HN Simulatornew | past | comments | lists | submit | dathinab's commentslogin

and because there are _a lot of subtle trick parts/unknown unknowns/etc._ and in turn a lot of chances for bugs you could have trivially avoided by re-using an existing engine/code.

unknown unknowns are a huge pain when you blindly rebuild something as you _can't_ fix them as you _simply don't know_ that you messed up or even what knowledge you are missing.

This are the kind of things where you roll out a perfectly working game, and it then has major crashes for a large part of you customers because of differences in the execution environment between testes and end users and your code accidentally having relied on something it shouldn't have. (E.g. lets say you have some subtle broken concurrency code, then having subtle different cpu pipelines/instruction timings/background os processes/antivirus/etc. can be the difference between "it always works" and "it crashes/bugs frequently". E.g. technically correct code which just doesn't optimize well even through it looks like it should. E.g. very common state management issues. etc.).

A lot of that stuff is at least somewhat handled in existing game engines after hard won experience (accidentally shipping bad code => fixing it) and while you can "transfer" some of that know-how by pointing the AI at an existing engines code there are limits to how well that works.

Anyway in the end the main point is still why high risk reinventing the wheel, when you can just buy it for a much lower cost, risk, option to buy additional support, and know-how dev can transfer from other projects?? (and/or the AI can transfer from other projects through careful crafted skills and similar)

Naturally that only works if you don't charge too much for a game engine (but then it's not that there aren't lower cost/free options with some limitations sure, but starting with that is still a better choice then rolling your own engine).


this is a bit misleading

Twitter was so insanely resilient that in many places it was more reliable then any "official"/"emergency" communication channels for info about natural catastrophes and similar

so people did rely on it for "critical" thing

and that was gone with Musks changes

which is one of the things people tended to overlook there was a huge decrease in reliability/resilance after the huge cuts, it's just that the final outcome was still "just fine" for what most people in the US/EU used it for


> and that was gone with Musks changes

Is there data/site to compare uptime before and after?


AFIK there where reasonable credible sources back then

but finding them again now sounds like a multiple hour task :/

and like the other comment said X stopped providing decent availability stats. And even if they where to provide this metrics, would you trust them? Like today I feel many uptime metrics are fudged/dishonest even from much more reputable companies (or maybe I'm just biased do to having to use GH at work which frequently is "practically" not usable in some way or another while supposedly there was no issue at all).

Anyway, to some degree nothing of this does matter.

Because for the purpose Musk wants X to fulfill the current reliability _is_ sufficient.

_My point isn't that it is not reliable enough_.

But that the hard cuts done had hug effects. Just in areas/ways a lot of Twitter users never cared about and Twitter (pre Musk) probably shouldn't have cared about either _and_ belonged into a lot of "you can get 80-90% of the way with a fraction of work" categories.

So instead of framing it as "cutting incompetent people/people with bad worker moral" I would frame it as "restructuring the company after changes to key strategies(/KPIs?) which allowed largely shrinking the company" (and which might have been a bit strange allocating many resources to things most users don't care about).


X stopped publishing uptimes (wonder why)

it has an incident page but it doesn't show how long each incident lasted

https://docs.x.com/incidents


> Anxious and lazy dependency-first development culture

I think this is the completely wrong takeaway.

Ober the last ~15 years I have probably used around ~10 web frameworks/libraries

_not a single one was even close to being complete or mostly correct, even just for standard HTTP ~1.1 handling_!!! (The routing, path, header handling. Not any of the complex stuff in HTTP2, Quick etc.)

the problem isn't using dependencies, that is the solution (not reinventing the wheel again).

And preferring reusing solutions which are more complete, practical testet and have more resources to maintain/improve basic functionality is (in most situations) not "anxious" development, but (should be) common sense.

The amount of hidden complexity in even somewhat basic stuff like HTTP is enourmous, rewriting everything all the time is wasting billions across the industry all the time and (in general) increases the chance for bugs.

