The Gordian Knot

[susliko]

A startling discovery

changing $.lambda_expression body from $._block to $._indentable_expression results in

• 11-fold decrease in parser generation times (from 110s to 10s on my laptop)
• parser size reduction from 41M to 22M
• resolution of Allow block before lambda_expression #54
• approximately the same smoke test accuracies (86% in scala2 compiler, 82% in dotty)

However, it brings the following problems:

Problem1: conflict of $.lambda_expression and $.self_type

Self-type in

trait A {
  self =>
  val a = 1
}

is parsed as lambda expression. Seems to be not much of a problem, as it can be resolved by disabling $.lambda_expression in $.templabe_body (that's also how it's in the ebnf)

Problem2: multi-line lambdas without newline before closing paren

x.map((x) => 
if a then b.c)

fails to be parsed, for $._indent token is emmitted and $._outdent is not.

x.map((x) => 
if a then b.c
)

is, on contrary, parsed correctly because of the newline before the closing paren.
For this issue - I can't come up with a solution and will be glad of any ideas

[eed3si9n]

scala> trait A
// defined trait A

scala> foo(new A:
     |   def x = 1)
val res0: A = anon$1@50df37eb

so at least in this case, the closing parenthesis works as an outdent, so maybe we should support that in a situation where the parser is looking for an outdent.

[eed3si9n]

Here's my PR for the outdent - #279

[SethTisue]

😮

[susliko]

@eed3si9n Thank you for the fix of outdents! It helped to solve remaining issues in this PR, it's ready for review
Key results from the initial description remained:

• ~11-fold decrease in parser generation times with 1Gb max memory usage ceiling
• parser size reduction from 41M to 22M with 1300 as grammar complexity ceiling

[eed3si9n]

Thanks @susliko!

[ckipp01]

Really great job on this @susliko!