danielxmoore / hera Goto Github PK

( Foo:foo )? never binds foo. Similar to pattern matching in Civet.

The example code in README.md does not work:

Hera = require("hera")
fs = require("fs")

source = fs.readFileSync("cool-grammar.hera")

rules = Hera.parse(source, {
  filename: path
})

// Generate parser.js source code
parserSource = Hera.generate(rules)
fs.writeFile("cool-parser.js", parserSource)
parser = require("./cool-parser")
// or generate parser object
parser = Hera.generate(rules, true)

parser.parse("text that my cool parser should parse")

The first problem is the first require. It should be Hera = require("@danielx/hera").
The second problem is that the hera API are outdated. The latest API are tested in https://github.com/DanielXMoore/Hera/blob/main/test/main.coffee.

Here's a small idea for Hera optimization: skip matching the empty string literal "". We use this ~70 times in the Civet grammar, mostly for declarative rules, which I imagine get hit a lot (you probably know better).

Given that it always matches, we shouldn't need to build/call $L at all. I think this optimization could be in compileOp but I haven't understood this code yet to know how to try it.

I recently wanted to do something like:

HashThis
  LengthShorthand:id &( ( _+ ( ( Not __ In ) / In ) ) )?:beforeIn ->
    if (beforeIn) ...

but ended up needing to do:

LengthShorthand:id ( &( _+ ( ( Not __ In ) / In ) ) "" )?:beforeIn ->
  if (beforeIn != null) ...

This is because assertions always return undefined since they fail the parse if not met. This causes issue with ? because it returns undefined if the parse wasn't matched so there's not a simple way find out whether the optional assertion matched or not.

It would be nice if optional assertions translated to true/false for whether the assertion passes or fails.

Rule
  A &B?:b ->
    // b is `true` or `false`

• -Foo omits Foo from $0, and doesn't assign it $i index.
• &/! do this automatically (this would be a breaking change, potentially requiring all grammars to be rewritten, as it shifts numeric indexing, unless we still assign them $i indices)
• Alternative: ^Foo selects Foo for $0 (omitting others not marked ^)