In PR sourcegraph#177, I added normalization logic inside setTrackedFiles.

However, it seems like the other file tracking already must be
taking care of passing normalized paths as necessary (hopefully?).

In order to minimize the diff size, I'm moving the normalization
to the call-site of setTrackedFiles inside the pyright-scip folder.
We want to do a merge of upstream changes sooner rather than later,
so it'd be good to reduce the diff.

Updates some docs based on recent changes,
and updates package.json so that we can
create a new release where we get meaningful
stack traces on errors.

`env` on different systems behaves differently. On macOS,
`env ABC XYZ` will invoke ABC with argument XYZ.
On some Linux variants, you need `env -S ABC XYZ` for that.
On Alpine, the `-S` flag is not supported, and there is seemingly
no way of passing arguments.

So this patch undoes the accidental breakage in sourcegraph#179
and uses the `getSourceMapsSupport` API.

Technically, both `process.setSourceMapsEnabled` and
`module.setSourceMapsSupport` are experimental functions,
so to be safe, we do a dynamic existence check before calling
them for portability across Node versions.

…graph#183)

So it looks like basically there was a bug where the indexer was
generating 0 documents on macOS (related to the recent bug fix in
https://github.com/sourcegraph/scip-python/pull/177/files where we
started normalizing the file paths passed to setTrackedFiles). 😬 

However, the tests were passing on macOS not because they were
correct, but because the zero documents case bypassed the
updating/diffing process. 😢

This patch changes the test diffing logic to be more robust, in handling
various scenarios such as 0 documents. It also introduces a couple
of flags, one for filtering tests, and one for running tests with fail-fast
behavior.

On fixing the tests, I discovered some bugs due to improper handling
of case-sensitivity, so I fixed those (which is why the snapshots are unchanged
and the tests are passing on macOS). For some of these, I'm not super
sure about Pyright behavior, so I added a little assertions library
inspired by Antithesis so that we can test the coverage of different
cases by our test suite on macOS.

See also: https://antithesis.com/docs/best_practices/sometimes_assertions/

Changes the method lookup logic to use the same
logic as Pyright internals instead of our own ad-hoc
implementation of name resolution, which does not
match Python's method resolution order (MRO), which
is based purely on names, and not on types.

This gets rid of a code path where we were doing a `!`
operation, which triggered a crash when indexing
a prospect's codebase.

Fixes GRAPH-1278.

…cegraph#185)

* chore: Add linting and formatting commands to pyright-scip
* chore: Add AGENT.md for pyright-scip development instructions
* chore: Format pyright-scip with prettier
* chore: Add .git-blame-ignore-revs to ignore formatting commit
* chore: Add CI workflow for pyright-scip linting and formatting checks
* chore: Disable eslint checking -- too many issues
* chore: Run formatting & test checks on all PRs & commits
* chore: Replace asdf->mise

…ourcegraph#188)

Bumps nikolaik/python-nodejs from `fcbc543` to `3c587f7`.

---
updated-dependencies:
- dependency-name: nikolaik/python-nodejs
  dependency-version: python3.10-nodejs16-alpine
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

…urcegraph#193)

execSync in NodeJS takes a maxBuffer argument; it will not dynamically allocate
the output. When doing `pip show`, we could run into an ENOBUFS error if the buffer
size was too small, as a part of path listing. This means there are two options:

1. Use exec (or spawn): This requires CPS-ing the code all the way up to main.
2. Use a bigger buffer.

We go with option 2 because it's not worth changing everything to callbacks
just for this. All the other I/O uses sync operations anyway.

For the bigger buffer case, we put in retry logic because that is 'free';
it doesn't make sense to provide a CLI flag for the user to customize this,
because what they'll do is look at the error and increase the value;
so we just do it for them.

Changes the logic for listing packages to use importlib
which turns out to be much faster.

https://ampcode.com/threads/T-20d560ec-6811-4993-b63e-8935fa3d07f5#message-122-block-0

Adds an "integration" test around the Python script
so that we can be more confident that it works. The test
has both global and virtualenv packages, and both
are correctly located.

Also tested against `rich` and looks like the results
are unchanged.

The code could probably be cleaned up a bit, but I'm
prioritizing speed so that we can cut a release.