/reader

What's a song that makes you want to learn an instrument?

I grew up in the midst of the Golden Age of Hiphop, in the 80s. Some might not consider it an instrument, but when I saw Grandmaster Flash in the seminal documentary Big Fun in the Big Town on his turntables... or the other DJ's in the movies Beat Street and Wild Style... I learned you can make music from anything, anywhere. Just use your imagination and go wild.

View Frank Meeuwsen's Crucial Tracks profile

Masked secret police is pure terrorism. We should ask NYPD what we have to do to protect ourselves from them, and then do it. #

I, like a lot of other people, assumed that Americans would be terrible at authoritarianism. Shows we have a lot to learn. Americans are pretty good at it it turns out.#

Krugman doesn’t understand what’s coming for NYC. And doesn’t understand the leadership Cuomo uniquely provided at the height of Covid. It’s pretty likely what’s starting in NYC will be worse than Covid or 9-11. Funny thing is Krugman did get it, a few weeks ago when the crisis in Los Angeles was peaking. He wrote a piece that was terrified, and realistic. The National Guard had been nationalized. Marines were invading California. But now that Calif hasn't been in the news, it's easy to swing back. I bet if the election had happened during the worst of it, Cuomo would've won. #

Bernie Sanders asks what the Democrats should learn from Mr. Z's victory in the NYC primary this week. Here's what I say. Forget about ever rising from the ashes of what remains of the party. Right now, the issue is how to defend the city from the coming war with the US government. They're already holding the current mayor hostage. This will be worse than 9-11 and Covid. We have no leadership. We're totally fucked, what the Democratic Party does or doesn't do, at this point, simply doesn't matter.#

It’s best to skip right to the technical overview on GitHub. There are currently four building blocks, or “elements”, that combined together are an addition to the usual Creative Commons licenses.

At the risk of over-simplifying it, here is my summary of these new elements:

• credit: You want credit back when your content is cited by AI.
• direct contribution: You want some kind of payment when your content is used.
• ecosystem: You want the broader ecosystem around your content to be supported by the AI company.
• open: You want your content to be used only for open weight AI models.

These are layered on as exceptions to a separate proposal at the IETF for completely opting out of certain kinds of AI training. For example, we could declare that we don’t want generative AI training except if “direct contribution” is made back to us.

If we don’t care how our content is used, presumably we simply do not declare any CC signals. My blog is licensed as CC BY, which means I only want attribution when my content is republished in another form. Especially in light of this week’s ruling in the Anthropic case, my blog posts can be used in AI training.

I’ll be watching how all of this shakes out. In Micro.blog, we have a couple plug-ins to block AI crawlers. I also created a plug-in for declaring Creative Commons licenses for a blog. When it appears that CC signals are stable, I’ll add support for signals to the Creative Commons plug-in, for any bloggers who want to be more explicit about how their posts are used for AI training.

That is, until now!

I created a page at /lookup that allows you to specify however many id search params you want and it will pull all the matching icons into a single page.

Here’s an example of macOS icons that follow the squircle shape but break out of it ever-so-slightly (something we’ll lose with macOS Tahoe).

It requires a little know how to construct the URL, something I’ll address later, but it works for my own personal purposes at the moment.

So how did I build it?

Implementation

So the sites are built with a static site generator, but this feature requires an ability to dynamically construct a page based on the icons specified in the URL, e.g.

/lookup?id=foo&id=bar&id=baz

How do I get that to work? I can’t statically pre-generate every possible combination^[1] so what are my options?

1. Create a “shell” page that uses JavaScript to read the search params, query a JSON API, and render whichever icons are specified in the URL.
2. Send an HTML page with all icons over the wire, then use JavaScript to reach into the DOM and remove all icons whose IDs aren’t specified in the page URL.
3. Render the page on the server with just the icons specified in the request URL.

No. 1: this is fine, but I don’t have a JSON API for clients to query and I don’t want to create one. Plus I have to duplicate template logic, etc. I’m already rendering lists of icons in my static site generator, so can’t I just do that? Which leads me to:

No. 2: this works, but I do have 2000+ icons so the resulting HTML page (I tried it) is almost 2MB if I render everything (whereas that same request for ~4 icons but filtered by the server would be like 11kb). There’s gotta be a way to make that smaller, which leads me to:

No. 3: this is great, but it does require I have a “server” to construct pages at request time.

Enter Netlify’s Edge Functions which allow you to easily transform an existing HTML page before it gets to the client.

To get this working in my case, I:

1. Create /lookup/index.html that has all 2000+ icons on it (trivial with my current static site generator).
2. Create a lookup.ts edge function that intercepts the request to /lookup/index.html
3. Read the search params for the request and get all specified icon IDs, e.g. /lookup?id=a&id=b&id=c turns into ['a','b','c']
4. Following Netlify’s example of transforming an HTML response, use HTMLRewriter to parse my HTML with all 2000+ icons in it then remove all icons that aren’t in my list of IDs, e.g. <a id='a'>…</a><a id='z'>…</a> might get pruned down to <a id='a'>…</a>
5. Transform the parsed HTML back into a Response and return it to the client from the function.

It took me a second to get all the Netlify-specific configurations right (put the function in ./netlify/edge-functions not ./netlify/functions, duh) but once I strictly followed all of Netlify’s rules it was working! (You gotta use their CLI tool to get things working on localhost and test it yourself.)

Con-clusions

I don’t particularly love that this ties me to a bespoke feature of Netlify’s platform — even though it works really well!

But that said, if I ever switched hosts this wouldn’t be too difficult to change. If my new host provided control over the server, nothing changes about the URL for this page (/lookup?id=…). And if I had to move it all to the client, I could do that too.

In that sense, I’m tying myself to Netlify from a developer point of view but not from an end-user point of view (everything still works at the URL-level) and I’m good with that trade-off.

1. Just out of curiosity, I asked ChatGPT: if you have approximately 2,000 unique items, how many possible combinations of those IDs can be passed in a URL like /lookup?id=1&id=2? It said the number is 2^2000 which is “astronomically large” and “far more than atoms in the universe”. So statically pre-generating them is out of the question. ⏎

Reply via: Email · Mastodon · Bluesky