Doug Parker 🕸️

https://techhub.social/@develwithoutacause

Overly opinionated software developer on the Angular team at Google. Web development is hard, so I build tooling that makes it easier. I love forcing web tools to do things they were never supposed to do.

Blog: https://blog.dwac.dev/
GitHub: https://github.com/dgp1130/
Bluesky: https://bsky.app/profile/develwithoutacause.bsky.social

Public Key

npub1hvt595dnyyfx57xf5edvhxuppfykj0qmw6myg6hqt87rluxv90eqdwjhta

NIP-05 Address

[email protected]

Profile Code

nprofile1qqstk96z6xejzyn20ry6vkktnwqs5jtf8sdhddjydts9nlpl7rxzhuspz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dskpcapv

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1hvt595dnyyfx57xf5edvhxuppfykj0qmw6myg6hqt87rluxv90eqdwjhta

Show more details

Last Notes

2026-07-19 00:35:22 UTC

- reply

I happened to see a #DesignDoc on this fight recently, but I didn't believe him on how terrible this was.
https://youtu.be/cbL7UuBfULQ?is=JN6GLsCG1-ftlrN5

2026-07-08 17:22:29 UTC

I really hate the pattern of the #browser / #OS removing permissions you haven't used in a while.
I get the idea behind this, but I granted each of those permissions for a reason. Auto-revoking them is just rejecting my original intent, adds to notification spam, and encourages blind approvals.
Is my privacy really improved by having to re-grant the clipboard permission to any app I don't use daily? Is this really where we want users' exceptionally limited privacy attention to be directed?

2026-07-07 21:33:24 UTC

- reply

The harness should separately shut down any unnecessary servers when you're done with the conversation (for some definition of "done").
Picking a random port would at least help prevent a leaked server from interfering with other work on the same host, though I get that makes it less likely to discover the leaked server.

2026-07-06 21:33:47 UTC

- reply

Following that to the logical extreme, it means that tools started from agents shouldn't have a "default port", but just default to "random port".
That would be crazy for a human, but feels like the easiest way to ensure the agent gets out of your way.

2026-07-03 04:43:55 UTC

I'm two palaces into #Persona5 and I'm definitely enjoying it, but I think my biggest issue so far is the focus on the psyche of the villains rather than the protagonists.
Everything in the metaverse just reinforces how shitty the villains are in a very blunt and uninteresting way.

2026-06-14 06:04:57 UTC

- reply

I dumped my extra stats into Arcane, and I'm still not totally clear that's actually having an impact, but going hard on bleed (Lord of Blood's Exultation, White Mask, Seppuku) seemed to help a lot.
With Mohg down I have access to DLC items now at least which will hopefully help with the rest.

2026-06-14 06:04:32 UTC

Waiting for the #Steam summer sale, I needed to fill a couple weeks time and make the mistake of starting yet another #EldenRing run, this time going for a +0 weapon run (basically can't upgrade any weapon).
Definitely harder than RL1. Thus far, Valiant Gargoyles and Mohg gave me the most trouble.

2026-05-16 17:39:36 UTC

The design of Docker Desktop dramatically overestimates the significance of #Docker in my life.

2026-05-16 03:17:11 UTC

- reply

But I think the core problem is that the game relies so much on sequences with no intrinsic feedback. The final puzzle requires *100* correct inputs, and if you misread just one of them, the whole thing fails for seemingly no reason.
My attempt had two trivial mistakes which I needed help to find.
Even worse if you identify the correct answer but *input* it incorrectly. You never know and think your answer is wrong!
Just some feedback about what I input and how close it was to the solution or where I messed up would go a long way (tricky to balance against brute force attacks though).

2026-05-16 03:16:16 UTC

- reply

Just finished #Tunic today and I'm really torn about it. There's a lot to love here and some genuinely incredible puzzle design. As soon as I saw the first page of the instruction booklet, I could see there was something special here and it definitely delivered on that premise.
Minimizing spoilers as much as I can, the door to the mountains puzzle is actually genius and something that's easy to miss but quite obvious in retrospect, like all good puzzles are.

2026-05-05 05:33:15 UTC

RE: https://social.treehouse.systems/@jnkrtech/116519264248279742
Did you know? #JavaScript recently became #Baseline Widely Available.
You can now use it without needing progressive enhancement!

2026-04-14 22:17:45 UTC

- reply

You still need some affordances for occasional manual edits by humans.
But this allows you to design a language much more freely for what works well with LLMs. It makes a different trade off between humans and agents.

