Hacker Newsnew | past | comments | ask | show | jobs | submit | pastelsky's commentslogin

Author here - thanks for all the support (and bug reports)!

I built pretty.fish with the belief that Mermaid diagrams don’t have to look like they’re from the 80s and to avoid squinting at Mermaid diagrams that coding agents create in tiny windows.

Having them on canvas also makes it easier to keep a track of, and even visualize agent steps as and when they are are being taken (if you connect an MCP you get a pretty nice view of agent building its knowledge graph for eg for a problem).

Didn’t expect it to take off like this—there are still rough edges (pan/zoom, mobile), but I’m working on smoothing those out soon.


Thanks for sharing! I haven’t looked into that before but looks neat.


- Write Mermaid diagrams with a live preview. - Arrange multiple diagrams on an infinite canvas. - Group diagrams into multi-page projects. - Better themes


Make no mistakes.


UPI is great in terms of UX, but I don’t think it’s super reliable – especially big public banks go down all the time. The UPI base service had multiple incidents in the last 1 month too.


Did not expect to see Emraan Hashmi in this post!


Indeed! Bollywood makes it to HN xD


One would’ve thought that after looking at modern day computers that being able to run a wire through two computers being able to transfer data quickly, reliably, easily would be a 101 feature these boxes offer.

It’s 2024 and it doesn’t really work!


It used to work for some applications.

Back in the day, one could connect a bog-standard FireWire (IEEE 1394) cable betwixt a pair of Windows boxes, and it would create a (quite fast for the time) network for them to talk on.

Addressing was automatic with RFC 3927, and names just worked too.

After that, any appropriately-shared things on one machine would be accessible on the other.


This exact functionality works out of the box with macOS and a thunderbolt cable (I believe going back to TB1, but I've only tested with TB3).

Does Windows not auto-provision thunderbolt networking automatically?


OTOH corporate security gets paged whenever you plug something that isn’t a power supply or a monitor into the corporate box, up to and including an Ethernet cable.


I've seen banks disable USB ports by means of hot glue.

Hardware attacks are super hard to defend against. Depending on your threat model, drastic measures might be warranted. Anything with a USB plug can be a keyboard with an attack payload (e.g. Rubber Ducky). And if you think you can whitelist devices based on USB class or some identifier, you're wrong as they can be spoofed. Heck, there are "USB C cables" that are really attack payload delivery systems (e.g. O.MG Cable).

It's a scary world out there. Stay safe.


Banks could ask known-good keyboard vendors to implement PKI auth. Intel has a spec for PCIe device authentication, based on USB-C authentication, https://www.intel.com/content/dam/www/public/us/en/documents...



Fantastic, thanks for the pointer!


I hadn't heard of either PCIe auth or USB-C auth. Thanks for sending me down that rabbit hole!


USB4 networking is supported by Windows and Linux, and afaik macOS too in some form.

> It’s 2024 and it doesn’t really work!

have you actually tried it?


Author here:

I created this because I found myself peeping inside type declaration files too often, and the only way to do that was by installing the package first.

tsdocs.dev helps you check the API surface of a good number of JS libraries and their past versions — usually a quick search away.

There's something powerful about speed and being able to answer questions in seconds that usually take minutes.

edit: The server might be overloaded with requests as we prime up our caches, but do visit back after HN's done hugging us to death.

You can show your support and help cover a part of server costs if this (or bundlephobia.com) saved you time.

https://github.com/sponsors/pastelsky


This is really awesome work!

This takes a big burden off of individual packages from publishing their own API docs (having done that I know how hard it can be!), and having a centralized API viewer can offer a lot of advantages over separate docs.

A couple of things I would suggest:

1. Track re-exports and cross-package references and allow crossref links to go into other packages. If package A uses rxjs, then links to the rxjs types should go to the canonical definitions in the rxjs package. (figuring out the canonical declaration can be tricky because it's not always the original declaration though)

2. Organizing by type isn't always a great introduction or way to navigate a package. On the Lit project at https://lit.dev/docs/api we re-organized the API docs by categories and the most important API surfaces. There aren't standard jsdocs for this, but a few straightforward things like @category could be used to offer an alternate top-level nav for a package. Also consider supporting the @packageDocumentation tag from tsdoc.

3. Consider showing files other than the README. Relative links to things like CONTRIBUTING.md currently break. Alternatively interpret all relative links as pointing to github.com or npmjs.com. It'd be great to have a way to link from the README into API docs to guide readers. The community doesn't have a great convention for this unfortunately.