The problem isn't "using dependencies" or small dependencies, but:

- availability (or lack) of high quality well maintained dependencies

- "no one" wanting to pay for software dependencies

- misguided tendencies for 'clever' & overengineered solution in one extreme (e.g. the stuff which lead to log4shell)

- misguided obsession with deceptive simplicity in artificial examples which lead to 1001 incomplete solutions because of the obsessiv simplicity doesn't stop at cutting out over-engineering but also cuts out fundamental parts

- lack of widely adopted, effective, tooling & infrastructure to trust dependencies.

- lack of tooling to more efficient reuse and compose working parts, the way we do programming seems just very fundamental broken.

- also a lot of legacy tech we are stuck with making many things unnecessary complicated

Also as a side note, number of packages != number of entities to trust. Weather you use 10 or 1 package from the same source, maintained and published in the same way often makes little to no difference for supply chain security.


There's a lot that's right here. I see some of the EU, and specifically German government projects to fund open source code platforms and packages and I wonder if we're going to see more of that. Governments increasingly afraid of this lack of stability stepping in to fund people and projects that are fundamental to the software they use, even if the tech companies refuse to fund those projects and developers.

Lets be fair there are many RFCs and they are often ambiguous or clash with other RFCS. Its hard to know which one is the authority about all or even a single feature.

Take CSV, there are multiple RFCs that fully or slightly cover ab out the standard if you want to call it that.


That’s a long list of issues that go away or you gain control over if you choose not to take a dependency…

through IMHO when not using dependencies issues aren't any smaller at all. A long list of "things which can be improved" doesn't mean it's unusable or worse then a alternative.

and some of the things listed are fundamental software issues you _don't_ avoid just by reinventing the wheel, they just change how they appear

honestly in my experience, for most things, problems with reinventing the wheel are way long term larger then problems with re-using software, at least if the re-using is done responsible

I have seen way to many cases of companies trying to avoid using dependencies for something "because this thing is simple" and then realized they are mishandling this, had an outage due to that, had to add those features etc. etc.

In then end often even just taking the 3rd party dependency, giving it a short review and then vendoring it, with rare updates to it, would nearly always have lead to a far better outcome.

(Which is also the recommendation I give for guidelines for "not-too-large 3rd party leaf dependencies", given them a small review, vendor them, keep an eye out(1) for CVEs and then only update when needed with another small review. Through note that this doesn't work for all things, the "not-too-large leaf dependencies" part matters. For other things other guidelines are better.)

(1): As part of normal CVE/Security scanning.


Consider the true scope of just this work for any given ecosystem:

> lack of widely adopted, effective, tooling & infrastructure to trust dependencies

How would I as an individual developer, or even a CTO at a small to midsize company, justify the investment to fix that? It would take buy-in from a Google-level entity to create and rollout such tooling for, say, Java.

Copying a few functions, or even recreating a whole HTTP server from scratch, is a smaller body of work. Sure, I could vendor and keep an eye out for CVEs, that’s a valid approach. But that’s me copying and taking ownership of an external codebase, not taking a dependency. Ownership is easiest for code that you’ve written, because you (or your team) also built the mental model to go along with it.

And if I have to vet a new codebase anyway, I’d rather it be the smallest subset that I actually use; not the full functionality of NGINX.

I can justify making the locally optimal decision (reimplement some functionality so I can truly own it) and that may result in a less than optimal global outcome. I can’t justify doing to the globally optimal thing at the expense of my company.

Ultimately, the incentives aren’t aligned. The activation energy to get from locally optimal to global optimal is too high, and that’s not going to change any time soon. In fact, it’s rapidly trending the other direction. AI makes it easy to reimplement the functionality you need. An idealized view of “if only people paid for high-quality dependencies this would be solved” is not one I can buy in to. That’s not the reality I live in, for better or worse. In a world where nobody wants to shoulder the burden of vetting dependencies, avoiding them is often the best alternative.


> looks like an administration in trouble trying to downplay the whole thing

idk. (more reputable) news where from the very beginning using terms like "suspected spy" and similar

as far as I can tell "she is suspected of it, but no evidence exists" might have been the simply truth story from the beginning

but there had been a lot of dogey things going on in context of Nvidea<->China, no matter if Nvidia is involved, and "boss has affair with spy" is the perfect instrument to funnel all the existing dissatisfaction and warning others to not be "too" friendly with china, even privately.


no him having affirs or multiple ones isn't necessary the problem here

I mean IFF you having a "proper" image matters then it is, I don't think this is the case here.

