* [CameLIGO/ReasonLIGO] The AST node [EAnnot] (expressions
annotated by a type) now records the region in the source
code for the colon.
* [CameLIGO/ReasonLIGO/PascaLIGO] I added the syntax
%token <...> TOKEN "lexeme"
* [ReasonLIGO] I changed the AST nodes [Mtz] and [Str] to
[Mutez] and [String], respectively (in accordance with the
PascaLIGO front-end). I changed token [DOTDOTDOT] to
[ELLIPSIS].
* [ReasonLIGO] I added what was missing to make a loca build
with my Makefile.
Refactoring of parsing command-line arguments
* The type [options] is now abstract and implemented as an
object type to avoid struggling with scoping and type
inference when record types share some common field names.
Refactoring of ParserLog for PascaLIGO and CameLIGO
* The immediate motivation behind that refactoring was to
remove the use of a couple of global references. A
consequence is that we have a nicer and more compact code, by
threading a state. The files [pascaligo/Tests/pp.ligo] and
[ligodity/Tests/pp.mligo].
* Another consequence is that the choice of making strings from
AST nodes depends on the CLI (offsets? mode?). After this
refactoring, that choice is hardcoded in the simplifiers in a
few places (TODO), waiting for a general solution that would
have all CL options flow through the compiler.
* I removed the use of vendors [x_option.ml], [x_map.ml] and
[x_list.ml] when handling optional values. (Less dependencies
this way.)
Refactoring of the ASTs
* I removed the node [local_decl], which was set to [[]]
already in a previous commit (which removed local
declarations as being redundant, as statements could already
be instructions or declarations).
* I changed [StrLit] to [String] in the AST of CameLIGO and
ReasonLIGO.
* I also changed the type [fun_expr] so now either a block is
present, and therefore followed by the [with] keyword, or it
is not. (Before, the presence of a block was not enforced in
the type with the presence of the keyword.)
Notes
* [LexerMain.ml] and [ParserMain.ml] for CameLIGO and PascaLIGO
are almost identical and differ in the same way (language
name and file extension), which suggests that they should be
in the [shared] folder and instanciated as a functor in the
future (TODO).
* I removed the blank characters at the end of many lines in
the parser of ReasonLIGO.
- AST.ml/AST.mli:
- The AST now distinguishes the constructors `None` and `Some` as being
predefined, as in PascaLIGO. See type `AST.constr_pattern`.
- I removed the nodes specific to Liquidity,
e.g. `let%entry`, and, in particular, the natural literals
ending with `p`. Now it should be `n`, as in `10n`.
- I renamed the node `TAlias` to `TVar`.
- I have applied the rule of expanding type expressions after `of` when
those were not records.
- The type of the argument to a data constructor is now
`type_expr`, instead of `cartesian`.
- I added the patterns for bytes (`PBytes`) and natural literals (`PNat`).
- I renamed the node `Sugar` into
`PListComp` (meaning "pattern of list comprehension").
- Record types in CameLIGO now must have at least one field declaration.
- Replaced the type `closing` and `opening` with one type `compound`,
which captures only the right combinations of opening and closing.
- Components of tuples in a selection must not be written
between parentheses. For example, `a.b.(1).(0)` is now
`a.b.1.0`, as in PascaLIGO.
- LexToken.mli/LexToken.mll
- I renamed the string literal `Str` into `String`.
- I added the tokens `C_None` and `C_Some` (to distinguish the
constructors `None` and `Some`. See AST.ml)
- Fixed the function `mk_sym` so it does not fail with `failwith`, but
with `Error Invalid_symbol`.
- Lexer.mll (shared)
- I removed the character `%` from the identifiers (used to
support Liquidity, like `entry%point` and `match%nat`).
- I adde to the hint on broken strings: "or insert a backslash"
(from a Gitlab issue).
- ParToken.mly
- I added the tokens `C_None` and `C_Some` (to distinguish the
constructors `None` and `Some`. See AST.ml and LexToken.mll)
- Parser.mly
- Fixed the order of declarations in the AST (it was reversed).
- I removed syntax support for Liquidity.
- I added user-defined constructor applications to irrefutable
patterns (the ones afer a `let`), even though only the type
checker can decide that they are truly irrefutable because they
are the only constructors of their types.
- I added natural numbers and bytes to patterns.
- Access of tuple components do not require parentheses now, like
`a.b.1.0`.
- I refactored the semantic actions.
- I added the empty sequence `begin end`.
- ParserLog.ml/ParserLog.mli
- I added a pretty-printer for the AST (with source locations).
- ParserMain.ml
- The CLI for the pretty-printer is now `--verbose=ast`.
- The old CLI `--verbose=ast` is now `--verbose=ast-tokens`.
- ligodity.ml (simplifier)
- I removed the constructions of sets, lists and maps with
`Set [...]`, `List [...]` and `Map [...]`, as there are already
better ways (that is, more like the OCaml's way), like
`Set.literal [...]` and `Map.literal [...]`. (The case for lists
was entirely redundant with the rest of the language as it is.)
- Everywhere there is now a non-empty list of elements, I made a
change. In particular, I removed a corner case ("let without
binding"), thanks to more precise OCaml types for non-empty
lists.
- I ported all the changes to the AST above.
- region.ml (vendors)
- I changed the method `compact` so the end-line is not repeated
if it is the same as the start line: this is even more compact. I
use this in the new pretty-printer for the AST (see above)
- I updated all the CameLIGO contracts.
Right now they're defined as '.bor' and '.band' because of a glitch
in CameLIGO's parser where 'X.Y' leads to a parse error if Y is a
keyword or reserved word in CameLIGO.
There was an error in the syntax of tuple projection (no parentheses
are needed, in fact).
Another issue was a wrong Markdown layout for lists (my fault).
I added a type annotation for the variable iterating a
collection, which is also now marked as "map", "set" or "list".
I fixed and refactored the pretty-printer for the AST.
ParserLog: Finished the AST pretty-printer.
ParserMain: The CLI "ast" is now "ast-tokens" and the new "ast"
calls the AST pretty-printer.
Bug: Added nat literals as patterns.
AST: Removed unary constructor TupleInj.
Parser and simplifier:
- The rule "cartesian" is now properly stratified.
- Parenthesised expressions now correctly create EPar nodes.
- I aligned the names of the tokens in common with Ligodity.
- I removed the "down" and "step" clauses in loops.
- Note: the stratification of the rule "pattern" in the
previous commit has the pleasant effect to remove a call
to "corner_case" in function "simpl_case" of the
file "2-simplify/pascaligo.ml".
- Added more cases to the pretty-printer of the AST.
LexToken, AST: Tiny refactoring.
Bug: Added the making of the AST node PBytes.
Parser: The rule "pattern" was not properly stratified (the
constructor "PCons" was always produced, even when no consing was
done (now a fall-through to "core_pattern").
Bug: When sharing the lexers between Ligodity and Pascaligo, a
regression was introduced with the lexing of symbols. Indeed,
symbols specific to Ligodity (like "<>") and
Pascaligo (like "=/=") were scanned, but the
function "LexToken.mk_sym" for each only accepted their own,
yielding to an assertion to be invalidated. Fix: I created an
error "sym_err" now to gracefully handle that situation and
provide a hint to the programmer (to wit, to check the LIGO
syntax in use).
WIP: Started to write pretty-printing functions for the nodes of
the AST.
CLI: The option "--verbose=ast" now calls that function instead
of printing the tokens from the AST. When the pretty-printer is
finished, the option for printing the tokens will likely
be "--verbose=ast-tokens".
