feat: "Update output" mode

[augustebaum]

Long shot, but it would be pretty cool to be able to run xdoctest with e.g. --xdoctest-update which would overwrite the current expected output with the actual output of running the doctests, since doctests are pretty much snapshot tests.

[Erotemic]

At one point I started to implement this, but it got hairy fast. There is a hidden experimental features that can be enabled by setting the environ XDOCTEST_INSERT_SKIP_DIRECTIVE_ABOVE_FAILURES=1, which will attempt to disable failing tests, but it is poorly tested.

I'd be very open to a PR exploring a nice way to implement this. The main thing I worry about is when the current expected output (i.e. "want string") has pattern matching components to it. How do those port over to the new output? Do they? Is it a case-by-case basis?

Another issue is that xdoctest is designed to ignore checking output that you aren't explicitly testing for, so would all output be inserted as the "want string" by default? Would it need to be configurable? Would it rely on a lot of manual editing afterwards? There are a lot of design questions that need to be answered and justified, but still I'm very interested in this feature.

[augustebaum]

xdoctest is designed to ignore checking output that you aren't explicitly testing for, so would all output be inserted as the "want string" by default?

You're right, maybe there are 2 distinct features here:

• the one I'm imagining which is closer to "given that my doctests are already fully written, when a test fails, then write the new output to the file" and
• "write any output to the file"

I feel like the first one raises fewer questions.

[Erotemic]

I think it would be useful to have a feature where it generically overwrites any output that already exists (so a user can seed a doctest they want populated with ...). Then if we add a flag that lets the user restrict this to failing doctests that implements your feature. If you want to write the PR I'll review it and work with you on it.