2026-04-09 17:02:50 UTC

Trying to run a couple dependent #GitHub #PRs in parallel and oh-my-god I forgot how awful this is. Just having to deal with "branch of a branch" is a nightmare on its own.
I've been experimenting with #Jujitsu which handles this way better, and maybe I need to finally bite that bullet.

2026-04-03 21:22:53 UTC

Exploring an #AI workflow for automatically fixing flaky tests, and having reasonable success with it.
https://github.com/angular/angular/pull/68026
#Bazel is actually a particularly nice fit with features like `--runs_per_test`. I expect any Bazel #web workspace could probably use this prompt mostly as-is.

2026-04-01 21:47:44 UTC

- reply

The amount of "This is obviously broken and is definitely a bug for someone" I encounter on a daily basis is just absurd and enormously exceeds the amount of time and energy I have to try to chase down proper fixes for any of them.

2026-04-01 21:47:41 UTC

#Idea, an #AI that records my screen, looks for any obviously broken UI bug, slices off that little segment of the video, pulls relevant diagnostics, finds the relevant owner of that UI, and reports a bug with all the information.

2026-04-01 04:45:17 UTC

Picked a bad decade for my #GPU to suddenly stop booting.

2026-03-31 16:59:47 UTC

- reply

It sounds really bad, but actually has a happy ending.
A few months earlier I had seen a "Testing on the Toilet" about setting up a prober to test if your production service was down and wanted to incorporate the idea, but we already had basic monitoring.
https://testing.googleblog.com/2007/01/introducing-testing-on-toilet.html
I looked for another opportunity and came away with sourcemaps as a potential use case and created the rare prober which asserts that a production URL fails.
So when we did ship sourcemaps to prod, the prober immediately alerted and paged the oncall who rolled back the release within an hour or so.

2026-03-31 16:59:20 UTC

RE: https://hostux.social/@defcon201/116324484796527679
Quick #story time: I was on a team which accidentally shipped sourcemaps to production!
The production server would serve sourcemaps for local dev, and we built sourcemaps in prod to deobfuscate errors but they weren't provided to the server in the prod environment.
You might see the hazard there. An innocuous change to our #Bazel build structure accidentally pulled in sourcemaps into the server, which it then helpfully served.
This was an app in the #Google monorepo too, so we were revealing all our source code as well as all our dependencies!
#Web #JavaScript #Sourcemaps
#note172q…ew76

2026-03-31 03:34:45 UTC

#TIL #Bash edition: `readonly FOO="..."` is *not* the same as `FOO="..."; readonly FOO` as `readonly` swallows errors.
```
set -euo pipefail # Abort on error
readonly FOO="$(exit 1)" # Does not abort
echo "foo" # Prints
BAR="$(exit 1)" # Aborts
readonly BAR
echo "bar" # Does not print
```

2026-03-28 00:15:35 UTC

Days since I pushed out of date changes to a #Git branch: 0
Seriously, how is `--force-with-lease` not the default (or even configurable)?

2026-03-26 04:14:00 UTC

I'm enjoying learning #Caitlyn in #2XKO, but I swear this game has the shortest input buffer in the #FGC.
It feels like every attack following a special is manually timed. Combining that with all the partial charge attacks and the timing feels absurdly tight.

2026-03-25 23:09:34 UTC

#TIL: In #JavaScript, constructing a date with a string uses a different timezone than with numbers.
```
> new Date('2024-01-01')
Sun Dec 31 2023 16:00:00 GMT-0800 (Pacific Standard Time)
> new Date(2024, 0, 1)
Mon Jan 01 2024 00:00:00 GMT-0800 (Pacific Standard Time)
```
https://github.com/angular/angular/pull/67845/changes/732f85c188cbb38e95d0e5a223bdafaea2a59547

2026-03-25 21:59:49 UTC

- reply

The painful part of flakey tests is the debugging.
If the AI does that debugging and then gives me a working PR with a theory of why it was flakey, I can pretty easily validate that without needing to go through the effort of debugging the test myself.

2026-03-25 21:58:15 UTC

- reply

I'm just proposing the AI create PRs to fix flakey tests, not that it necessarily needs to auto-merge them. We can still enforce as much or as little human review as necessary.

2026-03-25 04:04:31 UTC

Free #startup #idea: A #GitHub bot which daily/weekly runs all the tests in your repo, looks for flakey targets, and then creates #AI generated fixes to improve stability.
#Nrwl #Nx self-healing CI is the closest existing example I can think of, but maybe there's something else out there I'm not aware of.

