Fix normalized URI generation for UNC path

[Friendseeker]

Issue

IO.directoryURI and Path.absolute both calls toURI.normalize to convert a java.io.File to a normalized URI, which generates invalid URI for UNC path. As a result, sbt crashes when we try to open a project in a WSL directory.

Example: say we have a dir: java.io.File representing Ubuntu-24.04/home/jerrytan/testwsl/ in WSL, then calling dir.toPath.toUri.normalize yields file:/wsl$/Ubuntu-24.04/home/jerrytan/testwsl/ , which then leads to

java.lang.RuntimeException: Invalid build URI (no handler available): file:/wsl$/OracleLinux_9/home/user/project/
        at scala.sys.package$.error(package.scala:30)
        at sbt.internal.Load$.$anonfun$builtinLoader$1(Load.scala:489)
        at sbt.internal.BuildLoader.$anonfun$apply$3(BuildLoader.scala:244)
        at scala.Option.getOrElse(Option.scala:189)
        at sbt.internal.BuildLoader.apply(BuildLoader.scala:244)
        at sbt.internal.Load$.loadURI$1(Load.scala:554)
        at sbt.internal.Load$.loadAll(Load.scala:570)
        at sbt.internal.Load$.loadURI(Load.scala:500)
        at sbt.internal.Load$.load(Load.scala:479)

Fix

Call toPath.normalize.toUri instead. For above example, it yields file:////wsl$/Ubuntu-24.04/home/jerrytan/testwsl/ which can be loaded by sbt. Note that the new file URL still do not follow the best practice as documented in Eugene's blog post, but it indeed do not crash.

Closes sbt/sbt#7135

Credit to reporter of sbt/sbt#7135 for detailed investigation of the issue.

[eed3si9n]

Could we add unit test for this plz?

[eed3si9n]

The creation of the first URI object seems wasteful here. See also #132, which states that URI object can lead to hundreds of MB in heap.

[Friendseeker]

Shall investigate this once I regain access to a Windows Machine.

Indeed there is something fishy there. If dirURI follows u2 notation, then calling dirURI.normalize should be correct. While I think the other changes in the PR makes sense, this particular change is just a bandaid that somehow fixes the sbt crash without fixing the underlying issue.

I now actually suspect the hidden issue might be with IO.toURL, which is called by ProjectRef.apply on sbt side. This is just a hypothesis though.

[Friendseeker]

Btw going slightly off topic but your blogpost about file URI schema is so helpful! Without it I probably need to spend a lot of time researching about file URI.

[Friendseeker]

I must say this is more complex than I anticipated...

The PR, as of its current state, is inadequate.

The issue is that, the dollar sign actually matters when parsing a URI. Probably since $ is a reserved character.

val path = "\\\\wsl$\\Ubuntu-24.04\\home\\jerrytan\\testwsl"
val wslDir = new File(path)
println(wslDir.toPath.toUri)

// Output: file:////wsl$/Ubuntu-24.04/home/jerrytan/testwsl/

val path = "\\\\wsl.localhost\\Ubuntu-24.04\\home\\jerrytan\\testwsl"
val wslDir = new File(path)
println(wslDir.toPath.toUri)

// Output: file://wsl.localhost/Ubuntu-24.04/home/jerrytan/testwsl/

The u4 URI is fragile. After calling .normalize() on it, it becomes file:/wsl.localhost/Ubuntu-24.04/home/jerrytan/testwsl/ which sbt cannot parse. The right conversion is to convert it to u2, which does not break (remains u2) after normalization.

So here's the current state:

• Path.toURI generates invalid URI for UNC path containing $ sign
• IO.toFile is correct, other than the fact that it does not append "/" to the end of URI when file is directory
• 'IO.directoryURI(dir: File)` is broken.
• IO.directoryURI(uri: URI) is probably correct, as long as the uri follows spec.