Keep generated Dict and SeqDict migrations valid when nested

[CharlonTank]

Problem

Two of the Dict / SeqDict migration templates are only valid at the top level of a field migration:

1. Value migrations are emitted as Dict.map (\k -> <m>). When the value is itself a Dict/SeqDict that needs a migration, the output is

   old.x |> Dict.map (\k -> Dict.map (\k -> migrate_UserType))

   which Elm rejects with a SHADOWING error. The existing Migrate_All/Expected.elm snapshot already contains exactly this (nestedDictCustomType); it just never gets compiled.

2. Key migrations are emitted as a pipeline fragment, SeqDict.toList |> List.map (…) |> SeqDict.fromList. That works after old.field |>, but as soon as it is nested in a function position (Dict.map (\k -> …), Maybe.map (…), …) it is no longer a function. For example, Dict String (SeqDict CustomKey v) generates a migration that fails to type check.

We hit both in almost every production migration of our app that touches such a field, and had to fix them by hand each time.

Fix

• Bind \_ instead of \k. The key is never used, so nesting can no longer shadow anything.
• Compose with >> instead of piping with |>, as the Result template already does. At the top level, old.x |> SeqDict.toList >> List.map (…) >> SeqDict.fromList stays valid because >> binds tighter than |>.
• Parenthesise the Tuple.mapFirst argument, as Tuple.mapBoth already does, so a compound key migration stays a single argument. elm-format removes the parentheses when they are redundant.

Tests

• The Migrate_All scenario gets a new field, nestedSeqDictCustomKey : Dict String (SeqDict UserType Int), which adds lamdera/containers to the scenario's elm.json. The snapshots are regenerated.
• migration-scenarios is pending, so I ran it locally with testExamples on src/Migrate_All only: ✅ 2 tests passed. The Migrate_External_Paramed scenario currently throws first, on a Wire3 encoder type error while compiling the scenario itself, which is unrelated to this change.
• I compiled the regenerated Expected.elm with lamdera make. Before the change it fails with SHADOWING on nestedDictCustomType. After it, the only errors left are the intentional Unimplemented placeholders.

Related

• fix: use old type name in Union migration type signatures for renamed types #81 (also mine) fixes the second problem in a different place. Its second commit (a4605ae) wraps a raw pipeline migration in (\v__ -> v__ |> …) when it is passed as an argument. This PR fixes it at the source instead: the templates no longer emit |> pipelines, so that wrapper no longer triggers for them. Composition binds no name, so it stays valid at any depth, whereas a \v__ wrapper nested inside another one would shadow it. fix: use old type name in Union migration type signatures for renamed types #81 does not touch the first problem (\k). If both land, only the Migrate_Nested_Dict snapshot of fix: use old type name in Union migration type signatures for renamed types #81 needs regenerating.
• Add compiler support for BiSeqDict, MultiSeqDict, and MultiBiSeqDict #69 adds BiSeqDict, MultiSeqDict and MultiBiSeqDict with copies of these templates; it now gets the same fix (6cc5f58).