test: observational baseline for per-module rebuild counts by robinbb · Pull Request #14310 · ocaml/dune

robinbb · 2026-04-23T20:43:39Z

Summary

Adds a blackbox cram test that records the rebuild-target count for a
consumer module C when each of three entry modules (A1, A2, A3)
of an unwrapped dependency library has its interface edited. C
references only A2.

On current main, editing any of A1/A2/A3 rebuilds C (conservative
glob dep over the dependency's object dir). The recorded counts are
A1=1, A3=1, A2>0. Once per-module tightening (#4572) lands, A1
and A3 counts go to zero and a later PR promotes the test output.

Related: #4572, #14116

Copilot

Pull request overview

Adds an observational blackbox cram test to capture the current (pre-#4572) rebuild behavior for a consumer module when interfaces of entry modules in an unwrapped dependency library change.

Changes:

Introduces a new cram test that edits A1, A2, A3 interfaces in an unwrapped base library and records rebuild-target counts for consumer module C.
Records current baseline expectations (A1=1, A3=1, A2>0) intended to be promoted once per-module tightening lands.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Copilot

Pull request overview

Copilot reviewed 1 out of 1 changed files in this pull request and generated no new comments.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

robinbb · 2026-04-24T03:11:51Z

Looking for a volunteer to review.

Per rgrinberg review on ocaml#14310: replace the [length]-based jq pipeline with the idiomatic [targetsMatchingFilter] pattern used in sibling tests (e.g. alias-reexport.t). The recorded output now shows the actual target paths under [_build/default/consumer/] that are rebuilt when each of A1/A2/A3 is edited, rather than a bare count. This also unifies the three blocks: A1, A3, and A2 all use the same pipeline. Once ocaml#14116's per-module filter lands, the A1/A3 blocks promote to [], while A2 keeps the c.cmi/c.cmo/c.cmt list. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

ocaml#14310 (the dep PR that owns unwrapped-tight-deps.t) was updated per rgrinberg's review to print rebuild target names via [targetsMatchingFilter] rather than counts via [length], matching the idiom used by sibling tests like alias-reexport.t. Resync this PR's copy with ocaml#14310's current shape: take the canonical content from ocaml#14310, with A1/A3 assertions promoted to [] (the per-module filter drops unreferenced [base] modules from [c]'s deps) and A2 keeping its rebuild target list (the one module [c] references must still rebuild). The other dep PRs whose tests this branch borrows have only prose-style differences (post-fix prose on this branch vs observational-baseline prose on the dep PR) — not new churn — so they don't need a similar resync. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Add a blackbox test that records, for a consumer module C referencing only A2 of an unwrapped library [base] with three entry modules A1/A2/A3 (each with an explicit interface), the rebuild-target count for C when each of A1/A2/A3 has its signature edited. On current main, editing any of A1/A2/A3 rebuilds C because the consumer depends conservatively on a directory-wide glob over [base]'s object dir. The test records A1=1, A3=1, A2>0. The test anchors the tightening surface for ocaml#4572: once per-module filtering lands, editing A1 or A3 is expected to leave C untouched, at which point the recorded counts are promoted. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Previously [consumer] was a single-module library stanza. Dune skips ocamldep for single-module stanzas with no library deps as an optimisation, and the per-module-lib-deps filter depends on ocamldep output. So under the filter's feature branch the test's A1/A3 rebuild counts flipped from 1 to 0 for two independent reasons: the filter running on ocamldep output, and a concurrent tweak to the skip-ocamldep optimisation to run ocamldep for single-module stanzas that have library deps. Add an unused module [d.ml] to [consumer]. Ocamldep now runs unconditionally for [consumer], and the recorded rebuild count for [c] reflects only the per-module filter's work, not the skip-ocamldep behaviour. The probe regex still matches [c] only so [d] being in the rebuild set does not affect the count. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Per rgrinberg review on ocaml#14310: replace the [length]-based jq pipeline with the idiomatic [targetsMatchingFilter] pattern used in sibling tests (e.g. alias-reexport.t). The recorded output now shows the actual target paths under [_build/default/consumer/] that are rebuilt when each of A1/A2/A3 is edited, rather than a bare count. This also unifies the three blocks: A1, A3, and A2 all use the same pipeline. Once ocaml#14116's per-module filter lands, the A1/A3 blocks promote to [], while A2 keeps the c.cmi/c.cmo/c.cmt list. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Each of the dep PRs supplying a test in [per-module-lib-deps/] was renamed per Leonidas's review on ocaml#14309: generic [libA]/[modA1]-style names replaced by role-bearing names like [dep_lib]/[spurious_rebuild]. This commit applies the same renames to this PR's local copies of those tests so they stay convergent with the dep PRs' canonical versions. Touches: - single-module-unreferenced-lib.t (ocaml#14308) - sibling-unreferenced-lib.t (ocaml#14309) - unwrapped-tight-deps.t (ocaml#14310) - lib-vs-lib-name-collision.t (ocaml#14324) - opaque-mli-change.t, opaque-cmx-deps-local.t (ocaml#14331) - transitive-unreferenced-lib.t (ocaml#14332) - wrapped-reexport-via-open-flag.t (ocaml#14346) - auto-wrapped-child-reexport.t (ocaml#14347) Pure rename + path/dir adjustments. Count assertions preserved (this PR's promotions to the post-fix counts are unchanged). Test suite passes. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

