### Main differences with respect to `ParsleyHaskell`

- Primitive grammar combinators are extensible, including the optimization pass for which they are the current top-level combinator.

- Error messages are based upon the farthest input position reached (not yet implemented in `ParsleyHaskell`) and a there is a preliminary support for error messages based upon [labeled failures](https://dl.acm.org/doi/10.1145/2851613.2851750).

- Minimal input length checks ("horizon" checks) required for a successful parsing are statically computed using a [polyfix](http://okmij.org/ftp/Computation/fixed-point-combinators.html#Poly-variadic) to see beyond calls to subroutines, which is not (yet) possible in `ParsleyHaskell`.

- No dependency upon GHC plugins: `lift-plugin`, `idioms-plugin` or `parsley-garnish` for users. Those provide convenient syntaxic-sugar (by quoting an Haskell expression as itself and its `TemplateHaskell` equivalent) for writing grammar productions, but are experimental, but I do not understand them that much and do not feel confortable to maintain them in case their authors abandon them.

- No dependency upon `dependent-map` by keeping observed sharing in `shareable` and `ref` combinators, instead of passing by a `DMap`. And also when introducing the join-points optimization, where fresh `TemplateHaskell` names are also directly used instead of passing by a `DMap`.

- No support (yet?) for general purpose registers in the `Machine` producing the `TemplateHaskell` splices. Hence `symantic-parser` generates parser much slower than `ParsleyHaskell`, comparable to `attoparsec` in the Brainfuck benchmark.

- License is `AGPL-3.0-or-later` not `BSD-3-Clause`.

- Testing grammars have their generated machines and `TemplateHaskell` splices followed by golden tests.

### Main goals

- For me to better understand [ParsleyHaskell](https://github.com/j-mie6/ParsleyHaskell), and find a manageable balance between simplicity of the codebase and features of the parser. And by doing so, challenging and showcasing symantic techniques.

- To support the parsing of tree-like data structures instead of only string-like data structures. Eg. to validate XML using RelaxNG in [symantic-xml](https://hackage.haskell.org/package/symantic-xml) or to perform routing of HTTP requests in [symantic-http-server](http://hackage.haskell.org/package/symantic-http-server). This is currently done in those packages using `megaparsec`, but `megaparsec` is not conceived for such input, and is less principled when it comes to optimizing, like merging alternatives.

### Implementation techniques

#### Typed Tagless-Final
The syntax of grammars are term-level combinators defined in type-classes,
and their semantics are data-types having instances of those type-classes.
And the same technique is applied for machine instructions and grammar productions.

For automatic deriving, `DefaultSignatures` are supplied, see `Symantic.Typed.Derive`.

For pattern-matching, data-families indexed by the syntaxic type-class are supplied,
see `Symantic.Typed.Data`.