test: sibling-module consumer unreferenced-dep regression guard by robinbb · Pull Request #14309 · ocaml/dune

robinbb · 2026-04-23T20:43:30Z

Summary

Adds a blackbox cram test recording the rebuild-target count for a
consumer module spurious_rebuild in a multi-module unwrapped
consumer library whose sibling consumes_dep references a module of
the declared dependency but spurious_rebuild itself references
nothing from that dependency.

On current main the count is 1: the consumer library depends on a
glob over dep_lib's object directory, which is invalidated by the
cmi change on referenced_dep. A future fix implementing library-
level dep filtering at consumer-module granularity would drop
dep_lib entirely from spurious_rebuild's deps and tighten this
corner to 0.

This zero-reference-sibling-in-a-library corner is not covered by
existing tests:

test-cases/per-module-lib-deps/lib-to-lib-unwrapped.t probes
siblings that reference a different (non-edited) module of the dep
— one-reference, not zero.
test-cases/per-module-lib-deps/transitive.t and
test-cases/per-module-lib-deps/unwrapped.t probe zero-reference
modules but within executable stanzas, not library stanzas.

Related: #4572, #14116

Copilot

Pull request overview

Adds a blackbox cram test to guard against regressions in per-module inter-library dependency filtering: when only one module in a consumer library references a dependency library, edits to the dependency’s referenced module (without interface/signature change) must not cause recompilation of the consumer’s unreferenced sibling module.

Changes:

Add a new per-module-lib-deps blackbox test covering the “sibling module does not reference dependency” scenario.
Validate via dune trace that modB1 is not rebuilt after a non-signature-changing edit to libA’s modA2.ml.

💡 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 1 comment.

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

Per Copilot review on ocaml#14309: the test does not rely on any feature newer than 3.0; the [(lang dune)] version can gate dependency-analysis behaviour, so using 3.22 may unintentionally change what this test is guarding. Consistent with other baseline tests in this directory. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

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.

Per Copilot review on ocaml#14309: the test does not rely on any feature newer than 3.0; the [(lang dune)] version can gate dependency-analysis behaviour, so using 3.22 may unintentionally change what this test is guarding. Consistent with other baseline tests in this directory. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

Leonidas-from-XIV

I think the test is fine, my comments are mostly about how to make the test easier to read.

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via ocaml#14309 as an observational baseline (originating from @nojb's repro in ocaml#14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. Signed-off-by: Robin Bate Boerop <me@robinbb.com>

@nojb

The test came in via #14309 as an observational baseline (originating from @nojb's repro in #14116 (comment)). On trunk it asserts a count of 1 — [spurious_rebuild] re-runs once because the consumer's compile rule globs over [dep_lib]'s objdir and any [.cmi] change in there fires. [| length] hides which target actually re-ran. A future change that quietly swaps which rule re-runs (e.g. [.cmo]/[.cmt] dropped but [.cmx] picked up) could still show [length = 1] and slip past the test. Promote the assertion to inspect the array of [target_files] returned by [targetsMatchingFilter] — same pattern as [implicit-transitive-deps-false.t] and the [[]] / [true] asserts in [alias-reexport.t]. This is a style-only upgrade: the asserted shape (one rule re-running, producing [.cmi]+[.cmo]+[.cmt]) reflects trunk's current behaviour. On this PR the per-module filter narrows the deps, eliminating the rebuild — the diff produced by running this test against the PR (count = 0 on disk vs the asserted single rule) is the visible improvement, and a later commit will flip the assertion to [[]] alongside the rest of the post-fix test promotions. 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 20:46

Copilot started reviewing on behalf of robinbb April 23, 2026 20:46 View session

Copilot AI reviewed Apr 23, 2026

View reviewed changes

robinbb marked this pull request as ready for review April 23, 2026 22:31

robinbb requested a review from Copilot April 23, 2026 23:04

Copilot started reviewing on behalf of robinbb April 23, 2026 23:05 View session

Copilot AI reviewed Apr 23, 2026

View reviewed changes

Comment thread test/blackbox-tests/test-cases/per-module-lib-deps/sibling-unreferenced-lib.t Outdated

robinbb requested a review from Copilot April 23, 2026 23:52

Copilot started reviewing on behalf of robinbb April 23, 2026 23:53 View session

Copilot AI reviewed Apr 23, 2026

View reviewed changes

robinbb force-pushed the robinbb-test-sibling-unreferenced-lib branch from 6b184c5 to 7027ba0 Compare April 24, 2026 00:54

robinbb mentioned this pull request Apr 24, 2026

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

Closed

Leonidas-from-XIV requested changes Apr 27, 2026

View reviewed changes

Comment thread test/blackbox-tests/test-cases/per-module-lib-deps/sibling-unreferenced-lib.t Outdated

Comment thread test/blackbox-tests/test-cases/per-module-lib-deps/sibling-unreferenced-lib.t Outdated

robinbb mentioned this pull request May 10, 2026

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

Draft

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

test: sibling-module consumer unreferenced-dep regression guard#14309

test: sibling-module consumer unreferenced-dep regression guard#14309
robinbb merged 2 commits into
ocaml:mainfrom
robinbb:robinbb-test-sibling-unreferenced-lib

robinbb commented Apr 23, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

Copilot AI left a comment

Uh oh!

Leonidas-from-XIV left a comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 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!

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!

Leonidas-from-XIV left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

robinbb commented Apr 23, 2026 •

edited

Loading