CAMP is a small calculus for pattern matching. Its purpose is to
capture the semantics of rules languages that rely on the ability to
recover from match failure.
Two distinct evaluation functions for CAMP are defined. The first
is a standard evaluation function and serves as semantic
definition. The second is equivalent but maintains a trace of the
evaluation for debugging purposes.
Summary:
-
Language: CAMP (Calculus for Aggregating Matching Patterns)
-
Based on: "A Pattern Calculus for Rule Languages:
Expressiveness, Compilation, and Mechanization" Avraham
Shinnar, Jérôme Siméon, and Martin Hirzel. ECOOP'2015.
-
URL: http://drops.dagstuhl.de/opus/volltexte/2015/5237/
-
Languages translating to CAMP: TechRule, DesignRule, CAMPRule, cNNRC
-
Languages translating from CAMP: NRA, cNRAEnv, NRAEnv
Require Import String.
Require Import List.
Require Import EquivDec.
Require Import Morphisms.
Require Import Utils.
Require Import DataRuntime.
Require Import CAMPUtil.
Declare Scope camp_scope.
Section CAMP.
Local Open Scope string.
Context {
fruntime:
foreign_runtime}.
Abstract Syntax Tree
CAMP patterns are pure expressions. They match against a current
value in the context of an environment. The environment has a
globale component which is static, and a local component which can
be modified in the course of evaluation.
Operations include: pmap, a functional map which skips match
failure; passert, which turns match failure into a terminal failure,
and porElse which allows to continue to a new pattern in case of
match failure.
Compared to the published material, three new operators are
added: pgetConstant which accesses the global environment, pleft and
pright which support pattern matching against choice values.
Inductive camp :
Set :=
|
pconst :
data ->
camp (* Constant value *)
|
punop :
unary_op ->
camp ->
camp (* Unary operators *)
|
pbinop :
binary_op ->
camp ->
camp ->
camp (* Binary operators *)
|
pmap :
camp ->
camp (* Functional map *)
|
passert :
camp ->
camp (* Assert pattern-matching success *)
|
porElse :
camp ->
camp ->
camp (* Recover from failure *)
|
pit :
camp (* Current value being matched *)
|
pletIt :
camp ->
camp ->
camp (* Set the current value *)
|
pgetConstant :
string ->
camp (* Access to a global variable *)
|
penv :
camp (* Current environment *)
|
pletEnv :
camp ->
camp ->
camp (* Add bindings to the environment *)
|
pleft :
camp (* Matches if the value is a left choice *)
|
pright :
camp.
(* Matches if the value is a right choice *)
Tactic Notation "
camp_cases"
tactic(
first)
ident(
c) :=
first;
[
Case_aux c "
pconst"%
string
|
Case_aux c "
punop"%
string
|
Case_aux c "
pbinop"%
string
|
Case_aux c "
pmap"%
string
|
Case_aux c "
passert"%
string
|
Case_aux c "
porElse"%
string
|
Case_aux c "
pit"%
string
|
Case_aux c "
pletIt"%
string
|
Case_aux c "
pgetConstant"%
string
|
Case_aux c "
penv"%
string
|
Case_aux c "
pletEnv"%
string
|
Case_aux c "
pleft"%
string
|
Case_aux c "
pright"%
string].
Equality between two CAMP patterns is decidable.
Global Instance camp_eqdec :
EqDec camp eq.
Proof.
Evaluation Semantics
Evaluation takes a camp pattern, a global environment, a local
environment and a current value. It returns a presult which can be
either a successful evaluation holding a value, a match failure,
or a terminal failure.
The external context for evaluation includes a brand relation,
and a global environment.
Section Evaluation.
Context (
h:
brand_relation_t).
Context (
constant_env:
bindings).
