scalameta 4.9.x

[bjaglin]

https://github.com/scalameta/scalameta/releases/tag/v4.9.0

scalameta/scalameta#3425 is a non-backward compatible feature, as quasiquotes expand to code referencing scala.meta.trees.Origin which does not exist in 4.8.x.

This means that Scalafix rules containing quasiquotes compiled against scalameta 4.9.x will cause link errors if ran on older Scalafix versions. As a result, we need a minor Scalafix bump when taking that in, despite the fact that it's "just" a minor bump in a SemVer lib (we could argue that a major bump of scalameta was needed, but backward compatibility guarantees with macros is not very well defined).

Edit after v0.12.0 release

I originally warned rule authors of this in the release notes:

• Changes breaking forward binary compatibility
• Usage of quasiquotes for tree traversal or pretty-printing in rules built against Scalafix 0.12.x will result in runtime errors when executed with Scalafix 0.11.1 or earlier (quasiquotes: preserve origin or dialect if possible scalameta/scalameta#3425)
  • Rules using quasiquotes and built with Scalafix 0.12.x should signal this via a major bump
  • Users of Scalafix 0.11.1 or earlier loading such rules will get a warning as documented

But this turns out NOT to be an issue with old scalafix-cli, as scala.meta.trees.Origin is a brand new symbol, properly resolved by the tool classpath URLClassLoader which contains scalameta 4.9.x (which the parent scalafix-cli classloader containing scalameta 4.8.x reverts to).

[bjaglin]

6d8b555 causes core compilation to succeed, but docs2_13 to fail

[error] Caused by: java.lang.ClassNotFoundException: scala.meta.trees.Origin

That's because for some reason, coursier prefers 4.8.15 over 4.8.15+150-a7c0baa0-SNAPSHOT there:

sbt:scalafix> show docs2_13/evicted
...
[info] 	* org.scalameta:scalameta_2.13:4.8.15 is selected over {4.8.15+150-a7c0baa0-SNAPSHOT, 4.8.15+150-a7c0baa0-SNAPSHOT, 4.8.15+150-a7c0baa0-SNAPSHOT, 4.8.15+150-a7c0baa0-SNAPSHOT
[info] 	    +- org.scalameta:mdoc_2.13:2.5.2                      (depends on 4.8.15)
[info] 	    +- org.scalameta:mdoc-cli_2.13:2.5.2                  (depends on 4.8.15)
[info] 	    +- org.scalameta:semanticdb-scalac-core_2.13.12:4.8.15+150-a7c0baa0-SNAPSHOT (depends on 4.8.15+150-a7c0baa0-SNAPSHOT)
[info] 	    +- ch.epfl.scala:scalafix-core_2.13:0.11.1+108-789b602c-SNAPSHOT (depends on 4.8.15+150-a7c0baa0-SNAPSHOT)

Adding scalameta at the root does affect resolution, so it looks like Coursier is following nearest-win instead of latest-win (despite what's expected).

 lazy val docs = projectMatrix
   .in(file("scalafix-docs"))
   .settings(
+    libraryDependencies += scalameta,

sbt:scalafix> show docs2_13/evicted
...
[info] 	* org.scalameta:scalameta_2.13:4.8.15+150-a7c0baa0-SNAPSHOT is selected over {4.8.15, 4.8.15}
[info] 	    +- ch.epfl.scala:scalafix-docs_2.13:0.11.1+108-789b602c+20240211-1152-SNAPSHOT (depends on 4.8.15+150-a7c0baa0-SNAPSHOT)
[info] 	    +- ch.epfl.scala:scalafix-core_2.13:0.11.1+108-789b602c+20240211-1152-SNAPSHOT (depends on 4.8.15+150-a7c0baa0-SNAPSHOT)
[info] 	    +- org.scalameta:semanticdb-scalac-core_2.13.12:4.8.15+150-a7c0baa0-SNAPSHOT (depends on 4.8.15+150-a7c0baa0-SNAPSHOT)
[info] 	    +- org.scalameta:mdoc_2.13:2.5.2                      (depends on 4.8.15)
[info] 	    +- org.scalameta:mdoc-cli_2.13:2.5.2                  (depends on 4.8.15)

Forcing eviction via ea2e835

[bjaglin]

ea2e835 exposes an incompatibility between mdoc and the latest scalameta SNAPSHOT

[error] Caused by: java.lang.ClassNotFoundException: scala.meta.internal.prettyprinters.enquote$

Following-up in scalameta/mdoc#842

[bjaglin]

[info] warning: patch.md:168:22: local val dialectOnly$macro$4 in value qual$12 is never used
[info] Patch.removeImportee(importee"Future").showDiff()
[info]                      ^^^^^^^^

Following-up in scalameta/scalameta#3546

[bjaglin]

[info] error: patch.md:168:1: token not found: 
[info] Patch.removeImportee(importee"Future").showDiff()
[info] ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
[info] java.util.NoSuchElementException: token not found: 
[info] 	at scalafix.util.TokenList.$anonfun$tok2idx$2(TokenList.scala:27)
[info] 	at scala.collection.immutable.Map$WithDefault.default(Map.scala:181)
[info] 	at scala.collection.MapOps.apply(Map.scala:176)
[info] 	at scala.collection.MapOps.apply$(Map.scala:175)
[info] 	at scala.collection.AbstractMap.apply(Map.scala:405)
[info] 	at scalafix.util.TokenList.leading(TokenList.scala:15)
[info] 	at scalafix.internal.patch.ImportPatchOps$.remove$1(ImportPatchOps.scala:229)
[info] 	at scalafix.internal.patch.ImportPatchOps$.$anonfun$superNaiveImportPatchToTokenPatchConverter$34(ImportPatchOps.scala:264)
[info] 	at scala.collection.StrictOptimizedIterableOps.map(StrictOptimizedIterableOps.scala:100)
[info] 	at scala.collection.StrictOptimizedIterableOps.map$(StrictOptimizedIterableOps.scala:87)
[info] 	at scala.collection.mutable.LinkedHashSet.map(LinkedHashSet.scala:34)
[info] 	at scalafix.internal.patch.ImportPatchOps$.superNaiveImportPatchToTokenPatchConverter(ImportPatchOps.scala:264)
[info] 	at scalafix.internal.patch.PatchInternals$.treePatchApply(PatchInternals.scala:118)

scalafix/scalafix-core/src/main/scala/scalafix/internal/patch/ImportPatchOps.scala

Line 203 in 61dce03

if (toRemove.pos == Position.None) return Patch.empty

brings false negatives for quasiquotes-built trees since scalameta/scalameta#3450, resulting in an exception instead of the documented no-op.

Another similar guard is impacted, although it causes a slight regression that might not be worth fixing:

scalafix/scalafix-core/src/main/scala/scalafix/patch/Patch.scala

Line 103 in 61dce03

if (from.pos == Position.None) Patch.empty

• Patch.replaceTree(q"foo", "bar") is no-longer a no-op as the early-return no longer applies now that quasiquotes generate trees with positions
  • It is now the same as Patch.addLeft(q"foo", "bar") or Patch.addLeft(Term.Name("foo"), "bar") (unchanged behavior for them): it adds a leading string to any tree, as the BOF token has the same hash no matter the origin, causing a false positive
• Patch.replaceTree(Term.Name("foo"), "bar") remains a no-op thanks to the early-return

[bjaglin]

unused since 1ab1d7e

[bjaglin]

This brings false negatives now that quasiquotes have positions (since scalameta/scalameta#3425). This is effectively a very minor breaking change in the API.