| Age | Commit message (Collapse) | Author |
|---|
|
- prod_applist
- prod_applist_assum
- lambda_applist
- lambda_applist_assum
expect an instance matching the quantified context. They are now in
term.ml, with "list" being possibly "vect".
Names are a bit arbitrary. Better propositions are welcome. They are
put in term.ml in that reduction is after all not needed, because the
intent is not to do β or ι on the fly but rather to substitute a λΓ.c
or ∀Γ.c (seen as internalization of a Γ⊢c) into one step,
independently of the idea of reducing.
On the other side:
- beta_applist
- beta_appvect
are seen as optimizations of application doing reduction on the fly
only if possible. They are then kept as functions relevant for
reduction.ml.
|
|
It will later be used to fix a bug and improve some code.
Interestingly, there were a redundant semantic equivalent to
extended_rel_list in the kernel called local_rels, and another private
copy of extended_rel_list in exactly the same file.
|
|
|
|
|
|
direct aliases are ok, and indices should not be made polymorphic. Fixes NFix.
|
|
|
|
I was trying to be a bit too clever with not substituting the universe
instance everywhere: the constructor type/inductive arity has to be
instantiated before instantiate_params runs, which became true only
for constructor types since my last commit.
|
|
|
|
Avoid undeeded large substitutions, and add test-suite file for
fixed bug 4283 in closed/
|
|
Reviewed by M. Sozeau
This commit fixes template polymorphism and makes it more precise,
applying to non-linear uses of the same universe in parameters of
template-polymorphic inductives. See bug report and
https://github.com/coq/coq/pull/69 for full details.
I also removed some deadcode in checker/inductive.ml.
I do not know if it is also necessary to fix checker/indtypes.ml.
|
|
Missing universe substitutions of mind_params_ctxt when typechecking
cases, which appeared only when let-ins were used.
|
|
Oups, sorry, I should have compiled the stdlib in full. Not only
the ~polyprop wasn't propagated properly, but Matthieu made it be
false by default somewhere instead of true. Argl...
|
|
The ~polyprop argument wasn't propagated properly anymore,
leading the extraction to try to operate on situations it cannot
handle (yet). Cf Table.error_singleton_become_prop for more details.
Regression test added.
|
|
is reduced as if without let-in, when applied to arguments.
This allows e.g. to have a head-betazeta-reduced goal in the following example.
Inductive Foo : let X := Set in X := I : Foo.
Definition foo (x : Foo) : x = x. destruct x. (* or case x, etc. *)
|
|
|
|
One remaining issue: aliased constants raise an anomaly when some unsubstituted
universe variables remain. VM may suffer from the same problem.
|
|
|
|
|
|
for the record binder of classes. This name is no longer generated
in the kernel but part of the declaration. Also cleanup the interface
to recognize primitive records based on an option type instead of a
dynamic check of the length of an array.
|
|
Let r.(p) be a strict subterm of r during the guardness check.
|
|
Involves changing the [mind_finite] field in the kernel from a bool to the trivalued type [Decl_kinds.recursivity_kind]. This is why so many files are (unfortunately) affected. It would not be very surprising if some bug was introduced.
|
|
Now kernel/indtypes builds the corresponding terms (has to be trusted)
while translate_constant just binds a constant name to the
already entered projection body, avoiding the dubious "check"
of user given terms. "case" Pattern-matching on primitive records is
now disallowed, and the default scheme is implemented using
projections and eta (all elimination tactics now use projections
as well). Elaborate "let (x, y) := p in t" using let bindings
for the projections of p too.
|
|
My previous optimization of guard checking (f1280889) made it slightly stricter,
in the presence of dependent pattern matching and nested inductive types whose
toplevel types are mutually recursive.
The following (cooked-up) example illustrates this:
Inductive list (A B : Type) := nil : list A B | cons : A -> list A B -> list
A B.
Inductive tree := Node : list tree tree -> tree.
Lemma foo : tree = tree. exact eq_refl. Qed.
