Open sakekasi opened 1 month ago
This is a bug. The reconstruction tactics should either fail or succeed, never leaving any subgoals behind. Perhaps this is due to some changes in more recent versions of Coq. CoqHammer was originally written in 2018, and I haven't actively worked on it since 2020, only maintaining.
As a workaround, you can try wrapping up the reconstruction tactics in some combination of solve
(https://coq.inria.fr/doc/V8.19.0/refman/proof-engine/ltac.html#coq:tacn.solve) and unshelve
. I already did that, but maybe something needs to change there.
I'm trying to run an evaluation of how many proofs coqhammer can solve in the coq_wigderson project. To do this, I'm running the hammer tactic from coq-serapi, using coq 8.13. Running hammer on the lemma
nbd_2_colorable_3
(in the file 'coloring.v' on line 232) succeeds with the reconstruction tacticqauto use: PositiveMap.gss, WP.F.not_find_in_iff, WP.F.not_mem_in_iff, WF.mem_find_b, WP.F.
.However, instead of solving the goal it was run on, running this tactic takes me from this proof state:
to this one
I expected hammer and its reconstruction tactics to do one of 2 things:
I'm a little surprised to find that the
qauto
tactic does neither of these things. Is this a bug in the tactic, or intended behavior?Thanks for your help!