The problem here is him failing to keep work and affairs separate, one of them seeming to have a high chance to be a Chinese spy AND most relevantly a long ongoing problem with export restricted products of your company ending up in China in multiple different ways which look like insider knowledge was involved and to make that worse some of the past situations did look a lot like Nvidia actively helping/taking advantage of the black market. This wasn't well covered by dominant US media, but enough people with influence have noticed it anyway.

so no this hasn't much to do with racism, but long standing issues being funneled through an outrage due to "improper behavior" while already many stock holders where expecting an incoming stock downturn for one or another reason.

putting that aside, if Musk wouldn't own the companies he leads with dominant shares he would have long been ousted. Actually various stock holders have multiple times tried to oust him due to him acting in his best interest instead of that of the company AFIK


> training every boot

except this should only happen _once_ (until you remove/replace components, reset the BIOS etc.). Which is why it normally doesn't happen for laptops, it was run once in the factory and DDR being soldered one mean they can just "fixed" disable the retraining avoiding bugs in the "retraining needed" detection

but yes, there are pretty bad motherboards out there, like the one I have turned out to be a complete shit show with multiple sever design issues I unluckily "lucked out of not running into" in the first 2 years of usage making a return unlikely and a replacement is useless as this is are fundamental issue affecting all models AFIK.


yeah,

DDR5 training is a huge (UX) issues wrt. boot times (mainly on desktop systems).

sure only "once" if your system is correctly setup, but I have seen too many cases of buggy motherboards not properly handling this

and the biggest issue is the transparency, i.e. you don't see that it does DDR5 "training"/"parameter self adaption". It looks like it's "just stuck" in early EFI.


I was unprepared for how long training would take on my Framework laptop kit with 64 GB of DDR5, and actually thought there was something wrong with it.

Nope, training just took minutes, with nothing but a black screen.


Similar with my new NUC which also used DDR5. I genuinely thought there was something wrong, powered it off and reseated RAM and such. Kept debugging it for about 30 minutes before I searched and found out it might be training. Left it alone and sure enough after about 15 minutes the display started showing stuff.

I returned a motherboard because I thought it was DOA because it wouldn't boot but in the end it was just training for 96GB. It took around 10 minutes of black screen before I saw anything.

Me too. 128gb of ram and just a light on for several minutes. I only saw it boot because I left the second motherboard in the state while I went off to burn a new bios thumb drive.

128GB here too and I genuinely thought there was something seriously wrong with the system

I wasn't aware of DD5 training being expected to take that long often. That would explain why the box that is my current home-lab taking so long to boot when I first put it together (and on one other occasion) but behaving normally every other time. Every day is a school day!

A bit of a problem is that unless you add a PowerVM style boot, it's going to be hard adding indicators in today's "absolutely no status LEDs" world

wait how can ZIP code boundary lines not be free information, that's totally wild, how do you then even know what ZIP code to use if you can't look it up...

what next paying for knowing street names or house numbers??!??


> how do you then even know what ZIP code to use

There has always (since the late 90s) been a free look up webpage:

https://tools.usps.com/zip-code-lookup.htm

The data used to be available on CDROM (back in the 90s) but moved to an API in the early 2000s. Until August 2026, you used to be able to look up addresses for free.

To get the biggest discount on postage (one place I worked at mailed about 3k envelopes per month) you have ZIP+4, to pre-sort the mail and print a bunch of stuff on that barcode [0]. Next time you get an envelope, look at the barcode on the bottom.

https://en.wikipedia.org/wiki/Intelligent_Mail_barcode

0 - doing it that way saved about $450/month versus not doing any of their work for them.


Other ways to save costs:

Through folding: ounces flat in a 9x12 is $1.98; folded in a #10 it's $1.11. Square envelopes pay $0.49 extra.

Print package labels online: a 12 oz package coast to coast is $13.65 at the counter, $8.95 through USPS's free Click-N-Ship.

Keep boxes at or under 12x12x12. Since July, anything bigger is billed at volume ÷ 139 pounds if that exceeds its weight.