Fixpoint camp_eval (
p:
camp) (
bind:
bindings) (
d:
data) :
presult data
:=
match p with
|
pconst d' =>
Success (
normalize_data h d')
|
punop op p₁ =>
bindpr (
camp_eval p₁
bind d)
(
fun d' => (
op2tpr (
unary_op_eval h op d')))
|
pbinop op p₁
p₂ =>
bindpr (
camp_eval p₁
bind d)
(
fun d'₁ =>
bindpr (
camp_eval p₂
bind d)
(
fun d'₂ => (
op2tpr (
binary_op_eval h op d'₁
d'₂))))
|
pmap p₁ =>
match d with
|
dcoll l =>
liftpr dcoll (
gather_successes (
map (
camp_eval p₁
bind)
l))
|
_ =>
TerminalError
end
|
passert p₁ =>
bindpr (
camp_eval p₁
bind d)
(
fun d' =>
match d'
with
|
dbool true =>
Success (
drec nil)
|
dbool false =>
RecoverableError
|
_ =>
TerminalError
end)
|
porElse p₁
p₂ =>
match camp_eval p₁
bind d with
|
TerminalError =>
TerminalError
|
RecoverableError =>
camp_eval p₂
bind d
|
Success x =>
Success x
end
|
pit =>
Success d
|
pletIt p₁
p₂ =>
bindpr (
camp_eval p₁
bind d) (
camp_eval p₂
bind)
|
pgetConstant s =>
op2tpr (
edot constant_env s)
|
penv =>
Success (
drec bind)
|
pletEnv p₁
p₂ =>
bindpr (
camp_eval p₁
bind d)
(
fun rd'₁ =>
match rd'₁
with
|
drec d'₁ =>
match merge_bindings bind d'₁
with
|
Some bind' =>
camp_eval p₂
bind'
d
|
None =>
RecoverableError
end
|
_ =>
TerminalError
end)
|
pleft =>
match d with
|
dleft d' =>
Success d'
|
dright _ =>
RecoverableError
|
_ =>
TerminalError
end
|
pright =>
match d with
|
dright d' =>
Success d'
|
dleft _ =>
RecoverableError
|
_ =>
TerminalError
end
end.
End Evaluation.
Pretty Printing
Evaluation traces rely on printing support for CAMP's abstract syntax.
Global Instance ToString_camp :
ToString camp
:= {
toString :=
fix toStringp (
p:
camp) :
string :=
match p with
|
pconst d => "(
pconst " ++
toString d ++ ")"
|
punop u p1 => "(
punop " ++
toString u ++ " " ++
toStringp p1 ++ ")"
|
pbinop b p1 p2 => "(
pbinop " ++
toString b ++ " " ++
toStringp p1++ " " ++
toStringp p2 ++ ")"
|
pmap p1 => "(
pmap " ++
toStringp p1 ++ ")"
|
passert p1 => "(
passert " ++
toStringp p1 ++ ")"
|
porElse p1 p2 => "(
porElse " ++
toStringp p1++ " " ++
toStringp p2 ++ ")"
|
pit => "
pit"
|
pletIt p1 p2 => "(
pletIt " ++
toStringp p1++ " " ++
toStringp p2 ++ ")"
|
pgetConstant s => "(
pgetConstant " ++
s ++ ")"
|
penv => "
penv"
|
pletEnv p1 p2 => "(
pletEnv " ++
toStringp p1++ " " ++
toStringp p2 ++ ")"
|
pleft => "
pleft"
|
pright => "
pright"
end
}.
Fixpoint toString_camp_with_path (
p:
camp) (
loc:
camp_src_path) :=
match loc with
|
nil =>
string_bracket "<<<" (
toString p) ">>>"
|
pos::
loc' =>
match p with
|
pconst d => "(
pconst " ++
toString d ++ ")"
|
punop u p1 => "(
punop " ++
toString u ++ " " ++
toString_camp_with_path p1 loc' ++ ")"
|
pbinop b p1 p2 => "(
pbinop "
++
toString b ++ " " ++
(
if pos == 0
then toString_camp_with_path p1 loc'
else toString p1)
++ " " ++
(
if pos == 1
then toString_camp_with_path p2 loc'
else toString p2) ++ ")"
|
pmap p1 => "(
pmap " ++
toString_camp_with_path p1 loc' ++ ")"
|
passert p1 => "(
passert " ++
toString_camp_with_path p1 loc' ++ ")"
|
porElse p1 p2 => "(
porElse " ++
(
if pos == 0
then toString_camp_with_path p1 loc'
else toString p1)
++ " " ++
(
if pos == 1
then toString_camp_with_path p2 loc'
else toString p2) ++ ")"
|
pit => "
pit"
|
pletIt p1 p2 => "(
pletIt " ++
(
if pos == 0
then toString_camp_with_path p1 loc'
else toString p1)
++ " " ++
(
if pos == 1
then toString_camp_with_path p2 loc'
else toString p2) ++ ")"
|
pgetConstant s => "(
pgetConstant " ++
s ++ ")"
|
penv => "
penv"
|
pletEnv p1 p2 => "(
pletEnv " ++
(
if pos == 0
then toString_camp_with_path p1 loc'
else toString p1)
++ " " ++
(
if pos == 1
then toString_camp_with_path p2 loc'
else toString p2) ++ ")"
|
pleft => "
pleft"
|
pright => "
pright"
end
end.
Traced Evaluation
An alternative version of the compiler that keeps debug
information. While this forces some code duplication, this is
mitigated by the fact that the evaluation code is relatively
small and simple.