2026-03-19 18:12:24 UTC

I remain convinced that the #Git CLI is a terrible UX and an especially bad fit for AI agents. Evidence:
```
GIT_SEQUENCE_EDITOR="python3 -c \"import sys; p=sys.argv[1]; c=open(p).read(); open(p,'w').write(c.replace('pick 29507ade4d6', 'edit 29507ade4d6'))\"" \
git rebase -i 29507ade4d6^
```
I only asked it to take some local changes, break them up, and amend them to some existing individual commits and it is thrashing *wildly* but I'm scared to abort or it'll leave me in a potentially worse state.
I'm been playing with #Jiujitsu on the side, I wonder if that might work better?

2026-03-13 21:59:13 UTC

RE: https://techhub.social/@develwithoutacause/116223441415706537
Actually I'm realizing the logical extreme of this is to invent a new programming language for #AI codegen.
Except that language doesn't actually have to *exist*, it just needs to be transpiled as a post-processing step in the workflow to the real language you want.
#note17ns…9789

2026-03-13 19:10:39 UTC

- reply

I wonder if there would be value in a mechanism to generate imports mid-output. Something like:
```
import {foo} from './foo'; // Cool, know I need this.
foo.doSomething(bar()); // Oh crap, I actually need `bar` too.
%genImport bar './bar';%
// ^ Becomes an `import` statement above.
```
The AI isn't exactly generating #JavaScript. Instead it's a little meta-syntax on top of JS to insert instructions for the agent to tweak previous bits of the code as it's generated based on the existing semantics of the language.
I imagine the problem of extra/unused imports is solved well enough by a build loop/linters already such that this is overdesigning a solution. But it's an interesting thought experiment.
Is location-dependent code like actually bad for AIs? Should languages of the future allow importing anywhere?

2026-03-13 19:09:36 UTC

I notice that #AI agents are generally pretty bad about writing imports correctly. They either forget to import something or import too much and leave extras.
I suspect this is from the fact that you need to "predeclare" your imports long before you're actually using them.
Imagine you're writing a file, and you know what you want, but now you have to write out all your imports in one-shot before actually writing the code. You're definitely gonna miss some stuff.
AI seems pretty good at fixing imports it missed, but tends to leave unused imports in my experience.

2026-03-08 02:27:32 UTC

Who names a port flag `--bind`, honestly?!
https://files.techhub.social/media_attachments/files/116/191/188/763/120/273/original/b1184e6cbac0ca48.png

2026-03-07 19:26:07 UTC

This page intentionally left blank to consume an extra 1% of a tree.

2026-03-06 05:07:49 UTC

- reply

Ok, think I figured this out. A few things going on here.
1. I wasn't rebuilding dependencies with #Turbo correctly.
2. Local #PDS instances are limited to `.test` and `.example` by default.
3. Browsers create TIDs from `@atproto/common-web` (not the more widely-documented `@atproto/common`).
I still got some weird "invalid handle character" errors last night and a couple times today, but after resolving the `serviceHandleDomains` that seems to have gone away, so I'm a bit confused what happened with that one.
https://tangled.org/develwithoutacause.dwac.dev/pylon/commit/bbf4b513a9e2774c92764502de821a0e372ea9c5

2026-03-05 06:10:43 UTC

I'm trying to experiment with #atproto, just creating an account on a local test #PDS instance, but I keep getting "Not a supported handle domain" no matter what I provide.
Couldn't find much help online, curious if anyone has any immediate intuition here?
https://tangled.org/develwithoutacause.dwac.dev/pylon/commit/c30786bebd8ce516ba1a1066b8521967b8cd3d38

2026-03-04 02:32:06 UTC

#AI really seems to love grepping source files (especially `node_modules/`) when investigating a problem, but it makes me very worried that whatever conclusion it comes to is based on the implementation details of that library rather than its actual promised contract.

2026-03-02 19:38:27 UTC

- reply

I feel like if the show were to air now, I'd write Gilfoyle's character completely the opposite.
There would be an effort to get everyone at the company to write with AI. Gilfoyle would fiercely oppose this and say how terrible AI is and how he can do it better and faster.
There would be some competition or bet where he argues he can produce more code, at higher quality, and faster and then race him against the AI. He resolves to only use Vim and a Terminal.
Eventually the AI goes off the rails and wrecks everything, proving Gilfoyle at least somewhat right.
It's kind of nuts to see how much we've changed as an industry in such a short time period.
Am I off base here? Would that be an accurate interpretation of the character? I only really saw the last season, so I'd be curious how others more familiar with it saw his character.

