Slow inference with large Maps of literals

[adkian-sifive]

Compiler version

3.4.0, 3.4.1, 3.4.2

Minimized code

I put it in a gist here since the problematic Maps are quite large: https://gist.github.com/adkian-sifive/5c904054f400871f2fe712ab160be1b3

Output

The compiler hangs

Expectation

In this case the compiler should complain about not finding a main method, but it does not do that and just hangs.

[som-snytt]

3.4.2 takes just over two minutes to compile on my old box. (2.13.14 takes a few seconds. IIRC it checks for all the same primitive when lubbing.)

TIL 2.13 does not try to initialize a main-less object. 3.4 does indeed complain.

[adkian-sifive]

Thanks for checking! After retrying with a bit more patience it did indeed complain after around 5 mins on my old potato. Perhaps a performance enhancement opportunity but doesn't seem like a bug...

[jackkoenig]

This does seem like a pretty painful performance issue, is that something the Scala 3 developers track with Github issues?

[som-snytt]

For example #19907

That is already in 3.4.2.

On HEAD, the example is still over a minute (trying it locally, informally).

[jackkoenig]

Thanks for the additional context!

Do you think this is worth re-filing separately as a performance issue?

[som-snytt]

There are a couple of open PRs for performance of calculating intersection types, but I don't know if they will apply here.

It couldn't hurt to reopen this ticket and let normal triage take its course, so I'll do that.

Edit: the ticket was already triaged, so I updated the labels. I may be wrong about what causes the slow-down.