An adjustment is required.

scala> println:
     |   42
     |
42

scala> List(42).map: x =>
     |   x + 1
     |
val res0: List[Int] = List(43)

scala> List(42).map: x =>
-- Error: ----------------------------------------------------------------------
1 |List(42).map: x =>
  |^^^^^^^^^^^^^^^
  |not a legal formal parameter for a function literal

scala>

where res0 was from history, and the error shows "arrow eol" is still broken.

Also the scripted test does not actually pass.

In incomplete state, REPL re-parses on every character (IIUC); I expected to follow up with other improvements in that area.

Dotty ReplTest does only single-line inputs; at least "scripted" test should work like Scala 2 SessionTest.

The gold standard integration test would run a REPL process from test input, in order to test incomplete input:

List(42).map: x =>// at eof
List(42).map: x =>\n// at eof
List(42).map: x =>\n  x+1// at eof
List(42).map: x =>\n  x+1\n\n // complete

The JLine parser could be tested for internal behavior, but scripted or "session" tests are easier to create and read.

The failure was the flaky cats test. Of course, dog people know that all cats are flaky.

@mbovel @hamzaremmal Looks like this is ready to review, anyone could take a look?

Yes, I can have a look tomorrow 👍

Let me try to summarize.

Consider the following snippet without a trailing line break (i.e., the last three tokens are NEWLINE, SPACE, and EOF):

object Foo:
••val foo = 42
•

1. Following No outdent at eof #22435, no DEDENT token is emitted for the last line. That avoids emitting an indentation error, which is desirable since the user might be writing text on the last line and might not be finished yet (a space on the last line of the file causes a compilation error #22332). The test tests/pos/i22332.scala demonstrates that.

2. However, we still want to emit a DEDENT token if the last line is completely empty (i.e., the last two tokens are NEWLINE followed by EOF), otherwise the REPL never considers the input complete (optional braces no longer work in repl #22844). This is what this PR achieves. The new test i22844 regression colon eol in dotty.tools.repl.ReplCompilerTests demonstrates that second situation.

Is my understanding correct? If yes, then the implementation looks correct to me! 🙂

Out of curiosity, I also checked what happens when a non-empty blank line is not at the end—for example, when the trailing four tokens are NEWLINE, SPACE, NEWLINE, and EOF, i.e., the last line is empty but the one before has a space. This also doesn't report an error. handleNewLine is not even entered for the line with the space in that case. This seems coherent 👍

Thanks for your review effort! I will also try to understand my code and clean it up.

I'm not wedded to the first commit, which I see as scouting. I think "match-aligned case" is the future, but the future doesn't have to be now.

Ah yes, to rephrase, the additional condition !isTrailingBlankLine says the last line must not match \s+ for outdent.

The change to observeArrowIndented says no indent at EOF for REPL.

posCC failed.