> 1. Track re-exports and cross-package references This should already be tracked — play around with the member visibility widget on the sidebar for e.g. For convenience, I've inlined the types from re-exports, but might be a good idea to indicate they were re-exported.

Pointing to the canonical source can also limit the usefulness in a few cases (e.g. d3 is just a bunch of re-exports), and users may not care about internal package organization.

> 2. consider supporting the @packageDocumentation tag Organizing by @category and @packageDocumentation should already be supported. e.g. See the functions in — https://tsdocs.dev/docs/lodash-es/4.17.21/index.html

> 3. Relative links to things like CONTRIBUTING.md currently break Yeah, this is known, thanks! https://github.com/pastelsky/tsdocs/issues/9


Found a fun error "TypeDefinitionResolveError" for a package that includes Typescript sources. Best guess, it may be related to the package.json uses a modern "bare" exports field ("exports": "./index.js"; no "main", no "types") where "index.ts" exists in the package as well. (Not a lot of projects use this today, but it is a modern way to publish projects that will likely increase.)

Typescript itself works just fine with this with this type of package with "moduleResolution": "node" (tsconfig/CLI options) with recent version using recent enough ES targets or with "moduleResolution": "node16" in older versions/older ES targets/non-ES targets.

https://tsdocs.dev/search/docs/butterfloat

Hope that's useful enough to debug the error.


Hey, Leo from Deno here.

We recently released a static HTML documentation generator via our `deno doc` subcommand.

We also have a similar system for our registry (here an example: https://deno.land/std@0.209.0/http/mod.ts?s=Server).

But we have an outdated system for general use, which is similar to what you have built: https://doc.deno.land/.

Did you know about this, and did you want to built your own system regardless, or did you not know about it and would this have been something you would have used instead of creating your own system?


Cool, fwiw I didn't know about the general use and I use deno heavily, here's a comparison of Three.js:

Deno Doc: https://doc.deno.land/https://esm.sh/v135/@types/three@0.159...

TSDocs.dev: https://tsdocs.dev/docs/three/0.159.0/modules.html

The search UX and documentation UI of TSDocs is better. The availability of Deno is better atm.

Though TSDocs was speedier and a better flow when it was working.


yes, the doc.deno.land is relatively outdated and uses an old system. our static generator via the doc subcommand and registry are the newer systems


Does this service do anything different/better than jsdocs.io?

I like the way jsdocs.io puts everything on one page.

I was initially confused when I looked up a package on tsdocs.dev and just got the readme. The actual type defs were tucked away out of sight in the hamburger menu. May just be a problem with the mobile site.


This is useful, the search especially.

There's a bug that only happens on this page [1]?

    Uncaught TypeError: Failed to execute 'observe' on 'MutationObserver': parameter 1 is not of type 'Node'
[1] https://tsdocs.dev/docs/svelte/4.2.8/functions/_svelte_compi...


Hi, nice site! I found a small ux issue on ios: the search field has capitalization on by default. It can be turned off with https://stackoverflow.com/a/5171812


I attempted to build this a number of years ago but quickly got frustrated with the inconsistency of how packages are laid out and the unreliability of the information in package.json. Good job if you sorted it out though.


Congrats, this looks fantastic and will be useful for many!


There isn't a LICENSE file in the repo.

Can you add one?


tsdocs.dev helps browse reference typescript documentation for any package or version of a library.

So you don't have to rely on library authors to provide upto date reference docs for different versions of their library


So I gave this a go for my canvas library[1]. The results are better than I expected, especially as (I assume) everything beyond the readme is being generated from my library's (handwritten) .d.ts file!

One question I do have: are there annotations and/or additional details that I could be adding to the file, or library, which could make the output more useful for anyone wanting to use this as a reference to the library? Also: how permanent is the output - is it going to regenerate every time someone hits that endpoint or is it robust enough for me to add a link to it from the library's website?

[1] - https://tsdocs.dev/docs/scrawl-canvas/8.10.4


If I understand correctly, the limited amount of "expressiveness" a quantum computer has restricts its ability to solve useful problems.

Couldn't you simply express useful problems as a function of problems that the quantum computer can already solve? Doing so might be extremely in-efficient, but give that there's so much performance leeway, it still might end up being similar to super computers of today?


On a related note – ‘networkIdleCallback’ : Preload assets and data whenever network is idle —

https://github.com/pastelsky/network-idle-callback

Helpful for progressively hydrating features in a page, or loading assets for next interaction


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: