parsetl: speedup parsing of n-ary operators with many operands

Issue #500, reported by Yann Thierry-Mieg.

* spot/parsetl/parsetl.yy, spot/parsetl/scantl.ll: Use variant
to store a new pnode objects that delays the construction of n-ary
operators.
* spot/parsetl/Makefile.am: Do not distribute stack.hh anymore.
* spot/tl/formula.cc: Fix detection of overflow in Star and FStar.
* HACKING: Update Bison requirements to 3.3.
* tests/core/500.test: New test case.
* tests/Makefile.am: Add it.
* tests/core/ltl2tgba2.test, tests/core/ltlsynt.test,
tests/core/tostring.test: Adjust to new expected order.
* NEWS: Mention the change.
This commit is contained in:
Alexandre Duret-Lutz 2022-03-26 15:57:56 +01:00
parent 46f3f5aaf4
commit 9c6a09890e
11 changed files with 374 additions and 181 deletions

9
NEWS
View file

@ -66,6 +66,15 @@ New in spot 2.10.4.dev (net yet released)
- purge_dead_states() will now also remove edges labeled by false
(except self-loops).
- When parsing formulas with a huge number of operands for an n-ary
operator (for instance 'p1 | p2 | ... | p1000') the LTL parser
would construct that formula two operand at a time, and the
formula constructor for that operator would be responsible for
inlining, sorting, deduplicating, ... all operands at each step.
This resulted in a worst-than-quadratic slowdown. This is now
averted in the parser by delaying the construction of such n-ary
nodes until all children are known.
Bugs fixed:
- reduce_parity() produced incorrect results when applied to