2026-03-02 19:37:54 UTC

I was reminded of #SiliconValley (the #HBO show) recently and it struck me how much things have changed and also how prescient it was with respect to #AI.
I only really watched the last season, but the subplot of Gilfoyle's "Son of Anton" AI is kinda crazy in retrospect.
The show ended well before #ChatGPT (2019), so this was very much unrealistic in its time.
Gilfoyle himself is a very "I'm smarter than you and I know it" kind of character, and the way the writers demonstrated that was by having him build his own AI agent to do stuff he couldn't be bothered to do.
This was borderline SciFi at the time and kinda absurd, but it's the way they sold how smart he was. Yet this workflow is increasingly standardized and taken for granted only a few years later. You could write any generic intern at Pied Piper doing the same thing today (albeit less extreme).

2026-03-02 03:56:10 UTC

RE: https://techhub.social/@develwithoutacause/116157563917745882
This is what I get for being skeptical of AI responses and asking it to cite its sources...
#note1erz…uusy

2026-03-02 03:56:05 UTC

I was just asking #Gemini about #BlueSky video hosting limitations and if I could work around them by self-hosting a #PDS, and it pointed me at "Bluesky's video.bsky.app transcoding sidecar" which is apparently... a Rick Roll?
Did I just get Rick Rolled by my #AI?
What's even more baffling is that it's not an arbitrary link. Someone at BlueSky had to make video.bsky.app redirect to that specific video. Someone got paid to do this.
https://files.techhub.social/media_attachments/files/116/157/558/861/687/846/original/c57881f0e819736e.png

2026-03-01 06:47:45 UTC

Whoo! Took down Consort (actually hitless!) and wrapped my RL1 All Remembrances run.
Had a lot of fun with this and I'm glad I did it, but I think now it's time to play something I haven't beaten multiple times already.
#EldenRing #RL1 #Consort #Hitless
https://files.techhub.social/media_attachments/files/116/152/574/924/497/408/original/b70fa211bf2bb70d.mp4

2026-02-28 07:00:50 UTC

- reply

Failed to wrap up today, took the whole night on Leda actually. Midra wasn't too bad, but NPC fights are always a crapshoot.
Cycled between a few builds (including Sword of Damnation) but they were all too random. Eventually I just got lucky with the Star Fists, didn't feel earned at all.

2026-02-27 07:06:06 UTC

- reply

Malenia and most of the DLC are down.
Bayle gave me the most trouble, finally reaching the limits to Ritual Shield.
Only Midra and Consort left. I'm actually not too worried about either, as I'm relatively familiar with both. Hoping to close it out this weekend.

2026-02-26 21:51:47 UTC

- reply

(To be clear, this only somewhat related to the drama going on today, these thoughts have been bouncing around in my head for a while.)

2026-02-26 21:51:36 UTC

- reply

So this is a combination of:
1. Unit tests becoming less important.
2. E2E tests becoming more important.
3. Humans validate E2E output.
4. Design more hard serialization boundaries into complex systems.
5. AI can now more effectively change individual components without tanking the whole system.

2026-02-26 21:51:24 UTC

- reply

I think this also leads into the "fungibility" of this code, as any integration with a hard serialization boundary avoids unexpected coupling.
You can swap out any one component and completely rewrite it, as long as you output the same thing, downstream systems literally can't tell the difference.

2026-02-26 21:51:09 UTC

- reply

Integration or e2e tests however become much easier for humans to reason about and review effectively.
I think this is especially true for anything which can have its output expressed as a "golden", such as a compiler. It's really easy to review that goldens are reasonable as a human.

2026-02-26 21:51:04 UTC

With #AI becoming more prominent, I wonder if this will shift the way we write our tests.
Unit tests will become less useful as the nuances of the implementation become less important. Maintainability is less of a concern and a human can't effectively review unit tests without a lot of context.

2026-02-25 06:07:27 UTC

- reply

Somehow this thread got disconnected, continued over here:
https://techhub.social/@develwithoutacause/116113001638060123
#note1p9e…5fdk

2026-02-25 06:06:31 UTC

- reply

Alright, managed to beat the base game by taking down Maliketh, Gideon, Godfrey, and RadaBeast.
Maliketh was definitely the trickiest, but straightforward once I figured out to stay close and behind him in phase 2.
I thought Elden Beast would give me trouble, but I brought him down second attempt.
Just Malenia left and then onto the DLC.

