fix(rewrite): keep pytest raw before a positional tail/head

[Cedric921]

Summary

Since #3171/#3343 the producer in front of a display-only consumer gets rewritten, so python3 -m pytest tests/ -q 2>&1 | tail -3 became rtk pytest ... | tail -3. rtk pytest prints its summary first and the tee pointers last. On a failing run, tail -3 then returned only the pointers, with no counts and no failure names.

• SafePipeConsumer gets a positional field: head and tail are positional, cat is not. analyze_pipeline records whether any consumer is positional.
• RewriteContext::PipelineProducer carries that flag. A producer listed in the new POSITION_SHIFTING_PRODUCERS (currently only rtk pytest) stays raw when a positional consumer follows it.
• No behaviour change for pytest alone, pytest | cat, other producers in front of head/tail, or final-stage grep/rg.
• src/discover/README.md documents the new rule.

This is the narrow form of the fix #4207 proposes for grep/rg. The table can take other producers whose rtk output moves lines (grep/rg from #4207, git log caps) as separate, reviewable changes. #4382 takes a broader route: it stops rewriting every pipeline producer. If that lands, this change is no longer needed.

Fixes #4445

Test plan

• New test_rewrite_pipe_reordering_producer_before_positional_consumer_stays_raw failed on develop (left: Some("rtk pytest t.py -q 2>&1 | tail -3")) and passes with the fix
• test_rewrite_pipe_producer_batch_rules_rewritten now checks pytest | cat (it used to expect pytest | tail -20 to be rewritten)
• cargo fmt --all --check
• cargo clippy --all-targets with zero warnings
• cargo test: 3894 unit tests plus the integration suites, 0 failures
• rtk rewrite "python3 -m pytest tests/ -q 2>&1 | tail -3" exits 1 (left raw); rtk rewrite "pytest -q" still gives rtk pytest -q

[rtk-wshm-sync-bot]

wshm · Automated triage by AI

📊 Automated PR Analysis

🐛 Type	bug-fix
🟢 Risk	low

Summary

Fixes a bug where rewriting pytest ... | tail -N to rtk pytest ... | tail -N caused tail to select the tee pointer lines instead of the pytest summary, since rtk pytest reorders output. The fix marks head/tail as positional consumers and keeps known line-reordering producers (currently rtk pytest) raw when followed by such a consumer.

Review Checklist

• Tests present
• Breaking change
• Docs updated

Linked issues: #4445

---

Analyzed automatically by wshm · This is an automated analysis, not a human review.

[Protocol-zero-0]

Independently validated this narrow fix against its own parent eb225a1 and head f365206 with real pytest 9.0.3 on Linux.

For a fixture with 12 failures and 5 passing tests, both python3 -m pytest test_probe.py -q 2>&1 | tail -3 and pytest test_probe.py -q 2>&1 | cat | tail -3 lose the session counts on the parent and return recall pointers. This head leaves those producers raw, so the selected output again contains the final failure names and 12 failed, 5 passed. All raw/parent/head executions retain exit 1 under bash -o pipefail.

Nine rewrite probes cover tail, head, an intermediate cat, cat-only, standalone pytest, a compound command, stderr piping, following tail, and final-stage grep. Cat-only and standalone pytest still rewrite; the compound's independent trailing pytest still rewrites; existing conservative stderr/following-tail behavior is preserved. Twelve real pytest executions cover the two failing positional pipelines plus cat-only and standalone controls. Both versions built with cargo build --locked --offline; 21 targeted pipeline unit tests and 36 hook-decision protocol integration tests pass on the head.

This approval covers the pytest positional-consumer fix for #4445. It does not establish that other producer filters preserve line positions, or replace the broader design discussion in #4382/#4207. I did not run the complete suite, clippy, or cross-platform tests.

[Protocol-zero-0]

Re-verified the develop merge at 2df1f37 against its current develop parent ffa5651 and the previously reviewed f365206. The conflict is resolved, and the fix still preserves the pytest summary before positional consumers.

On Linux with real pytest 9.0.3, a fixture with 12 failures and 5 passes retains the failure names and final counts through tail -3, cat | tail -3, and bash -c wrapping the tail pipeline. Current develop reproduces the missing-summary behavior in those three paths. Standalone pytest and pytest | cat still rewrite. Complete case statements stay raw in all three binaries. Exit behavior matches raw execution, including the nested bash pipeline's existing exit 0 when pipefail is not enabled inside that shell.

Validation: 14 rewrite cases across three binaries (42 calls), 32 real pytest executions, 680 discover tests and 36 hook decision protocol tests pass; fmt and clippy with warnings denied pass. The README retains develop's case-terminator and shell-wrapper wording alongside this positional-consumer rule. Full repository tests and other platforms were not run.

My approval stands for this merged head. #4382 remains the broader alternative; the integration choice stays with the maintainers.