Next.js Hacker News
CSS as a Query Language
74 points by evnc 18 hours ago | 24 comments
  • k1m
    I find CSS selectors a lot easier to write than XPath. I recently gave a talk on how PHP's new DOM API makes working with HTML and CSS selectors natively very easy (previously you had to convert CSS to XPath).[1]

    It's a shame that because CSS is still primarily for browser use and styling, we don't get nice things like the ability to select based on text content like we can with XPath. My understanding is that this was proposed but didn't make it into the spec because it could lead to performance issues in a browser rendering context.

    [1] https://speakerdeck.com/keyvan/parsing-html-with-php-8-dot-4...

    • lewisjoe
      What's more fun is: LLMs too are really good at CSS selectors. I've been building a document editing agent and I used this fact to present the document as HTML to LLM and make it query and pull pieces of documents into context by just specifying CSS selectors.

      Works like magic!

    • zerocrates
      Yeah, querySelector/querySelectorAll are totally widespread in client-side, it's nice to finally have them in PHP's newer DOM. Definitely what people are used to doing.
      • werdnapk
        document.evaluate is also widespread client-side.

        https://developer.mozilla.org/en-US/docs/Web/API/Document/ev...

        • zerocrates
          Sure, I was more speaking to usage... I'd have to imagine that querySelector* dwarfs evaluate in actual usage. Chrome's stats put evaluate at about 5% of loads; I'm not sure they have tracking for querySelector.

          Of course XPath has its place and is the better tool for some situations. On the PHP side XPath has been supported for a long time, while the querySelector stuff is quite new.

  • abathur
    I feel like we need a name for css-the-syntax (and maybe -the-semantics) as separate from css-the-body-of-rules/functions/units/etc-defined-by-csswg.

    There's juice in it, but it's hard to talk about and survey other uses without just searching GH for code using css parsers and just see what kind of shenanigans people are up to.

    I've been playing around with a weird thing that's kinda like a template engine, but driven by a mix of a lightweight node-based markup language, css selectors for expressing what goes into the template, and a css-alike for controlling exactly how all of these parts come together.

  • duncanfwalker
    Reminds me of seeing this presented at a conference years ago https://github.com/braposo/graphql-css

    It was a joke but I really like the way it pointed out how we copy and reapply patterns in different contexts and that might enable unexpected things.

    • evnc
      oh this is fun

      > we copy and reapply patterns in different contexts and that might enable unexpected things

      yeah, that's exactly what I am trying to do here. Mostly it doesn't go anywhere, but it's interesting for the hacker spirit within me :)

  • spookylukey
    The project pyastgrep https://pyastgrep.readthedocs.io/en/latest/ can use CSS selectors as a query language for Python syntax (default is XPath).

    e.g.:

    pyastgrep --css 'Call > func > Name#main'

    • evnc
      oh this is neat! Feels like exactly the sort of thing I was gesturing towards. Thanks :)
  • efortis
    Not sure I follow the scenario this would solve.

    For instance, currently you can conditionally change a parent based on its children. For example, this `pre` could either have 16px or 0px of padding. Zero when its direct child is a `code` element.

      pre {
    padding: 16px;

    &:has(> code) {
      padding: 0;
    }
  }
    • evnc
      tbh, this started as a connection of two disparate ideas ("hey, this thing looks like this other thing"), and then just kind of explores it in different directions.

      I think the conclusion (which I may not have made clear enough) is less like "These are limitations of modern CSS which ought to be fixed" and more "Maybe a CSS-like syntax could be added to a Datalog-like system and that would be helpful for making it more accessible to more engineers, navigating tree-shaped data, etc"

      thanks for the feedback, anyway!

    • tadfisher
      The article describes a syntax for modifying the underlying data (adding new child elements or attributes to the DOM) for matching selectors, not resolving style changes in a single pass like you've shown.
      • capitainenemo
        I suspect they are replying to this part of the article: "What you actually want to say is: “an element is effectively-dark if it has data-theme=”dark”, or if it has an effectively-dark ancestor with no effectively-light ancestor in between.” That’s a recursive relational definition. CSS cannot express it. CSSLog can:"

        The entire article doesn't seem to mention the existence of :has() which is rather surprising given how recently it was written. Not even in the footnotes.

