brr.fyi/posts/engineering-for-slow-internet we should send all framework developers to deploy for 3 years at the South Pole

dats fuking lame dip shit

in that article, "South Pole" means specifically Amundsen-Scott?

... yeah, I guess the current trend in JS development is to either have no error handling or to intentionally render it useless somehow

this is *not* an edge case, these are things that should work, at the very least don't handle "slow" or "unreliable connection" as a failure...

remember when programs could fit on floppies?

also, the poster forgot one entry in the list of people who could see issues like this: "There are a lot of people behind Broadcom WLAN NICs"

tomman: we could also replace these framework developer's network gear with 10-base2

njsg: ironically the dude that wrote those is a SV hipster that yes, he went to South Pole for a year

life at Amundsen-Scott precludes 50MB JS framework bloats and "Netflix and chill"

what I didn't knew was the hostile anti-caching measures webdevs were using for pissing off users on unreliable connections

at least in hindsight it does feel a bit obvious some people would come up with that

but yeah, I don't think I remember seeing that before, at least not so clearly

> Apple naively assumed that the 1 GB download would be so fast that they didn’t bother incorporating download speed feedback into the updater’s time estimate. They did not anticipate people installing macOS updates from a location where a gigabyte of downloads can take several hours, if not days.

Apple doesn't believe in users living in places with spotty/slow Internet, film at 11

the engineers at Cupertino have never seen anything but unlimited gigabit fiber since they started working there

> Credit where credit is due – Microsoft has a GREAT auto-updater built into Office for Mac!

now THAT'S an irony

apparently it's even better than Windows Update, but then MS usually has written a better Office for Mac than for Windows

Windows Update, at least in NT 5.1, had serious issues, I think somewhere that was attributed to throttling by MS, for older versions

and I don't think I recall cases in recent times where update progress bars tracked anything more than the difference between 0% and 100% for each update, even in update-specific "wizard"-like dialogs it seems to lack precision

I've seen the progress indicator on Google Play Store going _over_ 100% many times

at this point it looks like getting a list of downloads applied and a list of downloads available and manually fetching one by one is much, much faster

plus, at least for NT 6.1, Microsoft still hasn't understood the concept of dependency management

Windows Update will take ages to propose an update that requires another update, but without pulling in the second one, then repeating the process probably requires starting from scratch zero, and unless the missing dependency is installed, it'll all fail again

Windows Update has always been unreliable trash since its early days, back when it was named under the unfortunate name of Critical Updates Notification Tool

then the very first implementation of what we know today as Windows Update (delivered with 98SE/Me) was... well, there was a reason many people just ignored it :D

I'm not even sure I've ever known it by a name other than "Windows Update", that said I think either that was v4 or at some point was "updated" to v4, so not an early one, I presume

IIRC the story goes like this: Critical Updates Notification Tool/Utility -> WU v1 -> v2 -> v3 -> v4 -> v5

v1 was shortlived, v2 debuted with 2K/Me (and backported to 98), v3 came with XP

well, the one for windows 4.10 IIRC at least allowed a workflow of "go check for updates, you get a window". In my experience, more recent ones are like "go check for updates" [minimize] [force re-open] "checking for updates" [check completes, minimizes again] [reopen, go over list] "install" [minimizes again], all along with all the settings that'll nag you to no end unless you opt by having at least

something automated

uh? then why do I remember seeing "v4" in 98?

they really ruined it with Windows 10 and its complete inability for the end user to manage updates

njsg: IIRC v4 was the last to support 9x

ah, that's it, then. It's not like I used 98 for a short time.

hmmm, looks like V1 was the Notification Tool

WU debuted at v2

hmmm, for some reason video playback is breaking after some time here

videos load, but won't play or even show a frame at all

after restarting SeaMonkey, everything works OK

is scripting involved, or that can be reproduced when the tab content is the video itself?

here is the use case I know: open any tab with a video (so far I'm noticing that on Danbooru, but then I rarely visit video sites these days)

it will work fine

but after a few days (dunno how many), open the same tabs and video will not show at all, just the player controls

no sound will play either

but the network inspector says it actually loads

tomman: so the tab is always loaded anew?

or do you mean switching back to it again some days later?

no, a new tab

does it only happen with videos that were already played in the session, or does it affect any video?

say, would a certain green-on-black weblog what has moved to videos instead of gifs reproduce this too?