Fixpoint id (t : tree) :=
match t with
| Node l =>
let l := match foo in (_ = T) return list tree T with eq_refl => l end
in
match l with
| nil => Node (nil _ _)
| cons x tl => Node (cons _ _ (id x) tl)
end
end.
is accepted, but changing tree to:
Inductive tree := Node : list tree tree -> tree.
with tree2 := .
made id be rejected after the optimization.
The same problem occurred in Paco, and is now fixed.
Note that in the example above, list cannot be mutually recursive because of the
current strict positivity condition for tree.
|
|
In check_one_cofix, we now avoid calling dest_subterms each time we meet a
constructor by storing both the current tree (needed for the new criterion)
and a precomputed array of trees for subterms.
|
|
When dynamically computing the recarg tree, we now prune it according to the
inferred tree. Compilation of CompCert is now ok.
|
|
I had introduced it by mistake due to my OCaml dyslexia :)
Thanks to Enrico and Arnaud for saving my day!
|
|
variables.
Simplifies instantiation of constants/inductives, requiring less allocation and Map.find's.
Abstraction by variables is handled mostly inside the kernel but could be moved outside.
|
|
- realargs: refers either to the indices of an inductive, or to the proper args
of a constructor
- params: refers to parameters (which are common to inductive and constructors)
- allargs = params + realargs
- realdecls: refers to the defining context of indices or proper args
of a constructor (it includes letins)
- paramdecls: refers to the defining context of params (it includes letins)
- alldecls = paramdecls + realdecls
|
|
If the return predicate is not dependent, we avoid dynamically regenerating the
regular tree of the corresponding inductive type. This includes the commutative
cut rule. Should solve some performance issues observed in Compcert and Paco at
Qed time.
|
|
Used by the new guard criterion compatible with type isomorphisms.
|
|
|
|
This reverts commit ec1bb8a981fef14b58ab65483244fc42b05aef13.
|
|
This reverts commit 6a3bcd3ae320e65347cbd6ef4bac458f073d02ea.
Apply again if this kind of dependently typed programming idioms are needed.
|
|
commutative cut rule.
The error messages of the guard checker are now sometimes not
informative enough.
|
|
A pattern matching is can now be a subterm if:
- Every branch is a subterm
- The return predicate is a pattern matching whose return predicate is an arity.
- That pattern matching (in the return predicate) returns the same inductive
family in the conclusion of each branch.
The commutative cut rule hasn't been updated accordingly yet.
|
|
|
|
Following Bruno's suggestion, we check if the tree expected for the recursive
argument is included in the one which is inferred. This check is probably
not necessary in the current state of affairs, but might become so after
further extensions of the guard condition.
|
|
extensions.
|
|
|
|
|
|
|
|
instantiated in the return predicate are now taken into account. The resulting
recargs tree is the intersection between the one of the branches and the
appearing in the return predicate. Both the domain and co-domain are filtered.
|
|
Some fixpoints are now rejected in the standard library, but that's probably
because we compare trees for equality instead of intersecting them.
|
|
pattern matching.
This patch should be improved in two ways:
(1) Implement the same checks for the commutative cut subterm rule.
(2) When checking safe recursive subterms for each of the branches in a match,
instanciated parameters in the return predicate should be taken into account.
Step (1) should be enough to restore a correct guard condition, but (2) will
be required if we don't want to rule out some legitimate and practical examples.
|
|
|
|
- Add a tentative syntax for specifying universes: Type{"i"} and foo@{Type{"i"},Type{"j"}}.
These are always rigid.
- Use level-to-level substitutions where the more general level-to-universe substitutions
were previously used.
|
|
problem with hashconsing at the same time. This fixes bug# 3302.
|
|
|
|
A try at hashconsing all universes instances seems to incur a big cost.
- Do hashconsing of universe instances in constr.
- Little fix in obligations w.r.t. non-polymorphic constants.
Conflicts:
kernel/constr.ml
kernel/declareops.ml
kernel/inductive.ml
kernel/univ.mli
|
|
TODO fix interface on knowing_parameters to avoid useless array allocations.
|
