Decoy v3

[mcous]

Starting a discussion for the next major version of Decoy! Here's what's in my head so far

Very likely

• No more rehearsal syntax
  • It complicates the architecture
  • It's primary intent was to preserve type-safety and intellisense, and ParamSpec solves this use case
  • Without rehearsals, Decoy is able to distinguish between calls and stub configurations, unblocking richer warnings and debugability
• Promote various warnings to errors as already called out in the docs
• Bump minimum supported version of Python

Ideas / maybe

• debug helper like vitest-when
• Drop matchers in favor of hamcrest or other pre-existing library
• Drop global state
  • Might make call order verification difficult, and I think call order verification is valuable
• Drop custom mock object in favor of MagicMock?
  • Just a thought, not sure this is feasible or a good idea

[SyntaxColoring]

• Promote various warnings to errors as already called out in the docs

Early heads-up that we have hundreds of MissingSpecAttributeWarnings that seem spurious. I haven't had a chance to look into them yet. If you have release timing in mind, I can try to investigate before then, in case it's a Decoy bug. (It might not be—it wouldn't surprise me if we're just speccing a bunch of mocks wrongly.)

• Drop matchers in favor of hamcrest or other pre-existing library

This seems nice if the 3rd-party library is up to the task. Matchers seem like something that takes a surprising amount of care to handle all the edge cases.

Relatedly, I've been mixing unittest.mock.sentinel with Decoy and I haven't had any problems.

[mcous]

• Promote various warnings to errors as already called out in the docs
  Early heads-up that we have hundreds of MissingSpecAttributeWarnings that seem spurious.

😱. Let me know if you'd like any help looking into this. I peeped y'all's CI logs and, while my memory of the interfaces is not what it used to be, I wouldn't have expected to see those warnings

If you have release timing in mind...

I have no release timing in mind whatsoever. Opentrons is likely the only real use of this library, so I'd only pursue working on it if it was helpful for y'all. Otherwise I'm not currently using Decoy, so I don't have any real pressure for a release.

Relatedly, I've been mixing unittest.mock.sentinel with Decoy and I haven't had any problems.

Nice, yeah chalk that up as another point towards "what can Decoy re-use?"