The context is the same as for default evaluation, with an
additional printing flag.
Section EvaluationDebug.
Context (
h:
brand_relation_t).
Context (
constant_env:
list(
string*
data)).
Context (
print_env:
bool).
The following functions are used to produce error messages.
Definition mk_err (
desc:
string) (
p:
camp) (
bind:
bindings) (
it:
data)
:=
toString p ++ "
failed" ++
desc ++ "." ++
(
if print_env
then "\
n Current environment (
env): "
++
toString (
drec bind)
else "") ++
"\
n Current scrutinee (
it): "
++
toString it ++
"\
n".
Definition punop_err (
p:
camp) (
bind:
bindings) (
it:
data) (
d:
data) :
string
:= (
mk_err ""
p bind it ++ "
The operator'
s argument was: " ++
toString d ++ "\
n").
Definition binop_err (
p:
camp) (
bind:
bindings) (
it:
data) (
d'₁
d'₂:
data) :
string
:= (
mk_err ""
p bind it ++ "
The operator'
s first argument was: " ++
toString d'₁
++ "\
n The operator'
s second argument was: " ++
toString d'₂
++ "\
n").
Definition pmap_err (
p:
camp) (
bind:
bindings) (
it:
data) :
string
:=
mk_err "
because the scrutinee was not a collection"
p bind it.
Definition passert_err (
p:
camp) (
bind:
bindings) (
it d:
data) :
string
:=
mk_err "
because the argument was not a boolean"
p bind it
++ "
The argument to passert was: " ++
toString d
++ "\
n".
Definition pletEnv_err (
p:
camp) (
bind:
bindings) (
it d:
data) :
string
:=
mk_err "
because its first argument was not a record"
p bind it
++ "
The first argument to pletEnv was: " ++
toString d.
Definition pleft_err (
bind:
bindings) (
it:
data) :
string
:=
mk_err "
because the scrutinee was an incompatible type"
pleft bind it.
Definition pright_err (
bind:
bindings) (
it:
data) :
string
:=
mk_err "
because the scrutinee was an incompatible type"
pright bind it.
Definition pgetConstant_err s (
bind:
bindings) (
it:
data) :
string
:=
mk_err ("
because the given field (" ++
s ++
")
is not a valid constant") (
pgetConstant s)
bind it
++ "
The set of constants for this execution is: " ++
(
string_bracket "{" (
String.concat "; " (
domain constant_env)) "}")
++ "\
n".
The alternative traced evaluation is defined as follows.
Fixpoint camp_eval_debug (
loc:
camp_src_path) (
p:
camp) (
bind:
bindings) (
d:
data) :
presult_debug data
:=
match p with
|
pconst d' =>
Success_debug (
normalize_data h d')
|
punop op p₁ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug d' =>
match unary_op_eval h op d'
with
|
None =>
TerminalError_debug (
punop_err (
punop op p₁)
bind d d')
loc
|
Some x =>
Success_debug x
end
end
|
pbinop op p₁
p₂ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug d'₁ =>
match camp_eval_debug (1::
loc)
p₂
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug d'₂ =>
match binary_op_eval h op d'₁
d'₂
with
|
None =>
TerminalError_debug (
binop_err (
pbinop op p₁
p₂)
bind d d'₁
d'₂)
loc
|
Some x =>
Success_debug x
end
end
end
|
pmap p₁ =>
match d with
|
dcoll l =>
liftpr_debug dcoll (
gather_successes_debug (
map (
camp_eval_debug (0::
loc)
p₁
bind)
l))
|
_ =>
TerminalError_debug (
pmap_err (
pmap p₁)
bind d)
loc
end
|
passert p₁ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug d' =>
match d'
with
|
dbool true =>
Success_debug (
drec nil)
|
dbool false =>
RecoverableError_debug "
assertion failure"
|
_ =>
TerminalError_debug (
passert_err (
passert p₁)
bind d d')
loc
end
end
|
porElse p₁
p₂ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug _ =>
camp_eval_debug (1::
loc)
p₂
bind d
|
Success_debug x =>
Success_debug x
end
|
pit =>
Success_debug d
|
pletIt p₁
p₂ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug x =>
camp_eval_debug (1::
loc)
p₂
bind x
end
|
pgetConstant s =>
match edot constant_env s with
|
Some x =>
Success_debug x
|
None =>
TerminalError_debug (
pgetConstant_err s bind d)
loc
end
|
penv =>
Success_debug (
drec bind)
|
pletEnv p₁
p₂ =>
match camp_eval_debug (0::
loc)
p₁
bind d with
|
TerminalError_debug s loc' =>
TerminalError_debug s loc'
|
RecoverableError_debug s =>
RecoverableError_debug s
|
Success_debug rd'₁ =>
match rd'₁
with
|
drec d'₁ =>
match merge_bindings bind d'₁
with
|
Some bind' =>
camp_eval_debug (1::
loc)
p₂
bind'
d
|
None =>
RecoverableError_debug "
bindings could not be unfied"
end
|
_ =>
TerminalError_debug (
pletEnv_err (
pletEnv p₁
p₂)
bind d rd'₁)
loc
end
end
|
pleft =>
match d with
|
dleft d' =>
Success_debug d'
|
dright _ =>
RecoverableError_debug "
pleft called on a pright"
|
_ =>
TerminalError_debug (
pleft_err bind d)
loc
end
|
pright =>
match d with
|
dright d' =>
Success_debug d'
|
dleft _ =>
RecoverableError_debug "
pright called on a pleft"
|
_ =>
TerminalError_debug (
pright_err bind d)
loc
end
end.
The following theorem states that traced evaluation is equivalent to regular evaluation.
Theorem camp_eval_debug_correct (
loc:
camp_src_path) (
p:
camp) (
bind:
bindings) (
d:
data) :
presult_same
(
camp_eval h constant_env p bind d)
(
camp_eval_debug loc p bind d).
Proof.
End EvaluationDebug.
Toplevel
Top-level evaluation functions are used externally by the Q*cert
compiler. They take a CAMP pattern and a global environment as
input. The initial local environment is set to an empty record, and
the initial current value to unit.
The result of evaluation is lifted back from presult to and
optional value for consistency with evaluation functions for the
other intermediate languages.
Section Top.
Context (
h:
brand_relation_t).
Definition presult_to_result (
pr:
presult data) :
option data :=
match pr with
|
Success l =>
Some (
dcoll (
l::
nil))
|
RecoverableError =>
Some (
dcoll nil)
|
TerminalError =>
None
end.
Definition camp_eval_top_to_presult (
q:
camp) (
global_env:
bindings) :
presult data :=
camp_eval h (
rec_sort global_env)
q nil dunit.
The main top-level evaluation function for CAMP is as follows.
Definition camp_eval_top (
q:
camp) (
global_env:
bindings) :
option data :=
presult_to_result (
camp_eval_top_to_presult q global_env).
Definition presult_to_result_debug (
pr:
presult_debug data) :
option data :=
match pr with
|
Success_debug l =>
Some (
dcoll (
l::
nil))
|
RecoverableError_debug _ =>
Some (
dcoll nil)
|
TerminalError_debug _ _ =>
None
end.
Definition pr2op_debug (
pr:
presult_debug data) :
option data :=
match pr with
|
Success_debug l =>
Some (
dsome l)
|
RecoverableError_debug _ =>
Some dnone
|
TerminalError_debug _ _ =>
None
end.
Definition camp_eval_top_debug_to_presult_debug
(
debug:
bool) (
q:
camp) (
global_env:
bindings) :
presult_debug data :=
camp_eval_debug h (
rec_sort global_env)
debug nil q nil dunit.
Theorem camp_eval_top_debug_correct (
debug:
bool) (
p:
camp) (
global_env:
bindings) (
d:
data) :
presult_same
(
camp_eval_top_to_presult p global_env)
(
camp_eval_top_debug_to_presult_debug debug p global_env).
Proof.
Definition camp_eval_top_debug_to_data
(
debug:
bool) (
q:
camp) (
global_env:
bindings) :
option data :=
presult_to_result_debug (
camp_eval_debug h (
rec_sort global_env)
debug nil q nil dunit).
The main top-level traced evaluation function for CAMP is as follows.
Definition camp_eval_top_debug (
debug:
bool) (
q:
camp) (
global_env:
bindings) :
string :=
print_presult_debug q toString_camp_with_path
(
camp_eval_top_debug_to_presult_debug debug q global_env).
End Top.
End CAMP.
Tactic Notation "
camp_cases"
tactic(
first)
ident(
c) :=
first;
[
Case_aux c "
pconst"%
string
|
Case_aux c "
punop"%
string
|
Case_aux c "
pbinop"%
string
|
Case_aux c "
pmap"%
string
|
Case_aux c "
passert"%
string
|
Case_aux c "
porElse"%
string
|
Case_aux c "
pit"%
string
|
Case_aux c "
pletIt"%
string
|
Case_aux c "
pgetConstant"%
string
|
Case_aux c "
penv"%
string
|
Case_aux c "
pletEnv"%
string
|
Case_aux c "
pleft"%
string
|
Case_aux c "
pright"%
string].