Add a check for pre-normalize'd paths, to relativize

[dwijnand]

Fixes sbt/sbt#3286
Fixes sbt/sbt#3276

[eed3si9n]

Could you explain how this fixes sbt/sbt#3286?

[eed3si9n]

Is this happening because JIO Path is unable to handle . and .git correctly?
What if you have ../foo and .git? Don't they start the same?

1. It might make it easier for me to understand if the toPath and if expression part is factored out to an inner function named sharesPath.
2. After resolving the symlinks, maybe you should convert it to an absolute URI instead?

[dwijnand]

Of course.

FileExamples.files (link) lists the child paths of a given base directory and then relativizes in terms of that base directory:

val directory: File = ...
val childPaths = IO.listFiles(directory).toStream
val prefixedDirectChildPaths = childPaths map { IO.relativize(base, _).get } filter { _ startsWith prefix }

Because of the fundamental change to use getCanonicalPath instead of getAbsolutePath in sbt 1's IO in #18 (with subsequent fixes #33 and #34) using a base directory of "." doesn't relative its children:

Welcome to Scala 2.12.2 (Java HotSpot(TM) 64-Bit Server VM, Java 1.8.0_131).
Type in expressions for evaluation. Or try :help.

scala> val base = new java.io.File(".")
base: java.io.File = .

scala> val file = new java.io.File("./.git")
file: java.io.File = ./.git

scala> val basePath = base.toPath
basePath: java.nio.file.Path = .

scala> val filePath = file.toPath
filePath: java.nio.file.Path = ./.git

scala> filePath.normalize() startsWith basePath.normalize()
res0: Boolean = false

scala> basePath.normalize()
res1: java.nio.file.Path =

scala> filePath.normalize()
res2: java.nio.file.Path = .git

scala> basePath relativize filePath
res3: java.nio.file.Path = .git

So this change fixes the downstream issue, without breaking any of the existing regression tests for #18/#33/#34.

[dwijnand]

@eed3si9n I'm happy to iterate on a better fix for this, but I'm keen for this less-than-ideal fix to ship in RC1, please.

[eed3si9n]

I was concerned about correctness issue, but I can't think of counter example foo and foobar, so I guess we can merge.

[dwijnand]

I share your concern. This fix isn't satisfying in the least. But using it downstreams seems to solve the reported reproductions. Thanks for the merge.