This includes a few separate changes:
- pass visibility information of declarations (depending on wether the
declaration was in a ```catala-metadata block or not)
- add reasonable hash computation functions to discriminate the interfaces. In
particular:
* Uids have a `hash` function that depends on their string, but not on their
actual uid (which is not stable between runs of the compiler) ; the existing
`hash` function and its uses have been renamed to `id`.
* The `Hash` module provides the tools to properly combine hashes, etc. While
we rely on `Hashtbl.hash` for the atoms, we take care not to use it on any
recursive structure (it relies on a bounded traversal).
- insert the hashes in the artefacts, and properly check and report those (for
OCaml)
**Remains to do**:
- Record and check the hashes in the other backends
- Provide a way to get stable inline-test outputs in the presence of module
hashes
- Provide a way to write external modules that don't break at every Catala
update.
They are to become citizens of the same class if we want to allow
output-subscopes (without unnecessary complications like deconstructing and
reconstructing the same structure). And it's reasonable to assume that they
share the same namespace.
With this we should shortly collapse the (internal) ambiguity between
- `subscope.subvar`: access to a variable within a subscope
- `subscope.subfield`: access to a field of the output structure contained in a
subscope variable
With the subscope a variable, these should now become strictly equivalent, so
the plan is that the first could be removed.
This allows for retyping after monomorphisation: a new function just extracts
the return type of the operator, without checking the operand types.
Also to avoid multiplying function arguments around the typer, the flags have
been gathered in a record that is included in the typing environment; it's ok to
give them default values as long as these are the strictest.
This changes the `decl_ctx` to be toplevel only, with flattened references to
uids for most elements. The module hierarchy, which is still useful in a few
places, is kept separately.
Module names are also changed to UIDs early on, and support for module aliases
has been added (needs testing).
This resolves some issues with lookup, and should be much more robust, as well
as more convenient for most lookups.
The `decl_ctx` was also extended for string ident lookups, which avoids having
to keep the desugared resolution structure available throughout the compilation
chain.
rather than scattered in structures
The context is still hierarchical for defs though, so one needs to retrieve the
path to lookup in the correct context for info. Exceptions are enums and struct
defs, which are re-exposed at toplevel.
Two interdependent changes here:
1. Enforce all instances of Shared_ast.gexpr to use the generic type for marks.
This makes the interfaces a tad simpler to manipulate: you now write
`('a, 'm) gexpr` rather than `('a, 'm mark) gexpr`.
2. Define a polymorphic `Custom` mark case for use by pass-specific annotations.
And leverage this in the typing module
The module is renamed to `Mark`, and functions renamed to avoid redundancy:
`Marked.mark` is now `Mark.add`
`Marked.unmark` is now `Mark.remove`
`Marked.map_under_mark` is now simply `Mark.map`
etc.
`Marked.same_mark_as` is replaced by `Mark.copy`, but with the arguments
swapped (which seemed more convenient throughout)
Since a type `Mark.t` would indicate a mark, and to avoid confusion, the type
`Marked.t` is renamed to `Mark.ed` as a shorthand for `Mark.marked` ; this part
can easily be removed if that's too much quirkiness.
This is just a bunch of `sed` calls:
```shell
sed -i 's/ScopeSet/ScopeName.Set/g' compiler/**/*.ml*
sed -i 's/ScopeMap/ScopeName.Map/g' compiler/**/*.ml*
sed -i 's/StructMap/StructName.Map/g' compiler/**/*.ml*
sed -i 's/StructSet/StructName.Set/g' compiler/**/*.ml*
sed -i 's/EnumMap/EnumName.Map/g' compiler/**/*.ml*
sed -i 's/EnumSet/EnumName.Set/g' compiler/**/*.ml*
sed -i 's/StructFieldName/StructField/g' compiler/**/*.ml*
sed -i 's/StructFieldMap/StructField.Map/g' compiler/**/*.ml*
sed -i 's/StructFieldSet/StructField.Set/g' compiler/**/*.ml*
sed -i 's/EnumConstructorMap/EnumConstructor.Map/g' compiler/**/*.ml*
sed -i 's/EnumConstructorSet/EnumConstructor.Set/g' compiler/**/*.ml*
sed -i 's/RuleMap/RuleName.Map/g' compiler/**/*.ml*
sed -i 's/RuleSet/RuleName.Set/g' compiler/**/*.ml*
sed -i 's/LabelMap/LabelName.Map/g' compiler/**/*.ml*
sed -i 's/LabelSet/LabelName.Set/g' compiler/**/*.ml*
sed -i 's/ScopeVarMap/ScopeVar.Map/g' compiler/**/*.ml*
sed -i 's/ScopeVarSet/ScopeVar.Set/g' compiler/**/*.ml*
sed -i 's/SubScopeNameMap/SubScopeName.Map/g' compiler/**/*.ml*
sed -i 's/SubScopeNameSet/SubScopeName.Set/g' compiler/**/*.ml*
```
... and reformat
Many changes got bundled in here and would be too tedious to separate.
Closes#330
See changes in `shared_ast/definitions.ml` to check the main point.
- the biggest change is a modification of the struct and enum types in
expressions: they are now stored as `Map`s throughout passes, and no longer
converted to indexed lists after scopelang. Their accessors are also changed,
and tuples only exist in Lcalc (they're used for closure conversion).
This implied adding some more information in the contexts, to keep the mapping
between struct fields and scope output variables. It should also be much more
robust (no longer relying on assumptions upon different orderings).
- another very pervasive change is more cosmetic: the rewrite of the main AST to
use inline records, labelling individual subfields.
- moved the checks for correct definitions and accesses of structures from
`Scope_to_dcalc` to `Typing`
- defining some new shallow iterators in module `Shared_ast.Expr`, and
factorising a few same-pass rewriting functions accordingly (closure
conversion, optimisations, etc.)
- some smaller style improvements (ensuring we use the proper compare/equal
functions instead of `=` in a few `when` closes, for example)
Quite a few changes are included here, some of which have some extra
implications visible in the language:
- adds the `Scope of { -- input_v: value; ... }` construct in the language
- handle it down the pipeline:
* `ScopeCall` in the surface AST
* `EScopeCall` in desugared and scopelang
* expressions are now traversed to detect dependencies between scopes
* transformed into a normal function call in dcalc
- defining a scope now implicitely defines a structure with the same name, with
the output variables of the scope defined as fields. This allows us to type
the return value from a scope call and access its fields easily.
* the implications are mostly in surface/name_resolution.ml code-wise
* the `Scope_out` struct that was defined in scope_to_dcalc is no longer
needed/used and the fields are no longer renamed (changes some outputs; the
explicit suffix for variables with multiple states is ignored as well)
* one benefit is that disambiguation works just like for structures when there
are conflicts on field names
* however, it's now a conflict if a scope and a structure have the same
name (side-note: issues with conflicting enum / struct names or scope
variables / subscope names were silent and are now properly reported)
- you can consequently use scope names as types for variables as well. Writing
literals is not allowed though, they can only be obtained by calling the
scope.
Remaining TODOs:
- context variables are not handled properly at the moment
- error handling on invalid calls
- tests show a small error message regression; lots of examples will need
tweaking to avoid scope/struct name or struct fields / output variable
conflicts
- add a `->` syntax to make struct field access distinct from scope output var
access, enforced with typing. This is expected to reduce confusion of users
and add a little typing precision.
- document the new syntax & implications (tutorial, cheat-sheet)
- a consequence of the changes is that subscope variables also can now be typed.
A possible future evolution / simplification would be to rewrite subscopes as
explicit scope calls early in the pipeline. That could also allow to manipulate
them as expressions (bind them in let-ins, return them...)
This was the only reasonable solution I found to the issue raised
[here](https://github.com/CatalaLang/catala/pull/334#discussion_r987175884).
This was a pretty tedious rewrite, but it should now ensure we are doing things
correctly. As a bonus, the "smart" expression constructors are now used
everywhere to build expressions (so another refactoring like this one should be
much easier) and this makes the code overall feel more
straightforward (`Bindlib.box_apply` or `let+` no longer need to be visible!)
---
Basically, we were using values of type `gexpr box = naked_gexpr marked box`
throughout when (re-)building expressions. This was done 99% of the time by
using `Bindlib.box_apply add_mark naked_e` right after building `naked_e`. In
lots of places, we needed to recover the annotation of this expression later on,
typically to build its parent term (to inherit the position, or build the type).
Since it wasn't always possible to wrap these uses within `box_apply` (esp. as
bindlib boxes aren't a monad), here and there we had to call `Bindlib.unbox`,
just to recover the position or type. This had the very unpleasant effect of
forcing the resolution of the whole box (including applying any stored closures)
to reach the top-level annotation which isn't even dependant on specific
variable bindings. Then, generally, throwing away the result.
Therefore, the change proposed here transforms
- `naked_gexpr marked Bindlib.box` into
- `naked_gexpr Bindlib.box marked` (aliased to `boxed_gexpr` or `gexpr boxed` for
convenience)
This means only
1. not fitting the mark into the box right away when building, and
2. accessing the top-level mark directly without unboxing
The functions for building terms from module `Shared_ast.Expr` could be changed
easily. But then they needed to be consistently used throughout, without
manually building terms through `Bindlib.apply_box` -- which covers most of the
changes in this patch.
`Expr.Box.inj` is provided to swap back to a box, before binding for example.
Additionally, this gives a 40% speedup on `make -C examples pass_all_tests`,
which hints at the amount of unnecessary work we were doing --'
Note that this is incomplete in the case of desugared/scopelang because we only
have typing for expressions yet, and the scope/program structure is different.
The code allows passing an environment of types for scope/subscope variables in
order to resolve `ELocation` terms, but that's unused until we implement
scopelang typing at the scope level.
This gives further uniformity in their interfaces and allows more common
handling.
The next step will be for all the `Expr.make_*` functions to work on expressions
annotated with the `'a mark` type, correctly propagating type information when
it is present. Then we could even imagine early propagation of type
information (without complete inference), which could for example be used for
overloaded operator disambiguation.
