- Isabelle uses eager evaluation, so should the elaboration of terms
which are evaluated.
The value of instances are now registered in the data tables
of Isabelle/DOF when fully elaborated, ie,
term annotation antiquotations proposed by Isabelle/DOF in
an instance value are replaced by its value before registration
in Isabelle/DOF data
A new field, input_term, stores the lazy elaboration
and is used when elaboration is not wished
(to print the original input term declared by the user, for example)
- Clean up the simplication mechanism of the internal trace attribute
(used by monitor classes)
Loosen the dependency of the implementation of value* and term*
on Toplevel.transition.
Toplevel.transition should be avoided as it has specific behaviors
like only allowing atomic transactions.
- Warning: the current implementation does yet not support
some use-cases, like invariant on monitors,
or the initialization of docitem without a class associated.
- Add first draft of the checking of invariants.
For now, it is disabled by default because some cases
are not yet supported, like the initialization of docitem
without a class associated.
ex: text*[sdf]‹ Lorem ipsum @{thm refl}›
- To enable the checking, one can use the theory attribute
"invariants_checking" by declaring it in a theory like this:
declare [[invariants_strict_checking = true]]
- A checking using basic tactics (unfolding and auto) can be enable
with the "invariants_checking_with_tactics" theory attribute
for specific use-cases
- The specification of invariants is now automatically abstracted,
so one must define an invariant like this now:
doc_class W =
w::"int"
invariant w :: "w σ ≥ 3"
The old form:
doc_class W =
w::"int"
invariant w :: "λσ. w σ ≥ 3"
is now deprecated.
The specification of the invariant still uses the σ-notation
and is defined globally by the name component "invariantN"
- Update the invariants definition in the theories to match
the new implementation
- Update the manual to explain this new feature
- Add small examples in src/tests/High_Level_Syntax_Invariants.thy
and src/tests/Ontology_Matching_Example.thy
- Add a new Term Annotation Antiquotation (TA)
to allow requests on instances.
Example:
@{C-instances} will return all the instances of the class "C"
defined in the generated theory
- Update ISA_transformers elaborate function signature to
take into account the case where the term argument
of a TA is irrelevant, for example when a TA has no argument.
Example with the TA of the instances of a class:
@{A-instances}
Here the TA has no argument and none second level type checking is
wished, so its associated check function can be the identity function
with respect to the ISA_transformers chek function type.
- Add some request examples in Evaluation.thy
- Fix typos
- Fix the generation of the record associated with
a class and used for the logic.
The old implementation generated a new attribute
for each attribute defined by a subclass,
even the ones that were overriding ones of the superclass.
The new implementation generates the attributes of the subclass
which are not overriding ones.
Warning:
It implies that overridden attributes in a subclass are not
new attributes added to the theory context.
So the base name of an attribute will refer to the attribute
of the last declared class where it is defined.
If ones wants to refer to atttributes, one should use
long names, even in the invariants of a subclass definition
which overrides the attribute used in the invariant.
For example,
in ~~/src/ontologies/scholarly_paper/scholarly_paper.thy:
doc_class technical = text_section +
definition_list :: "string list" <= "[]"
status :: status <= "description"
formal_results :: "thm list"
invariant L1 :: "λσ::technical. the (level σ) > 0"
type_synonym tc = technical (* technical content *)
doc_class example = text_section +
referentiable :: bool <= True
status :: status <= "description"
short_name :: string <= "''''"
doc_class math_content = tc +
referentiable :: bool <= True
short_name :: string <= "''''"
status :: status <= "semiformal"
mcc :: "math_content_class" <= "thm"
invariant s1 :: "λ σ::math_content. ¬referentiable σ ⟶ short_name σ = ''''"
invariant s2 :: "λ σ::math_content. technical.status σ = semiformal"
The class math_content overrride the attribute status
of the class technical, by using the type synonym tc,
but the base name of this attribute refers
to the attribute of the class example where it is last defined
and not just overridden.
So in the invariant s2 of the class math_content,
we must use the long name of the attribute,
i.e. the base name "status" with its qualifier which refers
to the superclass where it is defined, the class technical.
Type synonyms as qualifiers are not yet supported.
- Qualify classes that only override attributes of their superclass
as vitual classes by adding a virtual attribute.
This attribute is used to discriminate virtual classes and generate
an adequate make function to initialize their associated record.
The implementation uses an hidden attribute (the tag_attribute)
to force the virtual class to be concrete or the logic
by having a full new record definition associated with it.
For example:
doc_class W =
a::"int" <= "1"
doc_class X = W +
a::"int" <= "2"
The class X is tagged as a virtual class and
the record make functions of the classes W and X are:
W.make W_tag_attribute W_a
X.make X_tag_attribute X_a X_tag_attribute
So a record definition is added to the theory context for each class,
even though a virtual class only overrides
attributes of its superclass.
This behavior allows us to support definitions of new default values
for attributes in the subclass, as shown in the example.
- Factorize make name components
- Use Record name components instead of strings to refer to Record
components
- Fix typos
- Add a first implementation of a referential equivalence
for the built-ins term annotations (TA)
- Some built-ins remain as unspecified constants:
- the docitem TA offers a way to check the reference of
class instances without checking the instances type.
It must be avoided for certification
- the termrepr TA is left as an unspecified constant for now.
A major refactoring of code should be done to enable
referential equivalence for termrepr, by changing the dependency
between the Isa_DOF theory and the Assert theory.
The assert_cmd function in Assert should use the value* command
functions, which make the elaboration of the term
referenced by the TA before passing it to the evaluator
- Update the Evaluation test theory to test the referential equivalence
and expose some of current implementation limitations
- Add a warning about the docitem TA in the TermAntiquotations theory
The philosophy is for the tag attribute to be unique
for each class.
So this commit updates the implementation of this attribute
to match the philosophy.
The previous implementation associated a tag attribute
with a class but also with each super-class of this class
up to the top (default) class "text".
Now a class with super-classes has only one tag attribute.
Add a command value*
- The value* command uses the same code as the value command
and adds the possibility to evaluate
Term Annotation Antiquotations (TA)
with the help of the DOF_core.transduce_term_global function.
The DOF_core.transduce_term_global function,
in addition to the validation of a term
(also called a second level type checking),
is now able to make a so called elaboration:
it will construct the term referenced by a TA before
passing it to the evaluator.
- For a term to be evaluated, it must not be contain
the "undefined" constant whose evaluation always fails.
(See the code generation documentation).
Furthermore, the instance class generation is updated in such a way
that each of its attributes is initialized with a free variable
whose name shows to the final user that this attribute
is not initialized.
It implies that an instance class evaluation will be pass
to the normalization by evaluation (nbe) evaluator by default
if the final user does not specify a class instance entirely,
i.e. by specifying each attribute of the instance.
This choice is considered a decent compromise, considering
the speed and trustworthiness of the nbe evaluator.
(See the article
A Compiled Implementation of Normalization by Evaluation from 2008)
- Update the ISA transformer tab to add a function
which is used for the elaboration of the term referenced by the TA.to pass
- Add a first really basic draft of the implementation
of the elaboration of the built-ins TA and of an instance class:
- For the built-ins, the term referenced by the TA is returned
as it is;
- For an instance class, the value of the instance is returned.
- Make the tag attribute global by moving it to DOF_core structure
- Add a first draft for some evaluation tests
and expose the limitations of the current implementation
in Evaluation.thy
- Add a term antiquotation for document classes
and add the term* command which mimics the classical term command
and adds the possibility to use a term antiquotation
which references document classes:
one can use @{A ''A_instance''} to reference
an instance of the class A in a term* command.
- Reuse and update the ML_isa_check_generic visitor pattern
to add the function which checks the class instance of a class,
used by the term antiquotation for document classes.
Also, the update_isa functions now expect long name
(See builtin term antiquotations setup)
- The merge of ISA_transformer_tab has been update to avoid conflicts.
Indeed, the merge is ultra-critical: the transformer tabs were
just extended by letting *the first* entry
with the same long-name win.
Since the range is a (call-back) function, a comparison on its content
is impossible and some choice has to be made.
An alternative may be to use Symtab.join
- As classes names as constants are already bound to
"doc_class Regular_Exp.rexp" constants by add_doc_class_cmd function,
we use a prefix "doc_class_" when adding
document classes term antiquotations
Nicolas Méric
Achim D. Brucker
Burkhart Wolff
Makarius Wenzel
Nicolas Méric
Burkhart Wolff
Simon Foster