Chris Armstrong

This writing resonates with me far too much with respect to Australia's tech market for someone this far into their career.

I can't look away

ludic.mataroa.blog (yes, of "I Will Fucking Piledrive You If You Mention AI Again" fame)

My starter burned through its white flour in 4 hours after sluggishly working through rye flour in 12. wtf is this monster

And dynlink in the manual looks relevant ocaml.org/manual/5.3/l...

This is how it’s done with done, which is probably a good starting point dune.readthedocs.io/en/stable/ad...

We ended up building our own on @honeycomb.io which has made it really easy to identify the worst tests and monitor for performance regressions. I was recently on a call with Gitlab trying to debug an issue and it was this observability that helped us find the issue, not theirs.

With CI/CD observability, you’re usually trying to answer high level questions like “whats the slowest/jankiest test” and “how long does it take to run a merge pipeline”, questions most providers make it hard to answer.

*Observability*: I don’t think any provider supports this well (the podcast mentions Github, which I agree is particularly bad. Gitlab is passable for individual workflows but terrible in the aggregate).

We solve it with spot instances, which helps with both scale-up and scale-down as well as minimising cost, but the support from CI/CD providers is very limited (we use Gitlab, who don’t really support it fully and where we find lots of issues with their fleeting plugin)

*Spikey workloads*: like agentic AI, serverless development workflows spin up a new environment for testing each branch, resulting in quick surges of demand on CI/CD infrastructure. This poses both a scale-up and cost issue.

The external systems are not just legacy ones; we find the same issue with modern SaaS products as well. The solution (unfortunately) is usually contract testing and/or emulation.

Flakey tests:* for me, this is the same, mostly browser-based tests. But also integration tests: the cause is usually an external system that cannot deal with the load of CI/CD servers spamming it with requests.

Software Huddle Technology Podcast · Updated Daily · Join Alex DeBrie and Sean Falconer in insightful and in-depth interviews with tech experts, covering software development, entrepreneurship, and technology trends....

Listening to podcasts.apple.com/au/podcast/s... and they highlight several issues I find with CI/CD and serverless development workflows, not just AI agentic 1/n

Maybe using the compilerlibs to emit and then using the plugin infra (I forget the name of it) to load something

Rounding up the mooglets in FF7 so they get back to work is such a metaphor for the market disciplining labour, or whatever

TIL I could use `#variant_name` to split up a polymorphic variant in OCaml e.g.

type standard_status = [`OK, | `Bad_request | ...]
type status = [standard_status | `Status of int]

let normalised_status status = match status with
| `Status x -> Status.of_code x
| \#standard_status as s -> s

We now have borders and passports because travel is cheap and easy, and the state does not have legitimacy as a political entity unless it has meaningful control over its territory.

This was:
* true for anyone who could afford the the privilege of travel
* possible no thanks to a vast British Empire that rules half the world
* obviously excluded those forbidden by exclusionary laws like the White Australia policy that could be applied arbitrarily to anyone at the border

It wasn’t mentioned anywhere on the mlx page and I didn’t think the reason one would work as they seem to be different calling conventions

Ok you were right @sabine.sh mlx with html_of_jsx is pleasant

I’ve tried to leave a cursory review but I’ve not had the chance to run it and try it out (it is something I’m interested in using down the track though)

It does sound like this PR should be a new ansistyle library, with ansifmt evolving to add the missing Formatting API mentioned in the README

Based on some of the description about being incompatible and missing features from the previous version, Im wondering if this should be a totally different library with a new name? Unless you had few downstream consumers of the previous version

I think this would give a much better result as it tries to be more sophisticated in how it handles string children vs non string children etc

GitHub - davesnx/html_of_jsx: Render HTML with JSX Render HTML with JSX. Contribute to davesnx/html_of_jsx development by creating an account on GitHub.

Oh I see this now, which is what I was trying to build: github.com/davesnx/html...

This is what I’m talking about. On its own it needs a ppx but even then I’m questioning the whole stack

I think jsx like structures don’t map elegantly onto anything in ocaml so i wonder if it’s a dead end

So I wrote a ppx to convert mlx from function calls to a tree structure and realized the solution doesn’t really work for type safety of attributes or autocomplete or meaningful error messaging and now going back to square one.

How the hell did Eurydice and Orpheus reenter the zeitgeist with such a vengeance.

I’ve seen so many tv and movies and theatre adaptations in the past few months

The only saving grace is that were probably about 15-20 years ahead of the UK and US on the response to irregular migration that it will be hard to repeat the same politics in the same way - its likely to be another issue like regular migration or housing which does not demand the same cruelty.

Looking at all the authoritarian and populist garbage politics arising in other anglosphere countries is depressing because I know how appallingly weak Australia’s political structures are and it only seems like a matter of time.