set code attribute on discriminator equations

MaSh error handling

repair invariant in MaSh when learning new proofs

killed more "no_atp"s

killed most "no_atp", to make Sledgehammer more complete

doc fixes suggested by Andreas L.

make sure that registered code equations are actually equations

accept very long lines in MaSh

make sure add: doesn't add duplicates, and works for [no_atp] facts

no fact subsumption -- this only confuses later code, e.g. 'add:'

generate a comment storing the goal nickname in "learn_prover"

clarified message

added thread-safety warnings

avoid confusion between outer ifs and inner ifs, that caused tactic to go berserk

handle nested tuples in 'let's

more exercises

also unfold let (_, _) = ... syntax

added helper function (that solves chicken-and-egg problem w.r.t. "callssss")

graceful handling of abort

generalized interface

choose facts to reprove more randomly, to avoid getting stuck with impossible problems at first

thread the goal through instead of relying on unreliable (possibly fake) state

added comment

tuning

verbose minimization when learning from ATP proofs

if slicing is disabled, pick the maximum number of facts, not the number of facts in the last slice

if slicing is disabled, don't enforce last slice's "max_facts", but rather the maximum "max_facts"

fast track -- avoid domain error in 0 case

remove overloading of "max_facts" -- it already controls the number of facts passed to ATPs for 'learn_prover'

merged

tuned exercises

use Code.abort instead of undefined in auto-generated equations

have MaSh minimize

tweaked signature

strengthened tactic to deal with 'False ==> ...'

made theorem extraction code not delve too far when looking at local fact, by relying on 'strip_thm' instead of (wrongly) reverse-engineering proof terms

improved duplicate detection in "build_name_tables" by ensuring that the earliest occurrence of a duplicate (if it exists) gets picked as the canonical instance

updated S/H docs

use MePo with Auto Sledgehammer, because it's lighter than MaSh and always available

drop only real duplicates, not subsumed facts -- this confuses MaSh

renamed relcomp_def to relcomp_unfold

fixed thm names

addressed rare case where the same symbol would be treated alternately as a function and as a predicate -- adding "top2I top_boolI" to a problem that didn't talk about "top" was a way to trigger the issue

added Nitpick limitations to docs

more defensive Nitpick setup -- exotic types of recursion are not supported yet in the model finder

keep temporary error handling in there until code equations are properly generated

tuning (simplified parts of 92c5bd3b342d)

tuning

stengthened tactic to cope with abort cases

tuned names

strengthened tactic w.r.t. "let"

more prominent MaSh errors

compile -- fix typo introduced in 07a8145aaeba

pass the right theorems to tactic

prove user-supplied equations for ctr and code reductions, preserving "let"s, "case"s etc.; generate code-style theorems (currently commented out since this still fails for many cases); filter tautologies (False ==> ...) out of generated theorems;

repaired confusion between the stated and effective fact filter -- the mismatch could result in "Match" exceptions

simplify fudge factor code

cleanup SMT-related config options

no isar proofs if preplay was not attempted

normalize more equalities