2026-02-24 02:33:08 UTC

RE: https://techhub.social/@develwithoutacause/116123233842059847
Someone at #Google #Docs:
"Yo, great point! But what if by default, any time you press any key, it quietly inserts random garbage into this public doc you don't have edit access to and sends an email to another human being."
#note14qc…9kmq

2026-02-24 02:25:43 UTC

- reply

Slightly more seriously, I am firmly against any hotkey which is only a single button press by default and I frequently touch a key or accidentally unfocus a text box frequently enough on GitHub that randomly changing the UI is genuine problem for me.
Just do Alt+. or something.

2026-02-24 02:25:30 UTC

Whoever at #GitHub decided to make the period key open #Codespaces clearly does not own both a numpad and a cat.

2026-02-23 07:58:28 UTC

- reply

Oof, stuck on Maliketh all night. It's really a combination of:
1. I never really learned him before.
2. Phase 1 has really fast swipes.
3. Phase 2 is harder to learn *because* it's behind phase 1.
By now I've got phase 1 mostly down, still working out punish windows for phase 2.

2026-02-22 07:03:19 UTC

I'm turning this into an "all remembrances" run and so far have knocked out everyone in the base game besides Maliketh and the endgame bosses + Malenia.
Actually making much quicker progress than I expected. Ritual Shield Talisman has been carrying hard, and I'm worried I'm nearing its limit.

2026-02-21 07:38:18 UTC

- reply

Managed to get to a +10 and kill Morgott, so making good progress. I imagine it'll slow down pretty soon though as I can't just bully future enemies around with an OP weapon.
I'm trying to avoid looking up any guides or strategies, just using my own game knowledge as much as possible (occasionally looking up specific items).
It's a different kind of routing experience and problem-solving than a typical playthrough.

2026-02-20 08:03:32 UTC

- reply

Doing ok so far, took down Godrick and have a seemingly viable build with Golden Halberd.

2026-02-20 08:03:22 UTC

With so many games to play it can be tough to decide where to start. Which is why I'm starting *another* #EldenRing playthrough, this time RL1 (rune level 1).
God damn, why can't I put this game down?

2026-02-19 02:01:43 UTC

As a
- Terminal user
I want to
- Press delete to generate "^[[3~"
So I can
- Save time whenever I need to type "^[[3~"

2026-02-18 18:24:36 UTC

I'm glad to see #AI #skills coming up as a more distributed, reusable form of context. It's almost like splitting up an application `AGENTS.md` file into individual libraries.
My biggest problem is that versioning is still a fairly unsolved problem.
When developing skills to define usage of a tool, I want those skills to be versioned with the tool. If a new flag is introduced or an old flag is removed, the skill should be up to date.
I want to ship skills not as GitHub repos, but deployed with the tool directly and automatically consumed.
I feel the ecosystem is moving the right direction here, but still not quite there yet.
Auto-install skills from `node_modules/@scope/pkg/.skills` you cowards!

2026-02-16 07:55:16 UTC

- reply

UPDATE: Player having trust issues after eliminating second health bar only for game to reveal a third.

2026-02-16 07:54:56 UTC

BREAKING: Player feels lied to upon "defeating" difficult boss only for game to reveal second health bar.

2026-02-14 19:13:02 UTC

- reply

Also, people tend to forget about #MGS2, but I feel like that game's message would resonate a lot better with modern audiences.
The game called it AI at the time, but I think you could find / replace in "social media" and "algorithms" and find the game's themes to be quite prescient.

2026-02-14 19:02:04 UTC

- reply

I feel like it's one of the few things you could do with the franchise without #Kojima, since you're just copying his work.
Only real requirement to me is that you need #DavidHayter, but I think #MGSDelta made that happen, so maybe it's actually possible.

2026-02-14 19:01:58 UTC

You know what games would benefit from a #ResidentEvil 2 Remake situation?
#MetalGearSolid 1 & 2
Just remake those stories with modern (MGS4/5) gameplay mechanics, rework the level design a little bit, instant hit.
#Konami, make this happen!

2026-02-10 16:50:14 UTC

Almost never in the history of #capitalism have productivity gains in an industry led to increased compensation for the working class. It always goes to corporate profits or cutting costs by reducing staff.
I don't expect #AI to be any different.

