Use a different type for primitives in clambda and lambda#1579
Use a different type for primitives in clambda and lambda#1579chambart merged 17 commits intoocaml:trunkfrom
Conversation
|
The CI failure are tests that does not compile to the exact same binary with ocamlopt.opt and ocamlopt.byte. |
|
I must have misdone something. I never intended to close this... |
|
I'm going to rebase this onto trunk since I've already had to do that once on the dev Flambda branch. |
22ae9d7 to
bb412a4
Compare
|
Rebased |
|
@alainfrisch @xavierleroy would you object to this ? Otherwise this is ready to merge. |
|
Well, I tried to read the diff. It's so big that I gave up. I'm fine with the idea; it's just that someone else will have to check all those changes and vouch that they are correct and not excessively intrusive. BTW: I see a .ml file containing type definitions only, and an identical .ml file. What's the policy du jour concerning those? It's no longer OK to have only the .mli? or only the .ml? |
|
It's problematic to have only the mli. It's ok to have only the ml, but the Makefiles do not seem to handle that well. Maybe a symlink is the right way to go. |
bb412a4 to
a217b22
Compare
|
There was some rebasing error. Pushed a cleaned version. |
c1379ba to
e67c2e2
Compare
|
Rebased |
|
I will review this once the CI passes. |
|
Notice that it might be easier to review the patches separately rather than the complete diff. |
|
On 64bits the build failure is a binary that is different in bytecode and native. |
The primites for Clambda are now different from lambda ones. This will now allow to clean those that don't make any sense in the backend.
2dfbe28 to
8c47eb0
Compare
|
Rebased on trunk. (thanks @lthls) |
lthls
left a comment
There was a problem hiding this comment.
I've reviewed the implementation, and I believe it is correct.
The actual changes in the rest of the code are minimal: the primitives for Lambda are left untouched, and the new type introduced for Clambda primitives is the same as the Lambda one except for the following changes:
- The
Pstring_load_*,Pbytes_load_*,Pbytes_set_*primitives and their Bigstring equivalents have been merged into primitives taking a size argument (and a variant argument for safety instead of a boolean). - The primitives
Pidentity,Pbytes_to_string,Pbytes_of_string,Pctconst,Pignore,Prevapply,Pdirapply,PgetglobalandPsetglobalhave been removed. A new primitivePread_symbolis added to replacePgetglobal, the others are expected to be dealt with when translating from Lambda (most were already dealt with anyway).
The rest are mostly mechanical changes induced by the change in types.
This pull request is rather important for simplifying other pull requests on the backend, so it would be helpful to get it merged early in the 4.09 cycle.
36a4d05 to
80ad15d
Compare
|
Moved |
|
@alainfrisch Do you have any objection ? Otherwise I think this is good to go. |
|
I haven't reviewed the PR, but I have no objection in principle. |
|
This has already been reviewed, that was the kind of non objection I expected. |
|
Currently the Change entry for this GPR says: Add a separate types for clambda primitives Is there onlyone type of are there several of them? Could the change entry be fixed accordingly, please? |
|
ping @chambart |
* Start of implicit-source-positions Expand [%src_pos] to hard-coded record (ocaml#1565) * Update codeowners * Revert "Update codeowners" This reverts commit 313cb2d1415a34971ea6e3cbd02d008e255ca888. * Start * backup * Hacky checkpoint with hard-coded record * Rename test * Shadowing works as intended * Simpler test case * Fix label layouts * Change name to lexing_position, fix record layout, add to shadowing test * Add to test * Formatting * Tests, fix predef lexing_position layout, add CR for untypeast shadowing * Fix test * Fix untracked file * Revert "Fix untracked file" This reverts commit 2a18a6aef66322e4e897aef29ace14eda25b7785. * reformat * Bootstrap, with predef lexing_position --------- Co-authored-by: Richard Eisenberg <reisenberg@janestreet.com> Separate `arg_label`s for Parse and Typed trees (ocaml#1589) * src_pos in fn decls and defns * Revert is_optional modifications for now * src_pos shouldn't work in arbitrary places * Tidy code * Whitespace * Add test * Comment on documentation * Whitespace * more whitespace * Amend test to use Position * Correctly print [%src_pos] rather than lexing_position and fix corresponding test * Store hard-coded string in variable * Separate parsed and typed labels * Delete fn_decl_and_defn.ml * Delete invalid_usages.ml * Minimize diffs in PR * Minimize diffs in PR * Update named_function.ml for less diffs * Clean up * Put arg_label in Outcome tree * Revert "Put arg_label in Outcome tree" This reverts commit cd5ace530c602975bac5f259e7d078a52dfabdaa. * is_optional for Parsetree arg_labels * Restore Asttypes.arg_label, re-export arg_label from Types * Document distinction between arg_labels * Fix whitespace * Add todo comment Add `arg_label` to `Otyp_arrow` (ocaml#1644) * Put arg_label in Outcome tree * Add comment for Outcometree.arg_label `[%src_pos]` in fn decls and defns (ocaml#1579) * src_pos in fn decls and defns * Revert is_optional modifications for now * src_pos shouldn't work in arbitrary places * Tidy code * Whitespace * Add test * Comment on documentation * Whitespace * more whitespace * Amend test to use Position * Correctly print [%src_pos] rather than lexing_position and fix corresponding test * Store hard-coded string in variable * Hacky outcometree printing without creating nontrivial node * Update documentation for Position labels * Update tests * Fix erroneously printing