| commit | 70ee8c77a8287612c038b9a6f14f6f3b38281bfc | |
| author | Akim Demaille <akim.demaille@gmail.com> | |
| Sun, 8 Nov 2020 12:30:14 +0000 (8 13:30 +0100) | ||
| committer | Akim Demaille <akim.demaille@gmail.com> | |
| Tue, 10 Nov 2020 06:08:07 +0000 (10 07:08 +0100) | ||
| tree | ddd61fcca4aaa9614b75dc80145cbaa3a4882c5e | treesnapshot (tar.gz zip) |
| parent | e9a43ed4ae9c9c4d6f4e250e4f929ac9602d7452 | commitdiff |
multistart: fix IELR computations
IELR needs to rule out the successors of the kernel items of the
initial state (`$accept: input • $end`). In the case of multistart,
this condition must be expressed differently: the mere item index does
not suffice.
* src/ielr.c (ielr_item_has_lookahead, ielr_compute_lookaheads): Don't
rely on the item index to check whether is_successor_of_initial_item.
It is certainly more costly than just checking the item index, but (i)
we need to compute the rule anyway, so it's not very much more costly,
and (ii) in ielr_item_has_lookahead, this situation is actually
impossible, so an optimizing compiler reading the assertions should
actually avoid this computation.
IELR needs to rule out the successors of the kernel items of the
initial state (`$accept: input • $end`). In the case of multistart,
this condition must be expressed differently: the mere item index does
not suffice.
* src/ielr.c (ielr_item_has_lookahead, ielr_compute_lookaheads): Don't
rely on the item index to check whether is_successor_of_initial_item.
It is certainly more costly than just checking the item index, but (i)
we need to compute the rule anyway, so it's not very much more costly,
and (ii) in ielr_item_has_lookahead, this situation is actually
impossible, so an optimizing compiler reading the assertions should
actually avoid this computation.
| src/ielr.c | diffblobblamehistory |