Skip to content

ref: second cleanup sweep — coverage/benchmark numeric bugs and duplication#880

Merged
Chemaclass merged 7 commits into
mainfrom
cleanup/sweep-round2
Jul 25, 2026
Merged

ref: second cleanup sweep — coverage/benchmark numeric bugs and duplication#880
Chemaclass merged 7 commits into
mainfrom
cleanup/sweep-round2

Conversation

@Chemaclass

Copy link
Copy Markdown
Member

🤔 Background

Related #879

A second 8-agent pass over the codebase. Four of the eight dimensions came back
genuinely clean (circular dependencies, unused code, defensive programming, legacy
code) — the earlier sweep in #862 and the CLI hardening in #871#878 had already taken
that ground. The remaining four found two real bugs and two duplication clusters.

💡 Changes

  • Validate BASHUNIT_COVERAGE_THRESHOLD_LOW/HIGH: a non-integer leaked a raw integer expression expected into coverage reports and mis-bucketed every file's class (90% reported as medium)
  • Compare @max_ms benchmark thresholds as floats: a decimal limit errored and rendered every row as failing regardless of the real average
  • Deduplicate the hits_by_line loader, copy-pasted into 7 coverage report writers, and the spy call-count lookup repeated across 3 assertions
  • Reuse env.sh's canonical threshold defaults instead of hardcoding 80/50 in 6 places, and strip 18 lines of step-narrating comments from the assert CLI path

Both bugs are not reproducible from inside a repo checkout: .env/.env.example list the threshold names with empty values, and an allexport source .env overrides an exported value (#865). The new tests use --skip-env-file for that reason.

exec_assert/exec_multi_assert in src/main.sh had a comment restating
almost every line ("Check if the function exists", "Set test title
for this assertion", etc.) added when multi-assertion mode landed.
Remove them along with matching Arrange-Act-Assert narration in
bashunit_init_test.sh and a redundant section header in
bashunit_assert_multi_test.sh. Load-bearing comments (Bash 3 array
guard, #877 reference, exit-code vs output assertion branches) are
left untouched.
They are compared with `[ -ge ]` in the coverage class lookup, which
errors instead of returning false on a non-integer value: a bad
threshold leaked a raw "integer expression expected" into the coverage
report and silently mis-bucketed every file's high/medium/low class.
Reuses bashunit::main::require_non_negative_int_or_exit, the same gate
already covering the other numeric BASHUNIT_* settings (#873).
`@max_ms` accepts a decimal (its parse_annotations regex is
`[0-9.][0-9.]*`, and the average itself can already be fractional),
but the status column used plain `[ "$avg" -le "$max_ms" ]`. That
errors instead of comparing on a fractional operand, so a well
under-budget row printed a stray "integer expression expected" line
and always rendered as failing (">") regardless of the real average.

Adds bashunit::math::is_le, mirroring bashunit::math::calculate's
bc > awk > integer fallback chain.
…them

get_coverage_class and the HTML legend independently hardcoded 80/50 as
the BASHUNIT_COVERAGE_THRESHOLD_HIGH/LOW fallback in six places, so a
future change to _BASHUNIT_DEFAULT_COVERAGE_THRESHOLD_HIGH/LOW in env.sh
(the single source of truth for every other BASHUNIT_* default) would
silently drift from these copies. Reference the canonical globals
instead, and pin the fallback behavior with a regression test.
Seven functions independently re-parsed get_all_line_hits into their own
local -a hits_by_line with an identical 5-line while-loop. Bash 3.0 can't
return an array from a function or copy a sparse array without losing its
line-number keys, so extract the parse into
bashunit::coverage::load_hits_by_line, which writes the shared
_BASHUNIT_COVERAGE_HITS_BY_LINE global instead (same return-slot pattern
already used elsewhere in this file). No behavior change; existing
coverage unit/acceptance tests cover every call site touched.
assert_have_been_called, assert_have_been_called_times and
assert_have_been_called_nth_with each repeated the same "resolve the spy's
times file, cat it, default to 0" sequence. Extract
bashunit::spy::times_to_slot (return-slot pattern, no added forks) and
call it from all three. Adds direct unit coverage for the new helper
alongside the existing spy assertion tests.
The two threshold tests passed in isolation but failed once integrated:
.env is gitignored, so it exists in the main checkout but not in the
worktrees they were written in. Both .env and .env.example list
BASHUNIT_COVERAGE_THRESHOLD_LOW/HIGH with an empty value, and an
allexport 'source .env' turns that listing into an unconditional
assignment that overrides the exported value under test (#865). CI does
'cp .env.example .env', so this would have gone red there too.

Run them with --skip-env-file, and correct an issue reference that
pointed at a number that did not exist.
@Chemaclass Chemaclass added the bug Something isn't working label Jul 25, 2026
@Chemaclass Chemaclass self-assigned this Jul 25, 2026
@Chemaclass
Chemaclass merged commit 5fc5766 into main Jul 25, 2026
37 checks passed
@Chemaclass
Chemaclass deleted the cleanup/sweep-round2 branch July 25, 2026 10:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

bug Something isn't working

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant