feat(fill,execute,benchmark): test selection via pytest's `testpaths` instead of markers via `--mode`

[danceratopz]

We now have 3 classes of test:

1. Regular tests.
2. Computationally intensive tests (./tests/benchmark/compute[*])
3. State intensive tests (./tests/benchmark/state/).

( [*] These paths don't reflect the current state, but the proposal here: #2208 )

The requirement, as I understand, is:

1. fill should only fill regular tests by default (same for execute).
2. fill -m benchmarks should only run tests from ./tests/benchmarks/compute/ (same for execute).
3. execute -m stateful should only execute tests from ./tests/benchmarks/state/ (fill is currently not required/realistically possible).

This requirement is implemented (via pytest markers) here:

• chore(benchmark): don't fill benchmark tests by default #1920, and
• feat: add bloatnet marker support #2169 (WIP).

Using pytest markers for this case results in a lot of very complex logic, and, as it stands in #2169, adds two additional modifications of pytest_collect_modifyitems() in two conftests under ./tests/.

There seems to be a much simpler solution: Just collect tests from the applicable sub-directories. This will:

1. Result in much simpler pytest logic.
2. Result in much faster test collection times.

This can be achieved by modifying the pytest ini file testpaths config for the fill and execute commands. This config tells pytest where to search for tests.

UX and Corresponding Required Pytest Flags

I'd suggest adding a --mode flag, which is None by default.

Relevant pytest config and docs:

• testpaths
  • See our ini files, e.g. https://github.com/LouisTsai-Csie/execution-spec-tests/blob/7e4d3e214252fa878a7ac5c381fcab0d48d1c40a/src/cli/pytest_commands/pytest_ini_files/pytest-fill.ini#L5
  • https://docs.pytest.org/en/stable/reference/reference.html#confval-testpaths
• -o:

  -o OVERRIDE_INI, --override-ini=OVERRIDE_INI
                        Override ini option with "option=value" style, e.g. `-o
                        xfail_strict=True -o cache_dir=cache`.
  

• -ignore:

  --ignore=path         Ignore path during collection (multi-allowed)
  

1. EIP testing (default behavior):

   fill
   

   should result in no changes to testpaths but automatically add these pytest flags:

   fill --ignore=tests/benchmarks
   execute --ignore=tests/benchmarks
   

2. Compute benchmark tests:

   fill --mode=compute
   

   is equivalent to modifying testpaths via these flags:

   fill -o "testpaths=tests/benchmarks/compute"
   execute -o "testpaths=tests/benchmarks/compute"
   

   Also possible:

   fill -o "testpaths=tests/benchmarks/instructions,tests/benchmarks/scenarios,tests/benchmarks/precompiles"
   

3. State benchmark tests:

   fill --mode=compute
   

   is equivalent to modifying testpaths via these flags:

   fill -o "testpaths=tests/benchmarks/state"
   execute -o "testpaths=tests/benchmarks/state"
   

Implementation

Two options, I think only the second works:

1. Add the flag to the execute_fill plugin. But I don't think this works as testpaths are so critical to pytest config, that it won't take effect early enough. But it'd be less code and cleaner.
2. Modify the behavior via the EEST CLI interface to our pytest commands in https://github.com/ethereum/execution-spec-tests/tree/a2f28413de99d10d349844872a860a340cb2f345/src/cli/pytest_commands . This would mean adding a new Processor class and appending the relevant flags based on the value of --mode. As a reference, you can see how --watch was added in ✨ feat(fill): watch mode #2173.