use '5 years ago' instead of absolute timestamp

[meejah]

Fix a test broken by the slow march of time

[codecov-io]

Codecov Report

❗ No coverage uploaded for pull request base (master@5e62118). Click here to learn what that means.
The diff coverage is n/a.

@@           Coverage Diff            @@
##             master    #453   +/-   ##
========================================
  Coverage          ?   90.1%           
========================================
  Files             ?     144           
  Lines             ?   27301           
  Branches          ?    3929           
========================================
  Hits              ?   24599           
  Misses            ?    1961           
  Partials          ?     741

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 5e62118...5f00842. Read the comment docs.

[exarkun]

Does this also work around the boundary conditions? In particular, I notice that secs_per_year is not always correct (since not all years have the same number of seconds). Will this cause the test to fail if run close enough to the year change-over? I guess any 5 year period will include either one or two leap years (intuitively; I haven't proven it). If it includes 1 leap year, then "5 years ago" is actually an additional 24 * 60 * 60 seconds ago. And if it includes 2 leap years, twice that addition.

So if this code runs on year X, Dec 31, 11PM then 5 * secs_per_year ago is around year X-4, Jan 1, 11PM (or even Jan 2 for ranges including two leap years). This is still pretty close to 5 years ago but I haven't checked to see what the code that actually computes the "N years ago" string does.

If it behaves correctly, maybe it's still worth a comment somewhere around here that it's okay. And if not,

now = datetime.now()
then = now.replace(year=now.year-5)
five_year_interval = (now - then).total_seconds()

gives the true number of seconds for that particular 5 year window.

[warner]

Can we also mock time.time() during the code that measures the interval, to make it deterministic?

BTW I filed https://tahoe-lafs.org/trac/tahoe-lafs/ticket/2901 before seeing this PR (I assumed it was already an end-of-year problem, but I guess it was a too-much-time-has-passed problem).

[meejah]

I don't know that we need to mock time -- IMO, the "human readable output code" should output "5 years ago" regardless of a few seconds (or even days) difference in the interval length -- that said I haven't looked at what it actually does ...

[meejah]

Okay, I used @exarkun's more-exact method of computing the number of seconds in 5 years. I don't think we should mock time().