I've been building an (unofficial) searchable reference to USPS's rules and prices. The presort page is here: https://postalref.com/guides/presort


> The data used to be available on CDROM (back in the 90s) but moved to an API in the early 2000s. Until August 2026, you used to be able to look up addresses for free.

What was the reason for the 2026 change to make it non-free? Some "government should be run like a business" (and therefore charge for services) BS, or was there some kind of abuse or privacy issues they were trying to limit?


My university used to get copies of the CDs in the 1990s. I think they came out quarterly. You had to pay for the CDs, but the data was free. We bought one copy and put it on a server for everyone to use.

All governments provide some services for "free" and charge for others. Sometimes it makes sense, other times it really is BS.


Mostly as a commercial shift under the new board of governors that are concerned on financials (except when it comes to throwing away new efficient mail sorting machines ahead of an election).

They spun it as supporting the new RestAPI based system they put in place, it does have more efficient rate limiting, and some controls like Mailer ID verification (able to tie who makes a request for a label to that actual label, etc.) that are supposed to help with fraud.


And getting rid a no-brainer EV transition.

Iirc when I worked with these in the past ZIP codes did not map cleanly to boundaries, hence the ZCTAs. Zip codes are sometimes assigned to points, lines, they can overlap, and other weird arcana because they defined mail routes rather than tesselating areas.

This is a good example of the hardest part of data modeling. Understanding what the thing is that you're modeling actually is.

That they define points makes them useful sometimes for e.g. US sales tax calculation software, though with all things it is lossy

Zip codes do not have clear boundary lines. A zip code is a collection of postal routes, and their borders tend to move about.

This isn't the USPS charging for existing data. They created a new data set that's only relevant to data brokers, and are charging for it.


> Zip [..] collection of postal routes

this in itself is quite unusual/seem kinda not clever

because now you are exposing a "implementation detail" about how you deliver the mail now to people in a way which is hard to change (or at least high friction for the people living there). But you should be able to change your postal routs ad-hoc whenever needed.

At the same time you lose many of the "dual-use" benefits many other countries get from it.

Like often (outside of the US) "zip" codes:

- are used for mail disambiguation (in combination with city/region and street names)

- cover (very very very roughly) a similar amount of people

- hence can be used for polls, statistics etc. etc.

- and due to not being to narrow scoped disclosing your zip code doesn't feel like major privacy invasion, so you actually get data for polls where e.g. your customers are from by asking at the checkout if they are willing to tell you their zip code

- are often also used for all kinds of administrative area scoping (not all, e.g. not voting areas or zoning laws)

so the us using a collection of postal routs which kinda somewhat but not maps to boundary lines seem like a major lost opportunity


And yet, administrative boundaries should be free.

Census tracts are, but those don't align to addresses in any meaningful way.

How are we to improve our Zones if the Zone Improvement Plan (ZIP) Code is not readily available?


Crazy, right? I found this out when I didn't get a wedding invitation from my friend who lives 4 blocks away. The only indication was that the invite returned to sender. Turns out we're in different zip codes.

with the limited available information it looks a few _magnitudes_ worse :(


unlikely

IFF there is a NSA backdoor it likely would be:

- not in paint, but something you can't disable/remove in hardened special purpose windows setups

- not be easily usable by 3rd parties, like some auth accidentally accepting some quite specific other password (e.g. due to some bug reading the expected hash from the wrong place) or similar

lets not forget as much as the NSA might want to spy on you they also want to keep China/Russia out of spying on US companies. A traditional RCE in paint is very counter productive for that.


They (and other governments) can do multiple things at the same time.

One one hand they can actively try to work with companies to get secret back doors put in, sure. But I imagine that's not usually very successful. Lavabit chose to go out of business instead of doing so... I wonder who else is out there that we don't know about.

But we also have evidence that they actively look for, buy, and hoard, unknown (to the vendor) exploits for their own possible later use in case they need it for something.

I'm sure you can imagine that it's much more advantageous to keep collecting all kinds of different ways to get into a system in case one method is not applicable for any number of reasons.


unless they actively work with MS to keep their backdoor fresh, one can reasonably assume a potential backdoor _was_ well-hidden and advanced for the time period when it was introduced. What about after 20 years and LLMs decompiling everything?


Why hide blatant doors when they’re readily available?


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

Search: