives an empty result here, while -t 0 means "no limit" in the C program. I think the behaviour here is the right one.)

[TheNetherWatcher]

Thanks for checking against the C program! I took both suggestions:

1. start_time after stop_time now raises a ValueError, checked before the annotation file is read.
2. Times now round half a sample up, as in the WFDB library. A new test case uses start_time=0.125 at 100 Hz, which gives sample 13 rather than 12.

I agree on stop_time=0, so I left it as it is. I also merged main and moved the changelog entry to changelog.d/ann2rr-time-range.rst.

[tompollard]

PhysioNet is having a moment, which I assume is why tests are failing. Sorting it out now...

[tompollard]

Thanks, this looks great! I re-ran the comparison with ann2rr and everything still matches.