Tej Chajed authored 4 years ago and Ralf Jung committed 4 years ago

The syntax re-uses the existing support for pure names, namely the % and
%H patterns. Using "[% H]" is like `iDestruct ... as (?) "H"` and using
"[%x H]" (with the string-ident plugin) is like `iDestruct ... as (x)
"H"`.

This changes how these patterns are parsed. Previously, both would have
been handled as conjunctions (using IntoAnd or IntoSep, depending on
whether the hypothesis is persistent or not). This means it was possible
for the user to use "[% H]" to destruct a pure hypothesis ⌜φ ∧ ψ⌝ and
put only the first conjunct in the Gallina context, leaving the other
one in the IPM; such patterns will now break, since iExistDestruct
does not handle this use case.

Tej Chajed authored 4 years ago and Ralf Jung committed 4 years ago

Fixes #404

This is an alternative to !591.

This reverts merge request !591

Tej Chajed authored 4 years ago and Robbert Krebbers committed 4 years ago

`iDestruct ("H" with "HP")` where "H" is persistent is supposed to
consume "H" if it isn't a forall. Due to a bug in the tactic, this
behavior was never triggered and "H" was always left alone.

This test is incompatible with Coq 8.8 and Coq 8.9, but Iris no longer
supports those versions.

put it into a type class `BiPureForall`.

This property does not hold for embeddings of classical logic into Coq.

A failing iIntros for implications should prettify the identifier before
printing, and iIntros on something that isn't a wand or implication
should say what couldn't be introduced (to clarify that `iIntros "HP
HQ"` failed because of the HQ in particular, for example).

`iDestruct H as "H1 H2"` produces an error that says the pattern should
contain exactly one proper introduction pattern. When multiple patterns
are provided, due to Ltac variable shadowing iDestructHypFindPat was
instead reporting only the first pattern in the error message (and even
that was printed as the parsed AST rather than the original string).

The error handling for `iIntro (?)` and similar tactics didn't correctly
report failures when the goal couldn't be turned into a universal
quantifier. This is something missing from !482 due to no test
triggering the error.

Tej Chajed authored 4 years ago and Robbert Krebbers committed 4 years ago

Fixes #337.

Preserve identifiers in binders where possible, analogous to the support
for destructing existentials in !479.

Fixes #336.

Tej Chajed authored 4 years ago and Robbert Krebbers committed 4 years ago

When running `iDestruct "H" as (?) "H"`, use the name of the binder in
"H". For example, if "H" has type `∃ y, Φ y`,  we now use `y` as the
name of the variable after freshening. Previously the name was always
the equivalent of running `fresh H`.

The implementation achieves this by forwarding the desired identifier
name through the `IntoExist` typeclass. Identifiers are serialized in
Gallina by using them as the name of a function of type `ident_name :=
unit -> unit`.