Re: [isabelle] qed and done take long for large goal states



Hi Larry,

With the plain matching, the processing times for the qed's and done's got down from about 4.5s to 0.8s on my machine (according to the Timing panel). Since there are dozens of them in that particular theory, the effect was quite noticeable.

Andreas

On 10/08/16 12:27, Lawrence Paulson wrote:
As the author of this code, which is already pretty tricky, I would be surprised if it could be fixed easily. If it were, the penalty might be much more permissive filtering, which would defeat the purpose of this code.

It would be useful to see âgot indeed a bit fasterâ quantified. Did the change make a really big improvement in some situations?

Larry

On 9 Aug 2016, at 22:40, Makarius <makarius at sketis.net> wrote:

On 03/08/16 09:34, Andreas Lochbihler wrote:

I manually changed the appropriate
line for Isabelle2016 and processing got indeed a bit faster.

Unfortunately, abbreviations are no longer contracted as they were if
eta-expanded terms are involved. For example,

 abbreviation "rel R â list_all2 (rel_option R)"

 term "list_all2 (rel_option undefined)"
 term "list_all2 (Îx. rel_option undefined x)"

Before the change, both "term" command output "rel undefined". After the
change, the second prints "list_all2 (rel_option undefined)".

This requires more rethinking of the approach.

Net.match_term explicitly says "MUST BE BETA-ETA NORMAL", in contrast to
Net.unify_term.

I am unsure if the net operations can be refined to cope with non-normal
terms.


	Makarius







This archive was generated by a fusion of Pipermail (Mailman edition) and MHonArc.