2026-02-08 08:10:29 UTC

Don't fall into the trap of "I can't figure out if this is a good idea, so we shouldn't do it."
Instead, think along the lines of, "I can't articulate why this is a *bad* idea, so we should try it and see what we learn."

2026-02-08 03:38:42 UTC

- reply

Demo
https://files.techhub.social/media_attachments/files/116/032/923/606/630/658/original/fb089757fca7b73e.mp4

2026-02-08 03:30:31 UTC

- reply

This was actually my first time vibe coding outside of a work context. It definitely helped as I'm not very familiar with #VSCode APIs, though there were definitely times it was faster to just get my own hands dirty.
#AI #LLM #VibeCoding

2026-02-08 03:30:14 UTC

Put together a quick #PR to add an "Open file on npmx.dev" command to the VSCode extension.
https://github.com/npmx-dev/vscode-npmx/pull/34
I've always wanted to deep-link to files in #npm, and now that I can with #npmx, I want to have an easy way to do it!

2026-02-05 17:36:13 UTC

A wise man once said "work smarter not harder" to which I say "skill issue."

2026-01-30 23:24:22 UTC

#AI can be kind of incredible, but I think my biggest disappointment is that it doesn't really seem to solve the problems I want it to solve.
If I have a weird test failing for complicated reasons and ask the AI to fix it, it always fails. If I can't confidently solve the problem, neither can it.
To some extent, I understand why it can't solve these complex issues (and I guess it's good for my job that it can't). But that means it's doing all the fun stuff and I'm left with all the hard problems I didn't want to deal with in the first place.
It's the developer version of:
https://files.techhub.social/media_attachments/files/115/986/624/416/099/420/original/b4fa07763a5ac1ea.png

2026-01-30 02:24:32 UTC

Couple nasty mixups I've been playing with for #Vi and #Caitlyn. #2XKO
Having a lot of fun with Caitlyn's combos in particular. Juggling the opponent with her rifle is so satisfying (if a bit janky from the physics).
https://files.techhub.social/media_attachments/files/115/981/669/747/487/942/original/09581deeac525634.mp4

2026-01-28 21:37:46 UTC

An underappreciated aspect of #software #design is symmetry.
If an abstraction acquires some kind of resource, it should also be responsible for releasing that resource.
Separating the responsibilities into different abstractions just makes it harder to understand the scope of each abstraction.
This goes beyond just managing resources to code as simple as:
```
if (node.isConnected) {
doSomethingWithAttachedNode(node);
} else {
doTheSameThingWithADetachedNode(node);
}
```
The purpose and goal needs to be aligned, even if you need to adjust the implementation for different use cases.
The code smells here are effectively:
1. `if` with no `else`.
2. `if` / `switch` branches with different effects.
3. Non-exhaustive `switch` statements.
4. A create/acquire operation without a comparable destroy/drop operation in the same class/file.

2026-01-21 19:42:53 UTC

Interested in #Microfrontends? I'll be chatting with @manfredsteyer.bsky.social tomorrow about how we're tackling them at #Google and #Angular. We hope you'll stop by!
https://www.meetup.com/angular-meetup-graz/events/312845056/

2026-01-17 06:27:22 UTC

Grinded out some #Artorias in #Nightreign. Having not played DS1 to completion, it's a first time for me. So he just has a move that ignores iframes? Unlike *every* other enemy in the game?
Playing with newly buffed Executor, ironically that move is his easiest because deflect hard-counters it.
You can fit in up to three guard counters and basically stagger him from scratch if he does all three.
Only weird bit is that he has two moves which seem to dead-angle and can be undeflectable in the wrong position, which is pretty nasty for this fight and hard to deal with.

2026-01-17 00:14:28 UTC

RE: https://techhub.social/@develwithoutacause/115906340207297016
I think this is the "Tiled windows have margins" setting and can be disabled.
Not sure I see the aesthetic appeal of margins like this, but I guess it made sense to some designer.
#note1pm8…v6sl

2026-01-17 00:10:36 UTC

- reply

If there's a setting to turn it off, I'd be very interested to learn about it. I'm not even sure what to search for myself.

2026-01-16 19:06:36 UTC

- reply

I've separately noticed that usually (not always?) when you maximize it leaves a little gap around the window. It annoys me visually, but whatever, not that important.
But now I'm wondering if that's working around this issue because if you click at the edge of the screen, you miss the window.
It's still not a fix, because I fail to scroll in that scenario, albeit in a slightly less painful manner. I've made this mistake at least 5 times in barely 15 minutes of coding. I haven't even written anything yet, I'm just scanning source code to figure out what I need to write.

