Skip to content

refactor(pl): pr bridges as recheckable per-logic rules - #1165

Open
strub wants to merge 1 commit into
pl/rndfrom
pl/pr
Open

strub wants to merge 1 commit into
pl/rndfrom
pl/pr

Conversation

@strub

@strub strub commented Oct 6, 2026

Copy link
Copy Markdown
Member

PR of the program-logic reorganization stack (see
src/phl/REFACTORING.md). It migrates the bridge tactics relating
judgements: bypr, pr_bounded, hoare, phoare split and
hoare split:

  • rules/hoare/ecHoareOfBdHoare.ml: hoare [c : P ==> Q] from
    phoare [c : P ==> !Q] = 0%r (no exceptional postcondition);
    rules/bdhoare/ecBdHoareOfHoare.ml: the converse. The hoare tactic
    on a phoare goal is derived: the bound-changing consequence to
    = 0%r, then the rule.
  • rules/bdhoare/ecBdHoarePr.ml: phoare [f : P ==> Q] ~ d from
    forall &m, 0%r <= d /\ (P => Pr[f(args) @ &m : Q] ~ d);
    rules/hoare/ecHoarePr.ml derives the hoare form from it and the view
    above. rules/equiv/ecEquivPr.ml: equiv [f1 ~ f2 : P ==> Q] from
    the equality of the distributions of two observed expressions; the
    node records their common type, re-checked by the builder.
  • rules/bdhoare/ecBdHoarePrBounded.ml: a probability is in [0, 1] and
    that of false is 0, plus the premise forms P => 1%r <= d /
    P => d <= 0%r (the bound-changing consequence would add a
    0%r <= d conjunct, so they are kept as rules).
  • rules/bdhoare/ecBdHoareSplit.ml: inclusion-exclusion on a conjunction
    or a disjunction, and the complement rule; the bound change and the
    phoare split b1 b2 : R case analysis are derived.
  • rules/hoare/ecHoareSplit.ml: hoare split, derived from the
    consequence rules (no rule of its own).
  • Each rule (statement and/or procedure form) is documented as an
    inference rule in its .mli and emits a node with a registered
    checker; side conditions (bound = 0%r, shape of the postcondition,
    bound combining the premises' bounds, ...) are part of the subgoal
    builders, so the checkers re-validate them.
  • EcPhlPr, EcPhlCoreView, EcPhlBdHoare, EcPhlHiBdHoare and EcPhlHoare
    are reduced to dispatchers and adapters; their interfaces are
    unchanged. EcPhlPr.t_prfalse, which has no caller, is not migrated
    and is flagged as unsound in a comment.

Behaviour is preserved: on the new test, goals after every bridge
tactic and the error messages of every failing invocation are identical
to those of the unmodified build.

A new test, tests/pr-bridges.ec, exercises every rule reachable from the
surface syntax and the error paths (the procedure forms of the split
rules are not reachable: phoare split cannot type its bounds on a
procedure goal). The stdlib, the unit tests and the examples pass under
EC_RECHECK=1 with no RecheckFailure. Each of the 14 checkers, when
deliberately broken, is caught only under EC_RECHECK (on the new test,
except the unreachable ones; on the stdlib where the rule is used
there: hoareF-of-bdhoare 2 files, bdhoareS-of-hoare 9,
bdhoareF-of-hoare 4, bdhoareF-pr 9, equivF-pr 8, bdhoareS-prbounded 6,
bdhoareF-prbounded 4, bdhoareS-split-or 1).

PR of the program-logic reorganization stack (see
src/phl/REFACTORING.md). It migrates the bridge tactics relating
judgements: `bypr`, `pr_bounded`, `hoare`, `phoare split` and
`hoare split`:

- rules/hoare/ecHoareOfBdHoare.ml: `hoare [c : P ==> Q]` from
  `phoare [c : P ==> !Q] = 0%r` (no exceptional postcondition);
  rules/bdhoare/ecBdHoareOfHoare.ml: the converse. The `hoare` tactic
  on a phoare goal is derived: the bound-changing consequence to
  `= 0%r`, then the rule.
- rules/bdhoare/ecBdHoarePr.ml: `phoare [f : P ==> Q] ~ d` from
  `forall &m, 0%r <= d /\ (P => Pr[f(args) @ &m : Q] ~ d)`;
  rules/hoare/ecHoarePr.ml derives the hoare form from it and the view
  above. rules/equiv/ecEquivPr.ml: `equiv [f1 ~ f2 : P ==> Q]` from
  the equality of the distributions of two observed expressions; the
  node records their common type, re-checked by the builder.
- rules/bdhoare/ecBdHoarePrBounded.ml: a probability is in [0, 1] and
  that of `false` is 0, plus the premise forms `P => 1%r <= d` /
  `P => d <= 0%r` (the bound-changing consequence would add a
  `0%r <= d` conjunct, so they are kept as rules).
- rules/bdhoare/ecBdHoareSplit.ml: inclusion-exclusion on a conjunction
  or a disjunction, and the complement rule; the bound change and the
  `phoare split b1 b2 : R` case analysis are derived.
- rules/hoare/ecHoareSplit.ml: `hoare split`, derived from the
  consequence rules (no rule of its own).
- Each rule (statement and/or procedure form) is documented as an
  inference rule in its .mli and emits a node with a registered
  checker; side conditions (bound `= 0%r`, shape of the postcondition,
  bound combining the premises' bounds, ...) are part of the subgoal
  builders, so the checkers re-validate them.
- EcPhlPr, EcPhlCoreView, EcPhlBdHoare, EcPhlHiBdHoare and EcPhlHoare
  are reduced to dispatchers and adapters; their interfaces are
  unchanged. EcPhlPr.t_prfalse, which has no caller, is not migrated
  and is flagged as unsound in a comment.

Behaviour is preserved: on the new test, goals after every bridge
tactic and the error messages of every failing invocation are identical
to those of the unmodified build.

A new test, tests/pr-bridges.ec, exercises every rule reachable from the
surface syntax and the error paths (the procedure forms of the split
rules are not reachable: `phoare split` cannot type its bounds on a
procedure goal). The stdlib, the unit tests and the examples pass under
EC_RECHECK=1 with no RecheckFailure. Each of the 14 checkers, when
deliberately broken, is caught only under EC_RECHECK (on the new test,
except the unreachable ones; on the stdlib where the rule is used
there: hoareF-of-bdhoare 2 files, bdhoareS-of-hoare 9,
bdhoareF-of-hoare 4, bdhoareF-pr 9, equivF-pr 8, bdhoareS-prbounded 6,
bdhoareF-prbounded 4, bdhoareS-split-or 1).
@strub
strub added this pull request to stack #1156 October 6, 2026 22:25
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant