GitLab

No measurable speed improvement.

Move call to [fold f qs] into loop, making it tail-recursive.
Then, hoist [loop] out of [fold].
The impact is measurable on [Coverage] (maybe 10% overall).

This seems to save a little time, but not as much as hoped.

Added [Lr1.nt_of_entry].
Implemented [ReferenceInterpreter.check_error_path].

This modifies the analyses in [Grammar] to use [Fix] as the fixed point
computation algorithm.

This reverts commit fc203a594b2d68f40b18966550e9c23460fc20e9.

   If the programmer did not provide a name [_i] is used (where [i] is
   the producer's position in the right hand side).

   This syntax modification actually simplifies the code:
   - There is no more on-the-fly naming of unnamed arguments.
   - Inlining is no more restricted to semantic actions that do not use $i.

   The patch is large because:
   - This change has an impact on many modules.
   - The expected results of the bench had to be updated.

Action: Use [Set.of_list].

This leads to slightly fewer questions in the main analysis (-10% on c.mly)
and many fewer nodes considered during backward search (4K instead of 60K on
c.mly), yet it is slightly slower (+10% on c.mly).