ocaml#14310 (the dep PR that owns unwrapped-tight-deps.t) was updated per rgrinberg's review to print rebuild target names via [targetsMatchingFilter] rather than counts via [length], matching the idiom used by sibling tests like alias-reexport.t. Resync this PR's copy with ocaml#14310's current shape: take the canonical content from ocaml#14310, with A1/A3 assertions promoted to [] (the per-module filter drops unreferenced [base] modules from [c]'s deps) and A2 keeping its rebuild target list (the one module [c] references must still rebuild). The other dep PRs whose tests this branch borrows have only prose-style differences (post-fix prose on this branch vs observational-baseline prose on the dep PR) — not new churn — so they don't need a similar resync. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Per Copilot review on ocaml#14332: [intermediate_lib] and the [main] executable were single-module stanzas. Dune's [dep_rules.ml] short-circuits ocamldep for single-module stanzas, so the test was exercising the cctx-wide glob fallback rather than isolating the per-module dependency filtering being baselined. A future change that flips the short-circuit (e.g. adding [not has_library_deps]) would change this test's count for an unrelated reason — the same conflation that ocaml#14310 had to undo retroactively. Add an unused [intermediate_dummy] module to [intermediate_lib] and an unused [main_dummy] module to the [main] executable so ocamldep runs unconditionally for both stanzas. Tighten the trace-probe regex to [dune__exe__Main\\.] so it only matches Main targets, not the new [Main_dummy]. Switch from a [length]-based count to the idiomatic [targetsMatchingFilter] JSON dump used elsewhere in this directory; the recorded targets are now the explicit [.cmi/.cmti] and [.cmo/.cmt] target_files lists for the two ocamlc invocations on Main. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR ocaml#14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR #14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR ocaml#14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR #14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR ocaml#14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR #14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR ocaml#14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

xyz-3 was branched from xyz-2 c71, which predates several test- only refinements that landed on xyz: library renames for clarity ([lib_a]/[lib_b] → [dep_lib]/[consumer_lib]), exact-[target_files] array assertions in place of [length] counts, and the format resync of [unwrapped-tight-deps.t] tracking dep PR #14310. This commit copies the xyz HEAD versions of the ten cluster tests that diverged. All assertions hold under the conservative wrapped- lib glob fix; no test outcome changes. Files: - auto-wrapped-child-reexport.t - implicit-transitive-deps-false.t - lib-vs-lib-name-collision.t - opaque-cmx-deps-local.t - opaque-mli-change.t - sibling-unreferenced-lib.t - single-module-unreferenced-lib.t - transitive-unreferenced-lib.t - unwrapped-tight-deps.t - wrapped-reexport-via-open-flag.t Signed-off-by: Robin Bate Boerop <me@robinbb.com>

robinbb self-assigned this Apr 23, 2026

robinbb requested a review from Copilot April 23, 2026 20:49

Copilot started reviewing on behalf of robinbb April 23, 2026 20:50 View session

Copilot AI reviewed Apr 23, 2026

View reviewed changes

Comment thread test/blackbox-tests/test-cases/per-module-lib-deps/unwrapped-tight-deps.t Outdated

robinbb requested a review from Copilot April 23, 2026 22:54

Copilot started reviewing on behalf of robinbb April 23, 2026 22:54 View session

Copilot AI reviewed Apr 23, 2026

View reviewed changes

robinbb marked this pull request as ready for review April 23, 2026 23:06

robinbb mentioned this pull request Apr 24, 2026

feat: per-module inter-library dependency filtering (#4572) #14116

Closed

rgrinberg reviewed Apr 25, 2026

View reviewed changes

Comment thread test/blackbox-tests/test-cases/per-module-lib-deps/unwrapped-tight-deps.t Outdated

robinbb assigned rgrinberg Apr 27, 2026

robinbb added 3 commits April 27, 2026 07:20

robinbb force-pushed the robinbb-test-unwrapped-tight-deps branch from 60fd49a to f56a633 Compare April 27, 2026 15:40

rgrinberg approved these changes Apr 27, 2026

View reviewed changes

robinbb merged commit ba4e542 into ocaml:main Apr 27, 2026
101 of 114 checks passed

robinbb deleted the robinbb-test-unwrapped-tight-deps branch April 27, 2026 22:01

robinbb mentioned this pull request May 10, 2026

feat: per-module inter-library dependency filtering (#4572) #14472

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

test: observational baseline for per-module rebuild counts#14310

test: observational baseline for per-module rebuild counts#14310
robinbb merged 3 commits into
ocaml:mainfrom
robinbb:robinbb-test-unwrapped-tight-deps

robinbb commented Apr 23, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

robinbb commented Apr 24, 2026

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Conversation

robinbb commented Apr 23, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

robinbb commented Apr 24, 2026

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

robinbb commented Apr 23, 2026 •

edited

Loading