2026-01-16 19:05:47 UTC

My latest #MacOS issue is that every time I try to grab the scrollbar in VSCode to scroll down, I end up somehow grabbing the whole window and dragging it.
I get that there's a space there to shrink the window width, but the designer clearly coded it so a drag down gets turned into a window drag.
But I don't know why I would *ever* want that? If I'm grabbing on the right side and pulling down, I think it's pretty clear I'm trying to scroll, not drag the window.
My intuition is that this is generic "window" behavior at the OS level unaware that a scrollbar exists at all.
https://files.techhub.social/media_attachments/files/115/906/331/098/316/650/original/5c6ba75ec58efd3d.mp4

2026-01-14 16:34:26 UTC

On my commute this morning, I saw a truck advertising gluten-free office equipment and umm...
I have questions.

2026-01-12 22:06:09 UTC

- reply

Alt+Tab (or Super+Tab to me now, I dunno what it actually is) selects apps that are open but have no windows?
Finder is always in the list because you can't quit it. This is just noise. Why do I care about apps I've closed? This just breaks my mental model of where Alt+Tab will take me.

2026-01-12 22:05:59 UTC

- reply

Scrolling is reversed by default which is just weird on mouse, but whatever.
What bothers me is that the setting to reverse applies to both touchpad *and* mouse, even though reverse scrolling on touchpad is quite reasonable IMHO. Is it that hard to have two settings?

2026-01-12 22:05:49 UTC

- reply

Home and End seem to work like "PgUp all the way" and "PgDown all the way".
It seems Cmd+Left/Right is what I would interpret as Home/End.
What I'm learning about myself is that I actually do use Home/End a lot (often with Shift) to edit lines, but that muscle memory is out the window now.
I'm sure these semantics are useful for somebody, but I'm confused why we needed to have two different semantics for Home/End? Could we not have just made up different keys for the two operations and let keyboard makers put whichever they want for their layout? That would at least work consistently.

2026-01-12 22:05:33 UTC

- reply

The keys are also out of order compared to the Mac embedded keyboard, making it impossible to know what I'm actually pressing.
If you tell me "Hit Cmd+J", I'm just going to try all three until one of them works.
Terminal still uses Ctrl, not Cmd. So my brain has to switch in and out of "Mac-mode" to make that work.
VSCode is especially confusing, because I need Cmd+P instead of Ctrl+P to open files, but then I need Ctrl+` to open the embedded Terminal? And then everything in there uses Ctrl?
This is maybe less of an observation about Mac so much as that it was maybe a historical mistake to print OS-specific keys on keyboards. I don't know the timeline here, but I feel like we screwed ourselves as soon as we started doing that.

2026-01-12 22:04:41 UTC

- reply

Being a lifelong #Windows / #Linux user, I'm trying to keep an open mind with the new #Mac, but this is really hurting my brain.
You use Cmd / Option instead of Ctrl / Alt, which is a well-known pain point and I'm trying to work with it, but it's so much worse than just learning new hotkeys.
Most Ctrl operations on Windows are Cmd operations on Mac, but that's not universal. Ctrl+Arrows on Windows moves words, but it's Option+Arrows on Mac, not Cmd+Arrows.
You can't just think of this as "Mentally swap Ctrl and Cmd".
I'm using an external keyboard with Ctrl, Win (Super), and Alt keys. But none of those exist on Mac. Ctrl means a completely different thing, Super seems to be Cmd, and Alt is Option?
Cmd+C / Cmd+V for copy paste feels reasonable, but that's not what I'm typing. I'm effectively typing Super+C / Super+V and just confusing my muscle memory even more when I need to use the embedded keyboard rather than my external one.
Maybe I need to internalize Super = Cmd?

2026-01-11 17:33:17 UTC

- reply

Can you expand on that? What specifically doesn't work with CSR but does work with SSR?

2026-01-11 04:34:21 UTC

- reply

I'm personally quite skeptical SSR even meaningfully impacts SEO at all. My understanding is that Google indexes CSR content just fine and SSR just helps a bit with indexing time.
I'm less certain of other social media sites or AI models, I'd be curious to hear others' experiences.
I know its hard to look past capitalistic profit motivations, but can we at least like, add performance to the list?
"So I need to improve my site's SEO *and perf*..."
It's only 8 characters...

