The only recent change to ninja that comes to mind is: #5393
Not clear whether it is related.

CC @cannorin

I suppose https://github.com/rescript-lang/ninja/pull/2/files is more relevant? (it changed how the #Done line is written)

Thanks -- indeed.

In fact, it seems unrelated to ninja, but related to the vscode extension.

CC @zth could you take a look -- this looks like an issue with the current extension.

CC @zth could you take a look -- this looks like an issue with the current extension.

On it.

I draw the same conclusions as the initial post - the compiler log is not terminated for some reason, so the extension does not clear errors. I also tested with ReScript 9.1.4 where I could not reproduce the issue. So I don't think it's in the extension itself.

Looks like both the compiler and the extension have changed behaviour in their current versions.

Was wondering whether the extension should be resilient to the compiler not printing the log terminator, just as the previous version of the extension seems to be doing. Though the normal situation where that would happen is when the compiler is in the process of building, and has not finished yet. I guess this is possible as the compiler log would be written incrementally to disk and the watcher could pick up an incomplete change.

So perhaps making the extension more robust w.r.t. unterminated output would also make it wrong.

So: isolated a more basic wrong behaviour, without using a watcher: rescript clean then rescript then rescript again gives a terminated log on first build, but unterminated on second build.

That's easy to add to a test too, which would be much more complicated if a watcher were involved.

Taking a less rushed look at ninja now.

Actually all of the above.
Fixing the missing closing log removes printing of cycles done in #5393