lexing_position for Position arguments * Clean up code * Reconstruct constraint for Position arguments in Untypeast * Update tests * Move comments around * Slightly prettify * Use Location.none in Untypeast reconstructed extensions * Add Ttyp_src_pos to Typedtree * Add CR * Comments * Clean up code * Update comment in printtyp.ml * Add documentation for Ttyp_src_pos, remove extraneous comments * *synonyms.ml * src_pos in fn decls and defns * Revert is_optional modifications for now * src_pos shouldn't work in arbitrary places * Tidy code * Whitespace * Add test * Comment on documentation * Whitespace * Correctly print [%src_pos] rather than lexing_position and fix corresponding test * Store hard-coded string in variable * Hacky outcometree printing without creating nontrivial node * Update documentation for Position labels * Update tests * Fix erroneously printing lexing_position for Position arguments * Clean up code * Reconstruct constraint for Position arguments in Untypeast * Update tests * Move comments around * Slightly prettify * Use Location.none in Untypeast reconstructed extensions * Add Ttyp_src_pos to Typedtree * Add CR * Comments * Clean up code * Add documentation for Ttyp_src_pos, remove extraneous comments * *synonyms.ml * Merge cleanup * Add label to Octy_arrow * Rename function * transl_label checks for Position * Pass None to transl_label for classes * Add comment * Delete comment * Consider Position when approximating type * Add tests for application, recursion * Error messages mention Position * Prettify * Comments * Rename function * Add comment * Remove extraneous calls to label translation * Test type-based disambiguation * Add comment for fn app labels * Add commuting tests * Remove duplicated src_pos match logic * Reduce instances of src_pos string * src_pos in fn decls and defns * src_pos shouldn't work in arbitrary places * Whitespace * Correctly print [%src_pos] rather than lexing_position and fix corresponding test * Store hard-coded string in variable * Hacky outcometree printing without creating nontrivial node * Clean up code * Slightly prettify * Add CR * Clean up code * Update comment in printtyp.ml * Merge cleanup * Add label to Octy_arrow * Rename function * transl_label checks for Position * Pass None to transl_label for classes * Add comment * Delete comment * Consider Position when approximating type * Add tests for application, recursion * Error messages mention Position * Prettify * Comments * Rename function * Add comment * Remove extraneous calls to label translation * Test type-based disambiguation * Add comment for fn app labels * Add commuting tests * Remove duplicated src_pos match logic * Reduce instances of src_pos string * Make things is_optional again * Add commuting tests * Add transl_label_from_pat * Whitespace * Parenthesize src_pos in error message * Fix test Create `Lexing.position` values when `[%src_pos]` in expressions (ocaml#1661) * Everything * checkpoint Construct lambda for src_pos directly * Revert now unneeded changes to predef * Clean up comments, whitespace Implicitly supply source position arguments (ocaml#1671) * Everything * Apply position arguments when expected type is nothing, refactor creating src_pos exprs * Add warning instead of modifying existing one * Fix test * Move test * Resolved comments Clearer classic mode label equivalence checks (ocaml#1692) * Everything * Apply position arguments when expected type is nothing, refactor creating src_pos exprs * Add warning instead of modifying existing one * Fix test * Move test * Resolved comments * Classic mode equivalence check * Refactor * Implicit source position merge conflicts (ocaml#2275) * Implicit Source Positions Conflict resolution This feature solely fixes merge conflicts that the implicit source positions project had after not having been rebased since the summer. Sadly, I mistakenly committed a new test! I meant to do it on a different branch, but I accidentally committed some more changes after that making it trickier to split... Please let me know if I should only keep this feature do "merge conflict resolution". Thanks! Testing ------- - I ran `make test`, and it passes. - `make install` works, I migrated bonsai to start using this and it works! - Added test on let operators. - Sadly it doesn't quite work, but I think it would be cool if it did work as it would allow codebases that use let+ and let* to get source code locations! I am unsure about its sound-ness though and left CR src_pos'es/question CR's. Please let me know if there is other additional testing/ci I should perform. Thanks! I also performed additional cleanup in this feature from self-questions I had during the merge. Please let me know if there are any changes I should perform. Thanks! * More cleanup after self-review * Made CI on Github pass - CI seems to run `make ci` instead of just `make test`. I've fixed more `make ci` changes, although not all, pushing to let ci run in case the current "compiler-libs.common" being missing is due to a misconfiguration on my environment/if it also fails on the github ci being missing is due to a misconfiguration on my environment/if it also fails on the github ci. - Performed a bootstrap - Fixed chamelon - Additionally turned my questions on let operator support into CR src_pos: * Removed empty file added accidentally * Updated let_operator support comment * Widened question on chamelon compatibility * Implicit source positions object system support (ocaml#2307) * Implicit Source Positions Conflict resolution This feature solely fixes merge conflicts that the implicit source positions project had after not having