        • tadfisher
          Yeah, that example is bad. The query doesn't require recursion, but they affirm it does by demonstrating a recursively-defined version of it. This is called "affirming the consequent"; "P -> Q" doesn't mean "Q -> P". Ironic, given the use of propositional logic throughout.
          • evnc
            doh, good point. will fix this, I acknowledge I sort of handwaved the example. thanks for the correction!
  • aisalwaysa
    LLMs suck at CSS now, very tempting to try this and see if its simpler for LLMs to reason about
  • logic-designer
    i cant think of any real use for this -> but cool
  • lostinplace
    Ummm... Isn't that just JQ?
  • shevy-java
    Hmmm. I kind of like CSS but I hate the creep-up of complexity.

    It's not that I don't understand the rationale - any programming language offers more power than a non-programming language. But I'd rather think here that something else could instead replace all of HTML, CSS and JavaScript, rather than constantly wanting to make everything more complex. I don't use most of the new elements in HTML5, largely because I don't see the point in using specialized tags for micro-describing a webpage. I succumbed to the "it is a div-HTML tag and it has a unique ID"; that's what I think mots of those containers actually really are. I even wanted to have aliases to such IDs, simply to use as navigational href intralink.

        [data-theme="dark"] [data-theme="light"] :focus {
      outline-color: black;
    }
    And I also don't like this. It takes my brain too long to process all of that. It is no longer elegant and simple.

    On the other hand:

       h2 {
     color: red;
   }
    That is still simple.

        So ancestor(X, Y) :- parent(X, Y). means: “For all possible values of X and Y, X 
    is an ancestor of Y, if X is a parent of Y.”
    See - I already don't want to have to think in such terms. What is the :- anyway, looks like a smiley.

        @container style(--theme: dark) {
      .card { background: royalblue; color: white; }
    }
    I stopped there.

    I think this is a problem with CSS - too many people are ruining it. It is strange to see how standards that used to work, are degraded over time.

    • evnc
      yeah I mean, to be clear, I'm less proposing "What if we add even more syntax and semantics to CSS" and more "what if we steal ideas from CSS, notice their similarity to logic / relational query languages, and use them to build something new". I probably could have articulated some of this better.

      eg this example:

        [data-theme="dark"] [data-theme="light"] :focus {
      outline-color: black;
  }
      means, in English/pseudocode, roughly: "If you have an element X with attribute data-theme="dark", and X has a child Y with attribute data-theme="light", and Y is focused, then the outline-color of Y is black".

      so we could write this also as, e.g.:

        outline-color(Y, black) if 
    data-theme(X, "dark") and
    parent(X, Y) and
    data-theme(Y, "light") and
    focused(Y)
      that's Datalog, except I went ahead and replaced :- with "if" and "," with "and".

      if we want even more syntax sugar, we could do:

        Y.outline_color := black if
    X.data-theme == dark and
    Y.parent == X and
    Y.data-theme == dark and
    Y.focused
      imagine `X.attr == val` <==> `attr(X, val)` as a kind of UFCS for Datalog to make it palatable to Regular Programmers, right

      the declaration and scope of these variables is implicit here; if you want something even more ALGOL-family, we could write

        forall Y {
    Y.outline_color := black if 
       Y.data_theme == "dark" and
       Y.focused and
       Y.parent.data_theme == "light"
  }
      here we've explicitly introduced Y, and made one of our joins implicit, and it looks even more like Regular Programming now, except the Datalog engine (or equivalent) is kind of running all these loops for you, every time one of their dependencies changes, in an efficient way ...
      • evnc
        one more, for completeness:

          SELECT 'black' AS outline_color
  FROM elements parent
  JOIN elements child ON parent.id = child.parent_id
  WHERE parent.data_theme = 'light'
    AND child.data_theme = 'dark'
    AND child.focused = true
        there's a lot of ways to express the same thing! it's interesting to notice the connections between them, I think, and their strengths and weaknesses, e.g. I probably wouldn't want to write my whole design system in SQL, but since it's relational queries over the elements structure and properties, you could.
  • zephyreon
    This is neat, but no. /s