2026-01-11 04:34:09 UTC

Rant: Whenever someone discusses #SSR, I swear the post *inevitably* starts with...
"So I need to improve my site's #SEO..."
Is that really where our priorities are as an industry?
I totally get why that's the case, but I need to say sincerely as someone who has supported and advocated for prioritizing #Angular SSR: I didn't do it for SEO.
I did it for user experience and #performance. Those are the metrics I want to see improved when you use it, not SEO.

2026-01-10 03:27:40 UTC

Setting up a new #Macbook and just going great!
1. Had to configure the Wi-Fi, timezone, etc. 3+ times before hitting the desktop.
2. Terminal settings hung multiple times, preventing force quit and full computer restart.
3. Switching to zsh completely borked my terminal because I was missing tmux.

2026-01-01 06:53:11 UTC

I'm glad the #EldenRing #Nightreign DLC added more content, but it seems to really highly prioritize that content over base game to the extent that it feels like there's *less* overall content.
I just did 5 runs in a row and only fought Death Knights, Red Bear, and Dancing Lion as my night bosses.

2025-12-21 05:44:26 UTC

- reply

I don't think I have any interest in pursuing the pantheons further. I only reached the fourth pantheon and actually downloaded a save file just to unlock the final bosses. I highly recommend that strategy for anyone aiming for an "all bosses" run like me.

2025-12-21 05:44:15 UTC

And finally took down #AbsoluteRadiance in #HollowKnight!
It's by far the hardest boss in the game and I don't think beating it has drastically changed my original opinion. I feel like I got fairly lucky in phase 4 and entered the end with basically full resources to burst it down belligerently.
It definitely wasn't entirely luck, I had a great phase 1 & 2 and 3 / 4 / 5 weren't *entirely* devoid of skill.
But I don't feel the same satisfaction as other bosses because the overlapping attacks lend themselves so much to randomness. I feel like I just kept trying until I happened to win. 🫤

2025-12-21 04:07:39 UTC

- reply

My only nitpick is that the double horizontal strike can start up while you just happen to be standing near the edge of the arena and hit you before the attack really starts. It was infrequent enough to not to be too much of an issue but was probably the only time my deaths really felt unjustified.

2025-12-21 04:07:22 UTC

- reply

In fact, I feel like this is actually a better version of #AbsoluteRadiance, because it significantly speeds up the fight in the same but doesn't quite overlap attacks into undodgeable scenarios. Feels much more fair in comparison.
https://techhub.social/@develwithoutacause/115750483161069468
#note1yc4…wda6

2025-12-21 04:06:43 UTC

Whoo! Just defeated Radiant #SistersOfBattle (meaning hitless).
This is probably my favorite boss fight in #HollowKnight. Took a lot of tries but I never really got frustrated because it's so well designed.

2025-12-20 07:31:50 UTC

- reply

Also giving it a unique phase not from the original fight right at the very end is just plain troll.
I think I might actually write a blog post expanding on that point.

2025-12-20 06:30:33 UTC

- reply

It's basically the same but faster, which would be fine in isolation. Except that each attack gives you only barely enough time to dodge, such that simply being in the wrong place sets you up for failure.
You can successfully dodge one attack only to be left in an undodgeable situation for the next attack you couldn't have predicted.
Some situations you can anticipate like the spikes, but an unlucky horizontal or vertical nail with the spikes in a bad spot is just an F-U to the player.
It makes the fight far more luck dependent than it had any need to be, and eliminates a lot of the clarity and simplicity the regular boss has.

2025-12-20 06:30:06 UTC

I really like #Radiance as a boss fight in #HollowKnight, it's really clean and reads so well with obvious answers for each attack which are just complicated enough to be hard to execute.
But #AbsoluteRadiance is too far I think.

2025-12-18 20:40:35 UTC

#TIL #Chrome kills browser extension service workers after 30s of inactivity and then silently drops any messages sent to them from another execution context?
https://developer.chrome.com/docs/extensions/develop/concepts/service-workers/lifecycle#idle-shutdown
https://github.com/angular/angular/pull/66166
I'm fine with killing unused service workers, but then I would expect them to be restarted in order to receive an event specifically sent to them. It feels broken that they just drop these events entirely? Is this really the intended behavior?

2025-12-18 19:40:29 UTC

#TIL `pnpm run my-script` is *not* the same as `pnpm my-script`.
The former only executes commands in the `package.json` `scripts` object. The latter will execute scripts or binaries installed from dependencies.
You need `scripts: {'my-script': 'my-script'}` to get them to behave the same.