Improve Aggregation

type:none

This MR improves a few this in the Aggregation pass :

• Use of add_prefix and internal_get_name_and_counter which should be avoided.
• Scope data are stored in a map for faster lock up
• Deduplication of all binder inside a module is changed by the deduplication of module binder before aggregating
• The current pass works in two step: Ast_typed -> list of locally defined decl -> Ast_aggregated. The new pass work in one step : Ast_typed to Ast_aggregated (and a preprocessing before). Therefore all functions are direct mappers.
• Expose function that doesn't take the local module as parameters.
• Separate the deduplication in an independent self-pass for redability.
• Homogenisation with the other pass

Side-effect : this change triggered a bug in the monomorphisation self-pass that is solved in the MR.

mentioned in merge request !1490 (closed)

assigned to @pewulfman

marked this merge request as draft

mentioned in merge request !1524 (merged)

added 121 commits

• cef10b6a...1cb6ddd0 - 104 commits from branch dev
• 77329f5f - add type_abstraction
• f871edc7 - wip:change cameligo abstraction; problem in typechecking
• 5c9bfafb - Fix type-checking & promote
• d67b05a6 - move detection of free polymorphic variable to the abstractors
• 0da7380f - remove dead code
• 934b348c - Move non-annotated test case to positives
• 46a09e00 - Fix regression
• c251a7b8 - fix some regression
• 826b4e25 - add test for big lambda
• f3f9233f - add self-ast-core pass
• 649d222f - add changelog
• 9ba6a91d - make variable uncompatible
• ab1cfca1 - change representation of generated variable
• 80b80634 - PP_annotated
• 6f764cc7 - remove trailling error
• ffefe1ee - add regression test
• 91696cae - aggregation done; remove_aliases break shadowing

Compare with previous version

I can really explain why, but I got the impression that the current aggregation pass is too complex.

Possibly, but I will believe it when I see it.. Right now all the matchings tests (where a lot of shadowing is happening) are breaking. Hint:

• All the generated variable here should be fresh, but when I make them fresh everything breaks big time
• I feel the module/record accesses case is the source of this

Also, it removes module aliases, which i don't think the pass should do.

I believe you meant that aggregation was not replacing module aliases by a let aliase ? if yes, we can agree on that. But then I will argue against the remove_alias self pass?

However, the monomorphisation pass doesn't work when there is too much indirection between the variable and the definition of the polymorphic function (or maybe I just don't propagate the types correctly)

That sounds extremely fishy. Whatever a pass does, it should preserve the soundness of the program it runs on. Your aggregation pass introduce type inconsistency that you are then partially " "fixing" " in remove_alias

added 1744 commits

• 91696cae...df561552 - 1740 commits from branch dev
• f9bcbc4a - PP_annotated
• 2ecbc962 - remove trailling error
• eee4127b - add regression test
• bbdbb49e - aggregation done; remove_aliases break shadowing

Compare with previous version

added 98 commits

• bbdbb49e...af55e551 - 93 commits from branch dev
• 3ae3f967 - PP_annotated
• 57589514 - remove trailling error
• 47e68789 - add regression test
• f501b23c - aggregation done; remove_aliases break shadowing
• a08872e8 - don't succeed. Redesign

Compare with previous version

added 277 commits

• a08872e8...f901f3f3 - 271 commits from branch dev
• d4fafd79 - PP_annotated
• 8d51be59 - remove trailling error
• 5dfdddba - add regression test
• 11337141 - aggregation done; remove_aliases break shadowing
• 3c091ebf - don't succeed. Redesign
• daae5c91 - wip

Compare with previous version

added 1 commit

• ef137ccb - reraise error

Compare with previous version

added 109 commits

• ef137ccb...10d954e1 - 102 commits from branch dev
• f1b5d630 - PP_annotated
• b806ec04 - remove trailling error
• fe957a41 - add regression test
• 0b5f13ae - aggregation done; remove_aliases break shadowing
• 41cba007 - don't succeed. Redesign
• dad63d5f - wip
• 9b81c8ed - reraise error

Compare with previous version

added 1 commit

• db2b7e1b - fix bug due to incorrect library

Compare with previous version

added 1 commit

• 3c4fd3c8 - fix bug due to incorrect library

Compare with previous version

changed the description

added 1 commit

• 19c739c6 - clean up monomorphisation

Compare with previous version

added 1 commit

• ad1f1ac9 - clean up monomorphisation

Compare with previous version

added 120 commits

• ad1f1ac9...2157dbd0 - 111 commits from branch dev
• 0041dc5b - PP_annotated
• b45d629e - remove trailling error
• b6c080af - add regression test
• 978842e9 - aggregation done; remove_aliases break shadowing
• f5226067 - don't succeed. Redesign
• 2fb0da9b - wip
• f6e2b636 - reraise error
• adeee189 - fix bug due to incorrect library
• 6c5009c9 - clean up monomorphisation

Compare with previous version

added 69 commits

• 6c5009c9...d5fd8ad6 - 60 commits from branch dev
• 85e63111 - PP_annotated
• 9a26a903 - remove trailling error
• 7fab3192 - add regression test
• b4000123 - aggregation done; remove_aliases break shadowing
• c5e68465 - don't succeed. Redesign
• 49fe1db6 - wip
• f5cd2ba2 - reraise error
• b16c84ff - fix bug due to incorrect library
• 0b20d323 - clean up monomorphisation

Compare with previous version

added 1 commit

• d04adfd2 - solve several bug. Still curious bug

Compare with previous version