GitLab

Remove --no-print-directory, because it prevents emacs from jumping to the file that contains the error.

Use [pvarlocated] when creating the mock file, so that warnings or errors about unused semantic values are reported in the .mly file.

Change [src/Makefile] so that [make] under emacs produces errors that emacs knows how to locate (yay).

A pseudo-token (declared e.g. by %left) could appear in a position where a
normal token was expected (e.g., as an actual parameter of a parameterized
symbol) and that was not detected, giving rise (later on) to an assertion
failure.

Infer: change the content of the .mock.ml file so that all occurrences of the type of a nonterminal symbol are shared. Also, use the type inferred by OCaml instead of the type declared by the user, as they do not necessarily coincide. This closes issue #37. Adjust the demo test/dynamic/misc/variants, where the OCaml type-checker now correctly requires one more branch in the [match] construct.

New test test/dynamic/misc/variants, which in the current version of Menhir causes a segmentation fault. This illustrates issue #37.

Change [Infer.ntvar] to return a type, not a string. Do not export [CodePieces.ntvar], which is unused.

This tells opam that a new version of menhirLib or menhirSdk cannot
coexist with an old version of the menhir package (from the time
when menhirLib and menhirSdk were not separate packages).

as suggested by Yann in merge request !9.
When a parameterized symbol does not use its parameter,
fewer graph edges are created.
Add two tests that were rejected but can be accepted
(expansion_ok, expansion_unused)
and another test which was accepted already
(self_ground).
The effect of this change is marginal: in practice,
very few graphs are affected by this change.
Nevertheless, with this change, the safe-expansion check
might be correct and complete (who knows?).

Add a test that must be rejected because expansion diverges,
and was not properly rejected until now.

[make speed]: perform internal time measurements. Do not subtract the time of the dry run. Perform 10 runs in succession. All this should contribute to more stable performance data.