mirror of https://github.com/danneu/html-parser.git synced 2024-11-25 08:19:17 +03:00

a lenient html5 parser written in Elm

Go to file

Dan Neumann e8c1cb6b99 Improve readme		2022-06-03 01:37:34 -05:00
review	Add and run elm-review	2022-05-26 16:54:34 -05:00
src/Html	Avoid text lookAhead with happy path	2022-06-02 23:46:18 -05:00
tests	Prohibit whitespace after open tag bracket	2022-06-02 21:36:38 -05:00
.prettierrc.json	Initial commit	2022-05-16 20:39:48 -05:00
elm.json	Bump version	2022-06-02 23:49:15 -05:00
LICENSE	Add license	2022-05-22 13:06:02 -05:00
package-lock.json	Add and run elm-review	2022-05-26 16:54:34 -05:00
package.json	Add and run elm-review	2022-05-26 16:54:34 -05:00
README.md	Improve readme	2022-06-03 01:37:34 -05:00

README.md

elm-html-parser

A lenient html5 parser implemented with Elm.

A lenient alternative to hecrj/elm-html-parser.

Experimental: Also contains undocumented, unpublished, work-in-progress node tree traversal, query, and transformation in Loc.elm using a Zipper data-structure.

Usage

run to parse an html string into a list of html nodes.
runElement to parse a single html element.
runDocument to parse <!doctype html>[...] into a root node.

import Html.Parser

"<p class=greeting>hello <strong>world</strong></p>"
|> Html.Parser.run Html.Parser.allCharRefs
-- Ok
--     [ Element "p" [ ("class", "greeting") ]
--          [ Text "hello "
--          , Element "strong" [] [ Text "world" ]
--          ]
--     ]

Rendering:

nodeToHtml or nodesToHtml to render parsed nodes into virtual dom nodes that Elm can render.
nodeToString and nodesToString to render parsed nodes into a string.
nodeToPrettyString and nodesToPrettyString to render parsed nodes into indented strings.

Goals

Leniency
- Avoids validating while parsing
- Prefers to immitate browser parsing behavior rather than html5 spec.
- Prefers to use the html5 spec only to handle ambiguous cases rather than to prohibit invalid html5
- Prefers to fall back to text nodes than short-circuit with parse errors
Handle user-written html
- Users don't write character entities like & and <. This parser should strive to handle cases like <p><:</p> -> Element "p" [] [ Text "<:" ].

Features / Quirks

Characters don't need to be escaped into entities.

e.g. <div><:</div> will parse correctly and doesn't need to be rewritten into <div><:</div>.
Tags that should not nest are autoclosed.

e.g. <p>a<p>b -> <p>a</p><p>b</p>.
Closing tags that have no matching open tags are ignored.

e.g. </a><div></div></div></b> -> <div></div>
Ignores comments in whitespace positions:

e.g. <div /> -> <div/>
Parses comments in text node positions:

e.g. div></div> -> Element "div" [ Comment "comment" ]

Differences from existing packages

Currently, there is only one html parser published to Elm packages: hecrj/elm-html-parser.

@hecjr has said that following the html5 spec is a goal of their parser, so their parser is stricter by design and rejects invalid html5.

Development

git clone and npm install.

npm test to run tests
npm docs to preview docs locally

Technical notes

(pre-v2.0.1) Parsing text

Note: This talks about the text parser pre-v2.0.1. Scroll to the next subheader to read about what changed.

One source of parser complexity is text.

Text in lenient html is basically "anything that wasn't parsed by the other parsers."

This means that you can't have a simple parser like:

parser : Parser Node
parser =
    oneOf
        [ element
        , comment
        , text
        ]

Because how would you define the text parser that doesn't underconsume ("parse anything until '<'") nor overconsume?

The best way I can think of accomplishing this with elm/parser is to, inside a loop, try all of your other parsers and then, if they all fail, consume a single character before looping again.

Something like this:

parser : Parser (List Node)
parser =
    loop [] <|
        \acc ->
            oneOf
                [ element |> map (\node -> Loop (node :: acc))
                , comment |> map (\node -> Loop (node :: acc))
                , chompIf (\_ -> True)
                    |> map (Text << String.fromChar)
                    |> map (\node -> Loop (node :: acc))
                , succeed ()
                    |> map (\_ -> (Done (List.reverse acc)))
                ]

It's not nice and simple anymore. And since it's not possible to make an exhaustive text parser, I've had to repeat this kind of logic in various places.

(v2.0.1) Parsing text

The text parser was changed from v2.0.0 to v2.0.1 to be stand-alone meaning that if you apply the text parser, it will return a text node that consumes text up until the next non-text node can be parsed, which is pretty useful and DRYs up some text parsing.

The basic idea is to use lookahead like this:

text : Parser Node
text =
    loop "" <| \acc ->
        oneOf
            [ -- Positive cases for text
              -- We want to be as greedy as possible here to avoid lookahead (slow).
              succeed (\s -> Loop (acc ++ s))
                |= characterReference
            , chompOneOrMore (\c -> c /= '<' && c /= '&')
                |> getChompedString
                |> map (\s -> Loop (acc ++ s))

            -- Now we use lookahead to enumerate negative cases.
            -- Notice how the "match until '<'" case above tries
            -- to avoid ever reaching here.
            , succeed (Done acc)
                |. lookAhead (openTag cfg) -- Starts with "<"
            , succeed (Done acc)
                |. lookAhead anyCloseTag -- Starts with "<"
            , succeed (Done acc)
                |. lookAhead comment -- Starts with "<"

            -- Finally, if we get here, we always consume one character.
            , succeed (\s -> Loop (acc ++ s))
                |= justOneChar
            , succeed (Done acc)
            ]

Between v2.0.1 -> v2.0.2, I realized how important it is to enumerate positive cases so that lookahead is minimized. For example, adding the chompOneOrMore bit above improved my benchmark 700x.

Special thanks

@hecrj and their contributors.
@ymtszw for their work on the Javascript <script> parser.