fix(ci,tooling,test-types,test-forks): import ethereum lazily so xdist fill coverage measures it

[danceratopz]

🗒️ Description

This PR improves the coverage measurement for ethereum and fixes the following warning that every fill job logs, once per pytest-xdist worker:

CoverageWarning: Module ethereum was previously imported, but not measured (module-not-measured)

Root cause

pytest-cov starts an xdist worker's coverage session late — in pytest_sessionstart. The execution_testing framework, however, imports the measured ethereum package at import time, well before that, from four module locations:

• base_types and test_types.trie import ethereum.crypto.hash.keccak256 at module level — executed as soon as the fill CLI and its plugins are imported;
• ExecutionSpecsTransitionTool imports the EELS t8n (T8N / ForkCache / create_parser / get_supported_forks / ethereum) at module level, and the forks plugin calls t8n.is_fork_supported() from pytest_configure.

So by the time a worker starts measuring, ethereum is already in sys.modules. pytest-cov's Central/DistMaster engines suppress this warning; DistWorker does not — hence it surfaces only on xdist workers.

Coverage impact

Pre-importing the source package also makes coverage under-count ethereum — in every mode, -n 0 and -n 1 alike.

On a small sample, deferring the import raises the measured statement total from 36,926 → 39,000 (+2,074 across 152 files). The xdist warning was the visible symptom of a measurement gap that affected all runs.

Update: the ~2,074 figure above was a measurement error (the "baseline" was a dirty checkout). A same-checkout A/B shows the statement denominator is unchanged (39,000). The real effect is small and constant: fill measures ~300 more ethereum statements — the import-time/header lines of modules loaded before the worker's coverage starts — ≈304 on a narrow Frontier slice and ≈303 on a 2,709-test Frontier→Cancun run, so it does not scale with the test set. At <1% it rounds to a ~0% codecov diff (see comment).

The fix

Defer all four ethereum imports to first use, so they run after the worker's coverage session has started (collection / test execution / the trylast report header). Specifically: lazy keccak256 imports; a lazy ForkCache property + in-function t8n imports + importlib.metadata for the tool version; and a lazily-computed, cached unsupported-fork set. Nothing is silenced — no filterwarnings, no --disable-warnings, no disable_warnings.

Results

• module-not-measured gone — verified on the minimal repro, repeated -n 4 runs, and a 2,709-test Frontier→Cancun fill.
• -n 0 and -n 1 report identical coverage; fill measures ~300 more ethereum statements (import-time lines) on the same 39,000-statement denominator — a constant <1% gain, not test-set-dependent.
• eels_base_coverage --cov-fail-under=85 gate passes (91.45%).
• 637 framework unit tests pass; just static clean.

⚠️ For reviewers: the statement denominator is unchanged; the coverage % moves by <1% (≈0% codecov diff), and the --cov-fail-under=85 gate still passes comfortably (91.45%).

Alternative considered

A runner-only approach was prototyped (start coverage earlier via COVERAGE_PROCESS_START + a dedicated coverage_fill.cfg in the Justfile, dropping pytest-cov's --cov and combining manually). It works, but we preferred fixing the source because it:

• removes the warning for every --cov=ethereum invocation, not just the just fill recipe (direct pytest/fill runs, json-loader, IDE runs, contributors);
• keeps the pytest-cov integration rather than hand-rolling coverage erase/combine/report/xml and a second coverage config to keep in sync with pyproject.toml;
• addresses the actual root cause — a test framework shouldn't import the package-under-test at import time.

_{The prototype lives on a local branch fix-coverage-warnings-start-coverage-earlier (currently based off the wrong base), kept only for reference.}

🔗 Related Issues or PRs

N/A.

✅ Checklist

• All: Ran fast static checks to avoid unnecessary CI fails, see also Code Standards and Enabling Pre-commit Checks:

  just static

• All: PR title adheres to the repo standard - it will be used as the squash commit message and should start type(scope):.
• All: Considered updating the online docs in the ./docs/ directory — no user-facing change.
• All: Set appropriate labels for the changes (only maintainers can apply labels).

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.22%. Comparing base (3f888bc) to head (7a858f2).

Additional details and impacted files

@@               Coverage Diff                @@
##           forks/amsterdam    #3059   +/-   ##
================================================
  Coverage            93.22%   93.22%           
================================================
  Files                  624      624           
  Lines                36926    36926           
  Branches              3377     3377           
================================================
  Hits                 34424    34424           
  Misses                1708     1708           
  Partials               794      794           

Flag	Coverage Δ
unittests	93.22% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.