Requires Elpi 2.0.3 and Coq 8.20.
The major change is the port to Elpi 2.0 that reports type checking errors to the location of the offending term and not its enclosing rule.
Elpi Accumulate Db Header <db>
to accumulate just theDb
declaration but no code added after thatElpi File <file> <code>
to name a piece of code without requiring an external fileElpi Accumulate File Signature <file>
to accumulate only the types declarations from a file.Elpi Typecheck
is deprecated and is a no-op sinceElpi Accumulate
performs type checking incrementally
- new
open-trm
argument for tactics with syntax ````(...)``` andltac_open_term:(...)
. Open terms can mention free variables. - new
{{:pat ...}}
quotations inside which_
is interpreted as a wildcard, not as a regular evar.
- Support export locality in
coq.TC.declare-instance
tc-instance
now just carries a priority, no matter if inferred or declared, and works for instances added asHint Resolve
to thetypclass_instances
database
- Support dune workspace build with
elpi
- Resolve
.elpi
files based on Coq's resolver. Paths are now expected to be of the form<coq_dir_path>/<rel_path>
, where<coq_dir_path>
part is a logical Coq directory (as mapped with-Q
or-R
), and<rel_path>
is a relative path from the corresponding directory.
Requires Elpi 1.19.2 and Coq 8.19 or Coq 8.20.
- New
coq.arguments.reset-simplification
- Change some speedup concerning universes
Requires Elpi 1.19.2 and Coq 8.19 or Coq 8.20.
- Fix release script to just publish coq-elpi (and not coq-elpi-tests)
- Fix opam constraints by adding upper bound
Requires Elpi 1.19.2 and Coq 8.19 or Coq 8.20.
- Fix type checking errors on inline code are now reported on the correct line in LSP based interfaces
- Fix various missing dependencies
- Fix rebuild before installation
- Change CI no more use of docker images
- Change silence
default-output-directory
warning
- Change organize the code inside a
tc
namespace
- Change do not leak
positive_scope
open
Requires Elpi 1.19.2 and Coq 8.19 or Coq 8.20.
- Change switch to dune
- New ppx_optcomp to support multiple Coq version
- New no need for dot-merlin-reader, OCaml's language server understands dune
- Change supports higher order unification
- Change syntax to register, enable and disable solver
- Change solutions found in Elpi are eta-contracted
- New
coq.debug
- New
coq.pstring->string
andcoq.string->pstring
- New
@warn!
attribute forcoq.notation.add-abbreviation
- New
coq.*.set.min
,coq.*.set.max
,coq.*.set.choose
,coq.*.set.fold
,coq.*.set.partition
- New
coq.*.map.fold
- New
coq.env.projection?
andcoq-env.primitive-projection?
- New
primitive (pstring S)
in Coq 8.20
Requires Elpi 1.18.2 and Coq 8.19.
- Fix initial synterp state of commands with a synterp phase
Requires Elpi 1.18.2 and Coq 8.19.
- New
Elpi Accumulate dbname File filename
allows to accumulate a file int a db - Change
Elpi Db
now only creates (and initialises) a database for the specified phase
- New
coq.parse-attributes
support for theattlabel
specification, seecoq-lib-common.elpi
for its documentation. - New
coq.goal->pp
- Replace
coq.replay-all-missing-synterp-actions
by (nestable) groups of actions - New
coq.begin-synterp-group
andcoq.end-synterp-group
primitives - New
coq.replay-synterp-action-group
primitive (replacescoq.replay-all-missing-synterp-actions
in conjunction with a group) - New
coq.replay-next-synterp-actions
to replay all synterp actions until the next beginning/end of a synterp group - New
coq.primitive.projection-unfolded
to fold/unfold a primitive projection. Note that unfolded primitive projections are still compact terms, but they are displayed asmatch
expressions and some Ltac code can see that.
Requires Elpi 1.18.2 and Coq 8.19.
- Fix
coq.elaborate-*
does not erase the type annotation ofLet
s (regression introduced in 2.0.1). This fix may introduce differences in generated names - Fix
coq.elaborate-*
are not affected anymore by printing options
- Fix install the right initial parsing state (the one before any synterp action is re-played)
- Fix evar instantiation loss when crossing the elpi/ltac border
- Fix encoding of "definitional classes" (
Class
with no record) - Fix order of implicit arguments of
Record
- Change requiring
elpi
does not load primitive integers nor primitive floats
- TC: avoid declaring options twice (could make vscoq2 fail)
- CS:
cs
now takes a context, a term that is the projection of some structure applied to the parameters of the structure, a term to put a structure on and the solution to return
Requires Elpi 1.18.1 and Coq 8.19.
This minor release adds compatibility with Coq 8.19.
Requires Elpi 1.18.1 and Coq 8.18.
This major release accommodates for the separation of parsing from execution of Coq 8.18 enabling Coq-Elpi programs to be run efficiently (and correctly) under VSCoq 2.0.
- New section about parsing/execution separation in the Writing commands in Elpi tutorial
- New
Elpi *
commands understand the#[phase]
attribute, see the doc in the README file, and the section about the separation of parsing from execution - New
Elpi Export
understands anAs
clause to rename or alias a program when exported
- Change
coq.elpi.add-predicate
now locality can be changed - Experimental
coq.toposort
returns a valid topological ordering of the nodes of a graph - Change
coq.TC.db-for
, now instances are returned sorted wrt their priority - New
tc-priority
, contains the priority of an instance and if the priority has been given by the user or computed bycoq
- Change
tc-instance
, now the type isgref -> tc-priority -> tc-instance
i.e. the priority is not an integer anymore - New
coq.ltac.fresh-id
to generate fresh names in the proof context - New
@no-tc!
attribute supported bycoq.ltac.call-ltac1
- New
coq.TC.get-inst-prio
returns thetc-priority
of an instance - New
synterp-action
datatype - New
coq.replay-all-missing-synterp-actions
- New
coq.replay-synterp-action
- New
coq.next-synterp-action
- New
coq.synterp-actions
(parsing phase only)
- New
tc
app providing an implementation of a type class solver written in elpi. This app is experimental
Requires Elpi 1.16.5 and Coq 8.18.
- Fix
Elpi Export
broken when used from VsCoq2
- New
ltac1-tactic
opaque data type - New
tac
argument constructor - Change
coq.ltac.call-ltac1
now accepts either a string (tactic name) or a tactic expression (of typeltac1-tactic
) - New
ltac_tactic:(...)
syntax to pass tactic expressions to Elpi tactics - New
coq.extra-dep
predicate
Requires Elpi 1.16.5 and Coq 8.18.
- Automate release process
Requires Elpi 1.16.5 and Coq 8.18.
- New
coercion
app providingcoercion
predicate to program coercions (thanks @proux01). This app is experimental.
- Removed option
@nonuniform!
as it disappears from Coq 8.18. (c.f. coq/coq#17716 )
Requires Elpi 1.16.5 and Coq 8.17.
- Mention the trace browser for VSCode in the Elpi tutorial.
- New
coq.elpi.accumulate-clauses
takes a list of clauses which share the same DB and accumulation site - New
coq.elpi.add-predicate
to declare the signature of a new predicate into a Db - New
coq.elpi.predicate
to build a term of typeprop
out of a predicate name and arguments - Change
coq.env.global
now relates a term with a gref, instead of working one way only - Change
coq.elpi.accumulate*
generalise clauses over global universe level, and error if algebraic levels are present. It used to warn if levels were present. - New
coq.elaborate*skeleton
support the@no-tc!
option to disable type class resolution - New
@global!
option forcoq.elpi.accumulate*
- New
coq.env.current-section-path
- New
coq.TC.db-tc
giving all type classes - New
coq.reduction.eta-contract
- Fix evar declarations were (rarely) generated at the wrong depth, possibly resulting in variable captures in types containing binders
- Fix
assert false
in evar instantiation readback (eta contraction code was incomplete) - Fix resiliency in case a goal is closed by side effect (was raising fatal errors such as "Not a goal" or "Not a variable after goal")
- Change assigning a hole linked to an evar always triggers type checking. This is necessary even if the term being assigned is well typed since one may still need to declare some universe constraints.
- Change propagate type constraints in
Prop
inward (Coq 8.17 only). Eg.Check (T -> _) : Prop
fails in 8.17 since_
is assumed to be inType
. We propagate the constraint ourselves across->
,/\
,\/
and~
. - Quotations
{{ ... }}
are now parsed by Coq ensuring the end of input is reached. Spurious text results in a parse error. For example{{ f ) }}
is no more accepted, as well as{{ _.x }}
- New
Elpi Print
also print the program in.txt
format
- Change compilation cache able to prevent most of lengthy compilations in
Hierarchy-Builder for MathComp 2.0. In some cases Coq-Elpi is more picky
about the order of accumulated files, in particular a file containing
the spilling of a predicate
{p}
needs to be accumulated after the type or mode ofp
is declared
derive Inductive i {A}
now setsA
implicit status globallylock Definition f {A}
now setsA
implicit status globally
Requires Elpi 1.16.5 and Coq 8.17.
- New
coq.int->uint63
andcoq.float->float64
- Fix bug introduced in 1.17.0 affecting
coq.ltac.call-ltac1
Requires Elpi 1.16.5 and Coq 8.17.
- New
coq.modpath->library
- New
coq.modtypath->library
- Fix
coq.env.*
APIs generating inductives, definitions and modules now emit metadata in the.glob
files so thatcoqdoc
can generate hyperlinks
- Add
NES.{List,Print}
. - Support relative paths in
NES.{Open,List,Print}
(path_.P
references top-level namespaceP
, paths without a leading_.
are relative to the current namespace)
Requires Elpi 1.16.5 and Coq 8.16.
The main change is the derive
app which must now be loaded
by importing derive.std
(just loading derive
won't work).
See the new derive documentation.
- Change
coq.env.module
andcoq.env.module-type
do not fail if the module (type) contains a mutual inductive. The resultinggref
is going to me unusable with most APIs, though. - Change
coq.env.module
returns a ADT describing the module contents - Change
coq.gref->path
andcoq.gref->id
do work ongref
which point to mutual inductives. - New
coq.env.term-dependencies
computing all thegrefs
occurring in a term. - New
coq.redflag
andcoq.redflags
types for@redflags!
option understood bycoq.reduction.lazy.*
and coq.reduction.cbv.norm
- New
coq.env.fresh-global-id
- Change
derive
usage. One should now importFrom elpi.apps Require Import derive.std
- Change derivations
eq
andeqOK
move toderive.legacy
- New derivations
eqb
andeqbOK
subsuming the previous ones
Requires Elpi 1.16.5 and Coq 8.16.
- Fix parse error location display for quotation code
- Fix HOAS of inductives with non-uniform parameters
Requires Elpi 1.16.5 and Coq 8.16.
- Fix parse error location display for inline code
- Fix HOAS of evars: pruning was not propagated to the type of the evar
Requires Elpi 1.16.5 and Coq 8.16.
- Fix lexical analysis inside quotations error location display
- Fix drop of universe constraints attached to automatically generates
universe levels (eg when
sort (typ X)
is passed to Coq) - Fix nix CI
Requires Elpi 1.16.5 and Coq 8.16.
- Fix parse error location display
Requires Elpi 1.16.5 and Coq 8.16.
- API:
- Fix
coq.env.indt-decl
correctly handles universes in parameters of universe polymorphic inductive - Fix
coq.typecheck-indt-decl
ignores non uniform parameters to compute the universe level of the inductive - Fix
coq.elaborate-indt-decl-skeleton
ignores non uniform parameters to compute the universe level of the inductive
- Fix
Requires Elpi 1.16.5 and Coq 8.16.
- API:
- Fix
coq.elaborate*skeleton
does refresh universes - New
@keepunivs!
attribute to force skeleton APIs to not refresh universes. This is useful to keep a link between a universe declaration and the declaration itself but still elaborate it - Fix Coq-Elpi is reentrant when commands call tactics
- Fix
Requires Elpi 1.16.5 and Coq 8.16.
The main changes are:
- experimental support for universe polymorphism. One can read and write
universe polymorphic terms and manipulate their constraint declarations.
Terms now have a new
pglobal
term constructor, akin toglobal
but for global references to universe polymorphic terms, also carrying a universe instance. The attribute@uinstance!
can be used to pass or retrieve a universe instance to/from APIs to access the Coq environment, as in@uinstance! I => coq.env.typeof GR Ty_at_I
. The meaning of@uinstance! I =>
depends ifI
is an unset variable or a concrete universe instance. In the former case the API generate a fresh universe instance (forGR
) and assign it toI
; in the latter case it uses the provided universe instance. See coq-builtin for the full documentation - command arguments are elaborated by Coq (unless told otherwise). As a consequence arguments can use the full Coq syntax, including deep pattern matching and tactics in terms. Raw arguments are (and will remain) available, but don't support that yet
- New experimental support for polymorphic definitions in
locker
- New example of
clearbody
tactic taking a list of names ineltac
- Change
derive
sets, globally,Uniform Inductive Parameters
. See https://coq.inria.fr/refman/language/core/inductive.html#coq:flag.Uniform-Inductive-Parameters for reference. The immediate effect is that inductive types uniform parameters don't have to be repeated in the types of the constructors (they can't vary anyway). Non-uniform parameters and indexes have to be passed, as usual. If the flag is unset by the userCoq-Elpi
will raise a warning since inference of non-uniform parameters is not implemented
- Change arguments to commands are elaborated by Coq by default
- New attribute
#[arguments(raw)]
to get arguments in raw format (as in version 1.14 or below) - Change raw inductive declaration using
|
to mark non-uniform parameters is expected to not pass uniform parameters to the inductive type (the same behavior applies to elaborated arguments, making the two consistent) - Change
coercion
attribute for record fields now takes valuesoff
,regular
orreversible
- New
pglobal
term constructor carrying agref
and auniv-instance
for universe polymorphic terms - New
upoly-indt-decl
argument type for polymorphic inductive types declarations - New
upoly-const-decl
argument type for polymorphic definitions - New
upoly-decl
data type for universe parameters declarations, i.e. the@{u1 u2 | u1 < u2}
Coq syntax one can use for inductives or definitions - New
upoly-decl-cumul
data type for universe parameters declarations, i.e. the@{u1 u2 | u1 < u2}
Coq syntax one can use for cumulative inductives - Rename
univ
->sort
i.e.(sort S)
is aterm
andS
can beprop
or(type U)
whereU
is auniv
- New
univ-instance
opaque type to represent how a polymorphic constant is instantiated, i.e.(pglobal GR I)
whereGR
is agref
andI
auniv-instance
- New
univ.variable
opaque type foruniv
which are not algebraic. This data type is used inupoly-decl
andupoly-decl-cumul
- New
coq.env.indc->indt
- New
coq.env.dependencies
to compute the dependencies of agref
- New
coq.env.transitive-dependencies
- New
@nonuniform!
and@reversible!
forcoq.coercion.declare
- New
@uinstance!
attribute supported by manycoq.env.*
APIs that can be used to read/write the universe instance of polymorphic constants. E.g.@uinstance! UI => coq.env.typeof GR Ty
can instantiateTy
toUI
if provided or setUI
to a fresh instance if not - New
@udecl!
attribute to declare polymorphic constants or inductives, like the@{u1 u2 | u1 < u2}
Coq syntax - New
@udecl-cumul!
attribute to declare polymorphic inductives, like the@{+u1 u2 | u1 < u2}
Coq syntax - New
@univpoly!
shorter version of@udecl!
, like the#[universes(polymorphic)]
Coq syntax (without giving any other@{u1 u2 | u1 < u2}
directive) - New
@univpoly-cumul!
shorter version of@udecl-cumul!
, like the#[universes(polymorphic,cumulative)]
Coq syntax - New
coq.env.global
API to craft aterm
from agref
. When used with spilling{coq.env.global GR}
gives either(global GR)
or(pglobal GR I)
depending onGR
being universe polymorphic or not. It understands the@unistance!
attribute for both reading or settingI
- New
coq.env.univpoly?
to tell if agref
is universe polymorphic and how many parameters it has - Change
coq.univ.leq
->coq.sort.leq
- Change
coq.univ.eq
->coq.sort.eq
- Change
coq.univ.sup
->coq.sort.sup
- New
coq.sort.pts-triple
computes the resultingsort
of a product - New
coq.univ.constraints
gives all the universe constraints in a first class form - Change
coq.univ.new
does not take a list anymore - New
coq.univ
to find a global universe - New
coq.univ.global?
tests if a universe is global - New
coq.univ.variable
links auniv
to auniv.variable
(imposing an equality constraint if needed) - New
coq.univ.variable.constraints
finds all constraints talking about a variable - New
coq.univ.variable.of-term
finds all variables occurring in a term - New
coq.univ-instance
links auniv-instance
to a list of ofuniv.variable
- New
coq.univ-instance.unify-eq
unifies twouniv-instance
(for the samegref
) - New
coq.univ-instance.unify-leq
unifies twouniv-instance
(for the samegref
) - New
coq.univ.set
OCaml's set foruniv
- New
coq.univ.map
OCaml's map foruniv
- New
coq.univ.variable.set
OCaml's set foruniv.variable
- New
coq.univ.variable.map
OCaml's map foruniv.variable
- New
Accumulate File <ident>
to be used in tandem with Coq 8.16From <path> Extra Dependency <file> as <ident>
Requires Elpi 1.15.0 and Coq 8.15.
- All
Elpi Bla
commands accept (and ignore with a warning) unknown attributes, to be forward compatible
Requires Elpi 1.14.1 and Coq 8.15.
- New 1 slot cache for context read back to improve the speed of FFI calls
needing to read back a large
coq_context
- New
Conversion.t
forgref
handwritten to minimize allocations - New terms of the form
(global ...)
are now hashconsed - New
extra_goals
postprocessing removingdeclare-evar/rm-evar
pairs which happen naturally writing code likecoq.unify-eq {{ f _ x }} {{ f y _ }}
(the_
are solved immediately, no need to declare them to elpi)
- New
coq.hints.opaque
- New
coq.hints.set-opaque
- Change load
coq.ltac.*
also in commands (and not just tactics) so that commands can easily turn holes into goals and inhabit them calling regular tactics. - New
coq.hints.add-resolve
- Fix
coq.option.add
survives the end of a file - New
coq.env.begin-module-functor
- New
coq.env.begin-module-type-functor
- New
coq.env.apply-module-functor
- New
coq.env.apply-module-type-functor
- New
coq.inline
with constructorscoq.inline.no
,coq.inline.at
andcoq.inline.default
- New
@inline-at! N
and@inline!
macros - Change
coq.env.add-axiom
honors@inline
macros
Requires Elpi 1.13.6 and Coq 8.15.
derive Inductive i {A}
now correctly setsA
implicit statuslock Definition f {A}
now correctly setsA
implicit status
- New
coq.arity->implicits
- New
coq.indt-decl->implicits
- New
coq.any-implicit?
Requires Elpi 1.13.6 and Coq 8.15.
- Change
{{ p x }}
is no more interpreted as a primitive projection even ifp
is the associated constant - New
{{ x.(p) }}
is interpreted as a primitive projection ifp
is a primitive projection - New
{{ x.(@p params) }}
is interpreted as a regular projection even ifp
is a primitive projection, since primitive projections don't have parameters and the user wrote some
- Fix globalization of
arity
inside a section - New
coq.option
type to access Coq's GOption system (Set/Unset vernaculars) - New
coq.option.add
- New
coq.option.get
- New
coq.option.set
- New
coq.option.available?
- New
coq.bind-ind-parameters
- New
locker
app providinglock
andmlock
commands
Requires Elpi 1.13.6 and Coq 8.14.
- Change
coq.bind-ind-arity
preserveslet
- New
coq.bind-ind-arity-no-let
to reducelet
, used incoq.build-match
- Fix
coq.build-match
puttinglet
bindings inmatch
return type - Change
coq.map-under-fun
preserveslet
Requires Elpi 1.13.6 and Coq 8.13.
- New
coq.env.informative?
to know if a type can be eliminated to build a term of sortType
- Fix
coq.warning
is synchronized with Coq's Undo machinery - Retire the venerable "elpi fails" message, replaced with something more precise inviting the user to report a bug: errors should be taken care of and reported nicely by the programmer.
- New
coq.uint63->int
- New
coq.float64->float
- New
coq.ltac.id-free?
tells if a given ident is already used to denote a goal hypothesis, or not.
- Fix derivation of induction principles for "data types" in
Prop
- Add derivation of
param1
for the equality testeq
with namet.param1_eq
- Fix
invert
andidx2inv
when dealing with containers - New datatypes from the Coq's prelude are derived in advance, no need to
to
derive nat
anymore.
Requires Elpi 1.13.6 and Coq 8.13.
- New node
proj
of typeprojection -> int -> primitive-value
holding the projection name (a Coq detail) and the number of the field it projects (0 based), eg:primitive (proj _ N)
stands for the projection for the Nth constructor field counting parameters. - Change
cs-instance
carries agref
- New
coq.notation.add-abbreviation-for-tactic
to add a parsing rule for a tactic-in-term, along the lines ofNotation foo := ltac:(elpi mytactic arguments)
but passingmytactic
the correctelpi.loc
of invocation. - New
@pplevel!
attribute to control outermost parentheses incoq.term->pp
and similar - New
coq.hints.add-mode
like theHint Mode
vernacular - New
coq.hints.modes
- New
coq.TC.declare-class
- Deprecate
coq.env.const-opaque?
->coq.env.opaque?
- Deprecate
coq.env.const-primitive?
->coq.env.primitive?
- Deprecate
coq.CS.canonical-projections
->coq.env.projections
- New
coq.env.primitive-projections
- Change
coq.warning
emits the same warning only once
Requires Elpi 1.13.6 and Coq 8.13.
- Cleanup
elpi.loc
attribute, which now carries a real loc and not a string. Thanks to elpi 1.13.6 we can project out the components without messing with regular expressions. Moreover loc are printed in a consistent way on Unix and Windows.
Requires Elpi 1.13.5 and Coq 8.13.
- Change
coq.gref->path
now (consistently) gives the path without the final id, which can be retrieved bycoq.gref->id
.
Requires Elpi 1.13.5 and Coq 8.13.
- Fix (reverse) the order of the context argument of
goal
. The head of the list is the most recent hypothesis and in the last to be loaded (the one with higher precedence) by implication when one writesCtx => ...
. - New
msolve
entry point for (possibly multi goal) tactics
- Fix argument interpretation for
coq.ltac.call-ltac1
, the context is now the one of the goal alone (and not the one of the goal plus the current one) - Rename
coq.ltac.then
tocoq.ltac.all
Requires Elpi 1.13.5 and Coq 8.13.
- New
lens
andlens_laws
for regular and primitive records with or without parameters derive
takes#[only(this, that)]
to select the desired derivations
- Fix
coq.elpi.accumulate
scopecurrent
, which was putting the closes in the current module for the current file, but was making them global for the files importing it - New scope
library
forcoq.elpi.accumulate
which links the clauses to the library, that is the module named after the file. - Fix databases are always available, no need to import files in the right order
when databases have named clauses. The error "Error: unable to graft this
clause: no clause named ..." should no more be raised in response to a
Require Import
. - New
coq.strategy.*
toset
andget
the unfolding priority of constants followed by the term comparison algorithm Coq uses at type checking time. - New
coq.env.record?
to test if an inductive is a record and if it has primitive projections - New
coq.env.recursive?
to test if an inductive is recursive - Change
coq.locate*
understands strings like"lib:some.name"
which point to global references registered via the CoqRegister
command - New
coq.ltac.fail
likecoq.error
but catch by Ltac - New
@ltacfail!
to be used like@ltacfail! Level => std.assert! ...
in tactic code to usecoq.ltac.fail
instead ofcoq.error
in case of failure - Change failure as is
elpi fails
(no more clauses to try) orelpi run out of steps
are not considered Ltac failures anymore, but rather fatal errors. Add a clausesolve _ _ :- coq.ltac.fail _
to preserve the old behavior. - New
coq.ltac.collect-goals
to turn unresolved unification variables into goals. - Fix
coq.env.add-const
now accepts an opaque definition with no given type. The body is assumed to be well typed and is quickly retypechecked.
- Fix handling of default case in
match
, now Coq'sif _ then _ else _
works just fine. - New quotation
{{:gref id }}
and{{:gref lib:qualid }}
that unfolds to thegref
data type ({{ id }}
and{{ lib:qualid }}
unfold to terms) - Change
solve
only takes 2 arguments (the arguments passed at tactic invocation time are now part of the goal) and the first argument is a single goal, not a list thereof. The second argument is now asealed-goal
. - Change
refine
now generates a list ofsealed-goal
s - Change
goal
now carries two unification variables standing for the raw solution to goal and the elaborated, well typed, one. Assigning a term to the raw variable triggers a call tocoq.elaborate-skeleton
which in turn assigns the other one to the (partial) proof term. Assigning the elaborated variable directly does not trigger a type check of the term.
- New
attributes
tactic argument (forTactic Notation
) - New
elpi tac
can receive attributes via the usual#[stuff] tac
syntax - New syntax to pass Elpi tactics arguments coming from Ltac variables:
ltac_string:(v)
(forv
of typestring
orident
)ltac_int:(v)
(forv
of typeint
orinteger
)ltac_term:(v)
(forv
of typeconstr
oropen_constr
oruconstr
orhyp
)ltac_(string|int|term)_list:(v)
(forv
of typelist
of ...)ltac_attributes:(v)
(forv
of typeattributes
) Example:
lets one writeTactic Notation "foo" string(X) ident(Y) int(Z) constr(T) constr_list(L) := elpi foo ltac_string:(X) ltac_string:(T) ltac_int:(Z) (T) ltac_term_list(L).
foo "a" b 3 nat t1 t2 t3
in any Ltac context. For attributes one has to placeltac_attributes:(v)
in front ofelpi
, as in:Here the delimitersTactic Notation "foo" "#[" attributes(A) "]" := ltac_attributes:(A) elpi foo.
#[
and]
are chosen for consistency, you can use any "delimited" syntax really. The usual prefix notation is also possible with the following limitations due to a parsing conflicts in the Coq grammar (at the time of writing):Tactic Notation "#[" attributes(A) "]" "tac" := ltac_attributes:(A) elpi tac.
#[ att ] tac.
does not parse(#[ att ] tac).
worksidtac; #[ att ] tac.
works
- Change
-qua.lid
is no more understood as the string"-qua.lid"
but as two strings (when passed to a command, syntax error when passed to a tactic)
Requires Elpi 1.13.1 and Coq 8.13.
- New attribute
#[skip="rex"]
and#[only="rex"]
for theElpi Acumulate
family of commands which let one accumulate a piece of (compatibility) code only on some Coq versions.
Requires Elpi 1.13.1 and Coq 8.13.
- New
coq.reduction.lazy.norm
- New
coq.reduction.native.norm
- New
coq.reduction.native.available?
- Rename
coq.reduction.cbv.whd_all
->coq.reduction.cbv.norm
- Rename
coq.reduction.vm.whd_all
->coq.reduction.vm.norm
Requires Elpi 1.13 and Coq 8.13.
- Commands, Tactics and Db cannot be declared inside sections or modules (it never really worked, but now you get an error message).
- Clauses which are accumulated via
coq.elpi.accumulate
and are not@local!
survive section closing if they don't mention the section variables being discharged.
- Warnings can be turned into errors by passing Coq
-w +elpi.typecheck
.
- New
coq.CS.db-for
to filter the CS db given a projection or a canonical value, or both. - New
coq.warning
likecoq.warn
but with a category and name, so that the message can be silenced or turned into an error.
Requires Elpi 1.13 and Coq 8.13.
- Calls to APIs that only read the global state are much faster (thousands of times faster)
- Fix compilation with OCaml 4.12
- Fix issue with
coq.env.add-abbreviation
when given a term with binders having overlappingname
s. - New
copy-indt-decl
- New
coq.coercion.declare
is able to infer the endpoints if omitted
Requires Elpi 1.13 and Coq 8.13.
- Fix issue with async-mode (Elpi commands can change the parser)
- New
attmap
attribute type to represent associative maps over strings, eg#[foo(x = "a", y = "b")]
Requires Elpi 1.13 and Coq 8.13.
- Fix
elpi.loc
computation when run in interactive mode. - New
@using! S
attribute forcoq.env.add-const
akin to Coq's#[using=S]
.
Requires Elpi 1.13 and Coq 8.13.
- Fix
coq.env.add-section-variable
andcoq.env.add-axiom
were not handling universes correctly.
- New target
build
which only builds elpi and the apps - New target
test
which runs all tests for elpi and the apps - OPAM package only calls
test
only if requested, hence the package typically installs faster
Requires Elpi 1.13 and Coq 8.13.
- Fix
coq.env.indt-decl
to generate arecord-decl
for records.
- Fix issue with the compiler cache when used in async-mode (via CoqIDE or vscoq).
- New type
coq.pp
andcoq.pp.box
to describe Coq's pretty printer box model - New
coq.pp->string
to turn formatting boxes into a string - New
coq.term->pp
to turn formatting boxes into a string - New
@ppall!
attribute to print terms in full details - New
@ppmost!
attribute to print terms in a reparsable way - New
@ppwidth! N
attribute to specify the maximal line length when turning formatting boxes into strings - New
fold-map
to map a term with an accumulator - New
coq.env.add-section-variable
- New
coq.env.add-axiom
- Deprecate
coq.env.add-const
for declaring axioms or section variables. The deprecation warning is calledelpi.add-const-for-axiom-or-sectionvar
and can be turned into an error by passing tocoqc
the option-w +elpi.add-const-for-axiom-or-sectionvar
- The
COQ_ELPI_ATTRIBUTES=text
parsestext
as Coq attributes#[elpi(text)]
and passes them to all commands. Attributes in theelpi.
namespace are silently ignored by commands not using them. - Attribute
elpi.loc
carries theloc
of the command being run (if exported withElpi Export cmd
). This location does not comprise control flags (egFail
,Time
) nor attributes. This limitation will be lifted in Coq 8.14 (8.13 does not expose this parsing information to plugins).
Requires Elpi 1.12 and Coq 8.13.
- Illformed terms like
global (const X)
(which have no representation in Coq) are now reported with a proper error message. Whe passed tocoq.term->string
, instead of a fatal error, we pick for the illformed sub term theunknown_gref
special constant.
Requires Elpi 1.12 and Coq 8.12.
- New
@primitive!
attribute forcoq.env.add-indt
allowing one to declare primitive records. So far no term syntax for primitive projects is supported, their "non primitive" version is always used instead.
- Best effort support for Coq's
let (x, y, .. ) := t in
in quotations.
- Fix
coq.term->gref
skips over casts
Requires Elpi 1.12 and Coq 8.12.
- New
primitive (uint63 <i>)
term constructor - New
primitive (float64 <f>)
term constructor
- New
coq.reduction.lazy.whd_all
- New
coq.reduction.cbv.whd_all
- New
coq.reduction.vm.whd_all
- New
coq.env.const-primitive?
- Fix argument
const-decl
is accepted even if the name is "_", allowing one to writeElpi command Definition _ : type := body
- Fix
coq.notation.abbreviation
gives an error if too few arguments are provided
Major reorganization of sources:
- src/ is for .ml files
- elpi/ for .elpi files
- theories/ for .v files meant to be installed
- tests/ for the test suite, not to be installed
- examples/ for tests (not to be installed)
Moreover the apps/ directory is for applications written in Coq-Elpi, their structure follows the same convention
- POC application emulating name spaces on top of modules
- Use Elpi 1.12 API to implement a compiler cache and avoid recompiling over and over the same programs.
Requires Elpi 1.11 and Coq 8.12.
- Display failures generated by
std.assert!
as errors
- Use the new
coq.elaborate-skeleton
API to insert coercions
- Embedding for sorts was incorrectly mapping
Prop
tosprop
coq.env.add-const
made 8.12 friendly with a workaround for coq/coq#12759
- New
coq.elaborate-skeleton
andcoq.elaborate-ty-skeleton
that run Coq's elaborator on a term obtained by disregarding evars and universes in the given input. Unfortunately Coq's elaborator does not take terms as input, but glob terms, and the conversion function is not lossless. See alsolib:elpi.hole
. - New
coq.elaborate-indt-decl-skeleton
to elaborate an inductive type declaration. - New
coq.elaborate-arity-skeleton
to elaborate an arity. - New
coq.env.current-path
to get the current module path. - New
coq.modpath->path
andcoq.modpath->path
to get access to the components of a module path. - Change
coq.elpi.accumulate
understands the@local!
attribute, which makes the clausesLocal
to the module into which they live.
- New
lib:elpi.hole
constant that can be used in place of a unification variable to denote an implicit argument when callingcoq.*-skeleton
APIs
Requires Elpi 1.11 and Coq 8.12.
Locality is now supported by coq.CS.declare-instance
Requires Elpi 1.11 and Coq 8.11.
- New option
@holes!
to be assumed (as in@holes! => ...
) before calling any Coq API. When this option is given unknown unification variables are interpreted as "implicit arguments" (linear holes that see all the variables in scope). If the unification variable is outside the pattern fragment the following heuristic is applied: arguments that are not variables are heuristically dropped; arguments which are variables but occur multiple times are kept only once (the first occurrence is kept, the others are dropped).
- New
coq.arguments.set-default-implicit
that behaves likeArguments foo : default implicits
- Change of arguments of type
@global?
attributes@local!
or@global!
. In order to pass a locality directive one has to do something like@global! => coq.add-something
Locality is understood by:coq.TC.declare-instance
coq.coercion.declare
coq.arguments.set-implicit
coq.arguments.set-default-implicit
coq.arguments.set-name
coq.arguments.set-scope
coq.arguments.set-simplification
coq.notation.add-abbreviation
coq.env.add-const
- Change of argument for deprecation to attribute
@deprecated! Since Message
. In order to pass a deprecation directive one has to do something like@deprecated! "8.11.0" "use this instead" => coq.add-something
Deprecation is understood by:coq.notation.add-abbreviation
- New macro
@transparent!
with valueff
to be used withcoq.env.add-const
engine/elaborator.elpi
is now installed (but not used by default). One canElpi Accumulate "engine/elaborator.elpi".
in order to load it. It is too experimental to use it in production, but it is also hard to experiment with it without having it installed.
- Switch to Github Actions and Coq Community's Docker workflow
- anonymous record fields are not given a generated name anymore
coq.typecheck
andcoq.typecheck-ty
API now ensure that all unification problems required by type checking are actually solved by Coq's unifier- some debug printings used to raise errors in corner cases, now fixed
Minor fixes
- Missing opaque data type declaration for
abbreviation
(could lead to confusing type errors) - Parse also "keywords" where
qualified_name
is expected.Elpi Export x.
turnsx
into a keyword, and that used to break commandsElpi Something x ...
. Parsing of all commands is now resilient to this.
Requires Elpi 1.11 and Coq 8.11 or 8.12.
The main visible change is the indt-decl
data type that now faithfully
represents an inductive type declaration (including the implicit status of
parameters). Also all the predicates implemented in coq-lib
are now in
the coq.
namespace.
- New
coq.notation.abbreviation-body
to retrieve the number of arguments and body of a syntactic definition. - New
coq.id->name
to convert a relevant id into an irrelevant pretty printing hint. - New
coq.mk-n-holes
to produce a list of flexible terms. - New
coq.env.indt-decl
to read for the environment an inductive type represented in HOAS form coq.env.indt->decl
renamedcoq.build-indt-decl
- New
coq.env.rename-indt-decl
- Change
coq.env.add-indt
now sets the imlicit status of the inductive type and its constructors (since theparameter
constructor can carry it) - New
coq.arity->nparams
to count the number of parameters - Change
parse-attributes
made deterministic - Change
coq.unify-leq
andcoq.unify-eq
now return a diagnostic - Change
subst-prod
->coq.subst-prod
- Change
subst-fun
->coq.subst-fun
- Change
prod->fun
->coq.prod->fun
- Change
count-prods
->coq.count-prods
- Change
prod-R-fun
->coq.prod-R-fun
- Change
safe-dest-app
->coq.safe-dest-app
- Change
arity->sort
->coq.arity->sort
- Change
term->gref
->coq.term->gref
- Change
fresh-type
->coq.fresh-type
- Change
build-match
->coq.build-match
- Change
map-under-fun
->coq.map-under-fun
- Change
iter-under-fun
->coq.iter-under-fun
- Change
bind-ind-arity
->coq.bind-ind-arity
- Change
with-TC
->coq.with-TC
- Change
valid-attribute
->coq.valid-attribute
- Change
is-one-of
->coq.is-one-of
- Change
parse-attributes
->coq.parse-attributes
- Change
mk-app
->coq.mk-app
- Change
mk-app-uvar
->coq.mk-app-uvar
- Change
mk-eta
->coq.mk-eta
- New support for
Type@{name}
in Coq{{ quotations }}
. - Fix more precise promotion of universe variables to universe global names
in builtins changing the Coq environment (eg
coq.env.add-const
). - New user error when
coq.elpi.accumulate
is given a clause that mentions universe variables: only global universes can be stored in a DB.
- Change
indt-decl
:- the
parameter
constructor carries anid
,imlpicit_kind
and a type - the
coinductive
constructor was removed, theinductive
one carries abool
,tt
for inductive,ff
for coinductive - the
inductive
constructor no more carries the number of non uniform parameters, and the inductive type arity (see below) is no more a simple term but rather anarity
(all its parameters are non uniform) - the
constructor
constructor now carries anarity
so that non uniform parameters can be represented faitfully
- the
- New
arity
data type, constructors areparameter
(shared withindt-decl
) andarity
. - New
indt-decl
argument type introduced in version 1.3 now supports the syntax of inductive types (not just records). EgElpi command Inductive P {A} t : I := | K1 .. | K2 ..
. - Change
context-item
now carries anid
and animplicit-kind
- Change
const-decl
now carries an arity to describe the parameters of the definition in a faithful way - New
@pi-parameter ID Ty p\ ...
to postulate a nominalp
with typeTy
and a name built out of the idID
- New derivations
derive.invert
andderive.idx2inv
now called byderive
- New global command
derive
taking in input the name of an inductive or an inductive declaration. In the latter case all derivations are placed in a module named after the inductive
Port to Coq 8.11, two API changes:
field
constructor ofindt-decl
takes an argument of typefield-attributes
rather than a simplebool
. The macro@coercion!
works in both versions, as well as omitting the attribute using_
. In 8.11 it is possible to disable canonical inference for a field using the(canonical false)
attribute.coq.env.add-abbreviation
takes an extra argument (deprecation info). Code working on both versions can be obtained as follows:if (coq.version _ 8 10 _) (std.unsafe-cast coq.notation.add-abbreviation F, F ... Abbrev) (std.unsafe-cast coq.notation.add-abbreviation G, G ... Deprecation Abbrev).
Requires Elpi 1.10 and Coq 8.10 or 8.11.
The main visible change is that opaque data types such as @constructor
,
@inductive
and @constant
are now written without the @
, since Elpi now
supports the typeabbrev
directive.
The main invisible change is that code accumulated into commands and tactics is
"compiled" by Elpi once and forall in the context in which it is accumulated. As
a consequence Coq code inside {{quotations}}
is processed in that, and only
that, context of notations, scopes, etc. Data bases are compiled every time it
is needed in the current Coq context, hence quotations should be used with care.
The file coq-HOAS.elpi
is now distributed as part of coq-builtin.elpi
.
- New
Elpi Export command
to makecommand
available without theElpi
prefix. Elpi command
(exported or not) can now access Coq's attributes (the#[option]
thing). See the HOAS section below.- Coq keywords or symbols passed to command and tactics are interpreted as
strings even if not quoted. Eg
Elpi command =>
is the same ofElpi command "=>"
. - The identifiers
Record
,Definition
,Axioms
andContext
are now reserved (see the HOAS section below). In order to pass them (as strings) one has to quote them.
- New
coq.typecheck-ty
to typecheck a type (outputs a universe) - New
coq.env.import/export-module
. - New
coq.env.begin/end-section
. - New
coq.notation.abbreviation
to unfold an abbreviation. - New
coq.locate-abbreviation
to locate abbreviations. - New
coq.locate-any
that never fails and gives a list of possible interpretations (term, abbreviation, module, module type). - Rename
coq.env.typeof-gr
tocoq.env.typeof
. - Rename
term->gr
togerm->gref
. - Rename
coq.gr->*
tocoq.gref->*string*
- Change
coq.typecheck
andcoq.typecheck-indt-decl
so that they never fail and have a 3rd argument of typediagnostic
(from Elpi 1.9) to signal success or errors (that can be printed). - Change
coq.elpi.accumulate
so that one can put the clause either in current module from which the program is started, or in the current module while the program runs (which can be different if one uses thecoq.env.begin-module
API). - Remove
coq.elaborate
andcoq.elaborate-indt-decl
. - Fix
coq.typecheck T TY
to uses Coq's unification to equate the type inferred forT
andTY
(when it is provided by the user). - Fix
coq.CS.*
w.r.t. default instances of canonical structures. - Fix all APIs changing the Coq global state to abort if they are used from a tactic.
- Fix
coq.gr->string
to not duplicate the label part of the name
- Change context entry
def
to not carry a cache for the normal form of the defined term (now cached by a specificcache
context entry).def
now carries the exact same information of alet
, asdecl
carries the same information of afun
. - New
indt-decl
argument type with a concrete syntax that mimics the standard one for records. EgElpi command Record x := K { f : T }
. - New
const-decl
argument type with a concrete syntax that mimics the standard one for definitions or axioms. EgElpi command Definition x := t.
. - New
ctx-decl
argument type with a concrete syntax that mimics the standard one for contexts. EgElpi command Context T (x : T).
. - Add to the context under which
main
is run the list of attributes passed to the command invocation (Coq syntax is for example#[myflag]
). See theattribute-value
data type incoq-builtin.elpi
andparse-attributes
helper incoq-lib.elpi
.
- New
coq.gr->path
to get the path components as a list of strings - Failure of
coq.ltac.call
is now turned into logical failure, as any other Elpi tactic - Fix
coq.end.add-indt
in the case of record (was not flagging the inductive as such) - Fix
coq.version
, wrong parsing of beta versions - Expose
set
andmap
from Elpi 1.8 (generic data structure for ground terms)
- Improve reflexive tactic demo
- Fix documentation of
coq.gr->*
APIs coq-HOAS.elpi
,coq-lib.elpi
andcoq-builtin.elpi
are now installed since they provide useful doc (but are not needed by the runtime, since they are embedded inelpi.vo
)
-
interface made consistent with other derivations:
derive.param2
takes in input optional suffix, instead of the full name of the derived concept -
storage of previous derivations based on Elpi Db
-
the derivation generates nicer types for relators over fixpoints (the new types are convertible to the old ones, but the fixpoint is not expanded). PR #84 by Cyril Cohen
- Improved documentation of
coq.typecheck
- First public release