Merge 5.2.0minus-5 #124
Merge 5.2.0minus-5 #124
Conversation
liam923
force-pushed
the
merge-5.2.0minus-5
branch
from
14d1873 to
2678281
Compare
|
I'm a bit confused by the CI failure. The same test passes when I run locally, and I haven't yet been able to reproduce. This test has been known to be flaky in the past, so I re-ran it twice, but there's two test cases that have consistently failed across all three runs. I'm tempted to comment out those tests for now so that they don't block rolling Merlin (the justification being that they were already flaky before). |
|
Those tests shouldn't be flaky -- they're an output of type-checking, which should be deterministic. That said, I'm not shocked that this has changed as we've evolved the type checked. The tests are looking to see where we allocate a cell that never gets used, which is a bit unusual. I think it's fine to accept the new output. But I wouldn't comment out the tests! |
| @@ -240,7 +240,7 @@ Syntax errors also shouldn't escape: | |||
| "line": 1, | |||
| "col": 41 | |||
| }, | |||
| "type": "parser", | |||
| "type": "typer", | |||
There was a problem hiding this comment.
Is there a good reason for this change?
| :: (_, node) | ||
| :: _parents -> | ||
| let loc = Mbrowse.node_loc node_for_loc in | ||
| (loc, Construct.node ~config ~keywords ?depth ~values_scope node) | ||
| | (_, (Browse_raw.Expression { exp_desc = Texp_hole; _ } as node)) | ||
| | ( _, | ||
| (Browse_raw.Expression { exp_desc = Texp_typed_hole | Texp_hole _; _ } |
There was a problem hiding this comment.
Does this really want Texp_hole? That surprises me a bit.
| let lbl = Env.find_label_by_name ident env in | ||
| let module Pack = struct | ||
| (* This gadt is used to mint the existential type [rep] *) | ||
| type t = P : 'rep Types.gen_label_description -> t |
There was a problem hiding this comment.
Why can't you use packed_label_description defined above?
| | `Description lbls -> | ||
| | No -> [] | ||
| | Maybe -> | ||
| Env.fold_labels Legacy add_label_description prefix_path env [] |
There was a problem hiding this comment.
Does it make sense to make a function the folds over both flavors of label? That will make it less likely to skip one flavor.
Merge flambda changes for 5.2.0minus-5. The first two commits run the import script and commit merge conflicts - I recommend skipping them when reviewing.
This PR was difficult for two reasons:
overwrite (3,4) with (_, 5). New nodes are added to the parsetree and typedtree:Pexp_holeandTexp_hole, respectively. This partially conflicts with Merlin's idea of typed holes. Before, Merlin used an extension node to represent holes in the parsetree, and in the typedtree, it also added aTexp_holeconstructor. I resolve this by using the newPexp_holein place of an extension node (soPexp_holerepresents either type of hole) and renaming Merlin'sTexp_holetoTexp_typed_hole(thus, there is now bothTexp_holeandTexp_typed_holein Merlin). I've submitted a PR to upstream Merlin to renameTexp_holetoTexp_typed_holeto mitigate conflicts in the future. (I also renamedTmod_holetoTmod_typed_holefor consistency - I included this in the upstream PR as well.)