been rebased since the summer. Sadly, I mistakenly committed a new test! I meant to do it on a different branch, but I accidentally committed some more changes after that making it trickier to split... Please let me know if I should only keep this feature do "merge conflict resolution". Thanks! Testing ------- - I ran `make test`, and it passes. - `make install` works, I migrated bonsai to start using this and it works! - Added test on let operators. - Sadly it doesn't quite work, but I think it would be cool if it did work as it would allow codebases that use let+ and let* to get source code locations! I am unsure about its sound-ness though and left CR src_pos'es/question CR's. Please let me know if there is other additional testing/ci I should perform. Thanks! I also performed additional cleanup in this feature from self-questions I had during the merge. Please let me know if there are any changes I should perform. Thanks! * Moves changes from original class-type support branch into a rebased branch * Removed lingering merge conflict markers * More tests + manually moved a commit from the original class type branch - For some reason I originally missed a commit that typed the argument on classes from the original branch. This feature also grabs the tests. Some of my questions revolving `Principal` are no longer needed as they seem to have disappeared! I suspect that `make ci` now passing in the parent feature is partly/transitively responsible somehow for `Principal` now no longer showing up. * Fixed incorrect merging of invalid_usages.ml * Removes weird whitespace observed after self-review * More tests! Found out that application on an inheritance call is unhandled! * Explicit passing positional argument in a pcl_apply works, erasure still does not. * Added Pcl_apply support * More cleanup + removed a question cr from the parent feature. * More tests. Found another bug! Class type arrows seeem to be left untranslated... * Fixed type annotation bug + added more tests * Removed duplicated test * Added more tests + fixed weird whitespace * Added question on the two class system environments (val_env vs. met_env) * Deduplicated more code after self-review * minor whitespace change * Removed resolved question CR and addressed new CRs * Implicit source positions directory locations (ocaml#2346) * Implicit Source Positions Conflict resolution This feature solely fixes merge conflicts that the implicit source positions project had after not having been rebased since the summer. Sadly, I mistakenly committed a new test! I meant to do it on a different branch, but I accidentally committed some more changes after that making it trickier to split... Please let me know if I should only keep this feature do "merge conflict resolution". Thanks! Testing ------- - I ran `make test`, and it passes. - `make install` works, I migrated bonsai to start using this and it works! - Added test on let operators. - Sadly it doesn't quite work, but I think it would be cool if it did work as it would allow codebases that use let+ and let* to get source code locations! I am unsure about its sound-ness though and left CR src_pos'es/question CR's. Please let me know if there is other additional testing/ci I should perform. Thanks! I also performed additional cleanup in this feature from self-questions I had during the merge. Please let me know if there are any changes I should perform. Thanks! * More cleanup after self-review * Made CI on Github pass - CI seems to run `make ci` instead of just `make test`. I've fixed more `make ci` changes, although not all, pushing to let ci run in case the current "compiler-libs.common" being missing is due to a misconfiguration on my environment/if it also fails on the github ci being missing is due to a misconfiguration on my environment/if it also fails on the github ci. - Performed a bootstrap - Fixed chamelon - Additionally turned my questions on let operator support into CR src_pos: * Removed empty file added accidentally * Updated let_operator support comment * Widened question on chamelon compatibility * Moves changes from original class-type support branch into a rebased branch * Created branch with working changes for directory positions. * Added a test sanity checking being able to pass in the flag. * Changed test to properly test that the right basename is supplied * Removed merge conflict markers upon self review. * Removed lingering merge artifacts upon self review * Renamed flag from -dir to -directory * Fix typo * Removed directory flag from ocamldoc options * [Implicit Source Positions] - Better Error Messages (ocaml#2364) * Rename src_pos -> call_pos Also left some self notes regarding the remaining CR src_pos * Improved error messages for %call_pos * Addressed a CR src_pos and removed an already addressed CR src_pos * Cleanup after self-review * Updated missing rename after self-review * Removed addressed c r questions * Fixed merge conflicts. `make ci` passes locally * bootstrap --------- Co-authored-by: jose r <45022810+Enoumy@users.noreply.github.com> Co-authored-by: enoumy <enoumy@gmail.com>
This patch defines a new
Clambda_primitivemodule containing the primitive for Flambda and Clambda. The aim of this patch is to allow to refactor the types to remove the parts that are meaningless in the backend.This does not yet contain a lot of the refactoring itself.
Note that this will be required by the future developments of flambda which includes a specific type for primitives in flambda that can't really be represented into the current primitive type of lambda without adding a lot of new cases useless for the bytecode implementation.