-
Notifications
You must be signed in to change notification settings - Fork 413
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
fix(melange): track immediate .cmj
deps as dependencies of JS rules
#10286
fix(melange): track immediate .cmj
deps as dependencies of JS rules
#10286
Conversation
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
.cmj
deps as dependencies of JS rules
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Great work, thanks for the fix!
I'm not sure I understand this issue exactly, but the purpose of the additional dependency is only to indirectly depend on the relative path of the module wrt to the library? If that's the case, isn't this a pretty big over-approximation of that? |
I don't think this is an over approximation. Here's my understanding:
|
Okay so I got that right, the melange compiler reads these files? |
Indeed it does |
Can you think of a better approach than this? My understanding is that the cmj dependencies are de facto dependencies of the JS emission step. |
If the melange compiler reads them, then I see no better approach. One should be able to reproduce the bug by turning on the sandboxing for the melange rules. Melange attempt to find some |
ded8bd5
to
fed936f
Compare
… rules Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
fed936f
to
58dd3ba
Compare
This should be ready for another look |
Signed-off-by: Antonio Nuno Monteiro <anmonteiro@gmail.com>
CHANGES: ### Added - Add link flags to to `ocamlmklib` for ctypes stubs (ocaml/dune#8784, @frejsoya) - Remove some unnecessary limitations in the expansions of percent forms in install stanza. For example, the `%{env:..}` form can be used to select files to be installed. (ocaml/dune#10160, @rgrinberg) - Allow artifact expansion percent forms (`%{cma:..}`, `%{cmo:..}`, etc.) in more contexts. Previously, they would be randomly forbidden in some fields. (ocaml/dune#10169, @rgrinberg) - Allow `%{inline_tests}` in more contexts (ocaml/dune#10191, @rgrinberg) - Remove limitations on percent forms in the `(enabled_if ..)` field of libraries (ocaml/dune#10250, @rgrinberg) - Support dialects in `dune describe pp` (ocaml/dune#10283, @emillon) - Allow defining executables or melange emit stanzas with the same name in the same folder under different contexts. (ocaml/dune#10220, @rgrinberg, @jchavarri) ### Fixed - coq: Delay Coq rule setup checks so OCaml-only packages can build in hybrid Coq/OCaml projects when `coqc` is not present. Thanks to @vzaliva for the test case and report (ocaml/dune#9845, fixes ocaml/dune#9818, @rgrinberg, @ejgallego) - Fix conditional source selection with `select` on `bigarray` in OCaml 5 (ocaml/dune#10011, @moyodiallo) - melange: fix inconsistency in virtual library implementation. Concrete modules within a virtual library can now refer to its virtual modules too (ocaml/dune#10051, fixes ocaml/dune#7104, @anmonteiro) - melange: fix a bug that would cause stale `import` paths to be emitted when moving source files within `(include_subdirs ..)` (ocaml/dune#10286, fixes ocaml/dune#9190, @anmonteiro) - Dune file formatting: output utf8 if input is correctly encoded (ocaml/dune#10113, fixes ocaml/dune#9728, @moyodiallo) - Fix expanding dependencies and locks specified in the cram stanza. Previously, they would be installed in the context of the cram test, rather than the cram stanza itself (ocaml/dune#10165, @rgrinberg) - Fix bug with `dune exec --watch` where the working directory would always be set to the project root rather than the directory where the command was run (ocaml/dune#10262, @gridbugs) - Regression fix: sign executables that are promoted into the source tree (ocaml/dune#10263, fixes ocaml/dune#9272, @emillon) - Fix crash when decoding dune-package for libraries with `(include_subdirs qualified)` (ocaml/dune#10269, fixes ocaml/dune#10264, @emillon) ### Changed - Remove the `--react-to-insignificant-changes` option. (ocaml/dune#10083, @rgrinberg)
CHANGES: ### Added - Add link flags to to `ocamlmklib` for ctypes stubs (ocaml/dune#8784, @frejsoya) - Remove some unnecessary limitations in the expansions of percent forms in install stanza. For example, the `%{env:..}` form can be used to select files to be installed. (ocaml/dune#10160, @rgrinberg) - Allow artifact expansion percent forms (`%{cma:..}`, `%{cmo:..}`, etc.) in more contexts. Previously, they would be randomly forbidden in some fields. (ocaml/dune#10169, @rgrinberg) - Allow `%{inline_tests}` in more contexts (ocaml/dune#10191, @rgrinberg) - Remove limitations on percent forms in the `(enabled_if ..)` field of libraries (ocaml/dune#10250, @rgrinberg) - Support dialects in `dune describe pp` (ocaml/dune#10283, @emillon) - Allow defining executables or melange emit stanzas with the same name in the same folder under different contexts. (ocaml/dune#10220, @rgrinberg, @jchavarri) ### Fixed - coq: Delay Coq rule setup checks so OCaml-only packages can build in hybrid Coq/OCaml projects when `coqc` is not present. Thanks to @vzaliva for the test case and report (ocaml/dune#9845, fixes ocaml/dune#9818, @rgrinberg, @ejgallego) - Fix conditional source selection with `select` on `bigarray` in OCaml 5 (ocaml/dune#10011, @moyodiallo) - melange: fix inconsistency in virtual library implementation. Concrete modules within a virtual library can now refer to its virtual modules too (ocaml/dune#10051, fixes ocaml/dune#7104, @anmonteiro) - melange: fix a bug that would cause stale `import` paths to be emitted when moving source files within `(include_subdirs ..)` (ocaml/dune#10286, fixes ocaml/dune#9190, @anmonteiro) - Dune file formatting: output utf8 if input is correctly encoded (ocaml/dune#10113, fixes ocaml/dune#9728, @moyodiallo) - Fix expanding dependencies and locks specified in the cram stanza. Previously, they would be installed in the context of the cram test, rather than the cram stanza itself (ocaml/dune#10165, @rgrinberg) - Fix bug with `dune exec --watch` where the working directory would always be set to the project root rather than the directory where the command was run (ocaml/dune#10262, @gridbugs) - Regression fix: sign executables that are promoted into the source tree (ocaml/dune#10263, fixes ocaml/dune#9272, @emillon) - Fix crash when decoding dune-package for libraries with `(include_subdirs qualified)` (ocaml/dune#10269, fixes ocaml/dune#10264, @emillon) ### Changed - Remove the `--react-to-insignificant-changes` option. (ocaml/dune#10083, @rgrinberg)
fixes #9190
Root cause assessment
foo.ml
->starting_dir/bar.ml
and we rebuild after movingbar.ml
toend_dir/bar.ml
, thefoo.cmj
file produced in both builds will be byte-by-byte identical.foo.js -> (dependencies of foo.cmj)
(foo.cmj -> bar.cmj
andbar.ml
changed)Discussion & Solution
this is only an issue if the rules change between builds, so I believe we only need to account for local builds (i.e.
melange.emit
entries and local libraries -- we assume that the installed world didn't change)the solution: we add the immediate dependencies of the
.cmj
we're compiling to JS asHidden_deps
of the JS target rule, which takes care of invalidating the JS builds if a source path has changedThis PR is best reviewed commit-by-commit
@rgrinberg I'm requesting your review on this one for the performance impact: we're now calling
Dep_rules.immediate_deps_of
for every JS target, which in turn callsOcamldep.read_immediate_deps_of
. My understanding is that the latter call is memoized so this change shouldn't have any noticeable impact. Please double check this assumption.