openvax / topiary / 34869603181-1

92%

master: 92%

Pull #311

github

Narrow to PredictorSetupError; bare RuntimeError masked real bugs

A second review of this branch caught it contradicting its own
principle: it argued a narrow type is how you avoid hiding bugs, then
caught bare RuntimeError across the whole predict call graph.

NotImplementedError and RecursionError both subclass RuntimeError.
Reproduced: patching predict_epitopes_from_args to raise a bare
NotImplementedError printed the usage block, then "topiary: error: "
with nothing after it, exit 2, no traceback. An abstract method left
unimplemented reported as a clean user error, with an empty message,
because str(NotImplementedError()) is "". Dependencies' internal
invariant failures - varcode's lazy SpliceOutcomeSet, mhctools TULIP's
row-count check - would have been treated the same way.

mhcflurry_composite_version now raises PredictorSetupError, a
RuntimeError subclass, exported, for the four conditions where the
predictor is reachable but unusable and the fix is the caller's
(mhcflurry absent, no __version__, unreadable release, no active
release). The CLI catches that type instead of RuntimeError. Same shape
as CachedPredictorCoverageError for KeyError, which is what this branch
should have done the first time.

The handler also falls back to type(e).__name__ when str(e) is empty,
so no caught exception can render as a bare "topiary: error: ". The
changelog claim that str(e) "is correct for every type the handler
catches" was false for argument-less exceptions and is corrected.

Test fixes from the same review:

- CachedPredictorCoverageError.__str__ had no direct test. The existing
  KeyError tests match substrings of str(excinfo.value), which pass
  whether or not the message is quoted, so the only test that would
  have caught its removal was the fixture-dependent CLI one. A direct
  test now covers the single-arg and multi-arg forms; confirmed failing
  with __str__ removed.
- The setup-failure test hard-coded a copy of the message and raised it
  f... (continued)

11 of 13 new or added lines in 3 files covered. (84.62%)

7129 of 7736 relevant lines covered (92.15%)

0.92 hits per line