Basically it seems to be "any video"

even ones previously played on that session, or being played for the first time

lemme check anywya

I suppose Chatzilla, Mail&News and/or Lightning might have "sound alerts" of some sort?

what I'm thinking is the question "is audio failing too?"

videos indeed do not play on THAT site too

the thumbnail loads fine

what environment is this on? were you on b1pre? pulseaudio on linux? or a custom build?

but once you click on Play... the player does nothing

Build ID 20240504213200

oh and actually, if you have an instance open where you can reproduce this, do you see any relevant error in the browser console?

the player doesn't even tell the proper video duration - it's all 0:00

I've seen that with videos playing, before they load, might be just how it starts

I never looked too much into it, most of my experience with video is Mastodon's insistence that everything must be transcoded to something that's not GIF even if it says GIF

the error console come clean

and videos indeed DO download, they just won't play or even decode the very first frame

(just to be sure, you checked the *browser* error console?)

there is also at least one preference controlling JS logging, I'm not sure if that might be hiding something.

yes, that browser error console is empty

but next time I restart the browser, I'll keep it open aside

with the debugging dump setting(s?) enabled, if you are on linux, starting from a terminal device should print stuff there too

(terminal emulator probably, as this is the kind of stuff where having a dedicated scrollback buffer with easy copy/select-paste helps :-)

but I do know some DEC VTs have enough scrollback and copy-paste too)

oh wow. I was having slowness issues in M&N today. I just realized: It's probably an emoji causing font fallback.

emoji fonts: one of the most backward decisions in tech

Anyway, I'll update my nightly now~

OK, backed up profile, updated, restarted (twice!), played a couple of videos, everything normal so far

but I noticed the error console is clearing itself after a few seconds

or to be more precise, after page navigation events

tomman: that looks like the page console in the devtools inspector

I noticed that this is not usable with keyboard

the browser does not allow opening that dropdown menu by the tab key

«input type=checkbox»?

even if I switch caret browsing on, it does not allow selecting that dropdown menu with keyboard

maybe that's something that should be possible for accessibility purposes

that is a very basic dropdown menu that is done completely without JavaScript

it has to be a checkbox so that it remembers its state

njsg: no, that's the error console

ah, I didn't notice at first that the checkbox isn't visible, so this is about the labels? how should keyboard focus work for these? any chance it's intended to work this way, i.e. labels are skipped in tab order because the tab order would also have the elements the labels apply too?

but keyboard usage is not possible even in caret browsing mode

tomman: more traditional, XUL-looking? Hm, here that console doesn't behave that way

Anyway, your broken website of the day: hmd.com/es_lta/nokia-215-4g

Sompi: an aside, I'm probably lacking some context, but why not <SELECT>?

when the caret is in the "Drop down menu" box and I press enter or space, nothing happens

"Nokia" doesn't want me learning about their cellphones, it seems

you see fragments of a site flashing briefly, then.. nothing

the only thing that works is the pointless GDPR cookie dialog

tomman: argh, nokia. ah yes, I've seen nokia sites failing recently too, didn't check HMD, though

oh, and the site uses React

njsg: the purpose was to write a "modern"-looking stylished dropdown menu without using javascript

the HTML SELECT element has some limitations in styles

the cookie prompt works flawlessly

but then I noticed that at least firefox and seamonkey don't allow using that with keyboard

everything else is a WSoD

tomman: also, I still can't think of Nokia without remembering passing by the Keilaniemi buildings and noticing it wasn't saying Nokia there anymore, and instead had some colored squares...

hence why I said "Nokia", as HMD is clearly *not* the home of Nokia phones

didn't it go three-way, Nokia kept all but phones, Microsoft got the smartphones, and HMD got feature phones?

it's... more complex

Initially all cellphones went to Microsoft

Sompi: do these need to be labels? any chance you could achieve this with inputs? Sorry if I keep asking about alternatives, this is not something where I'm able to give a good answer

then HMD got created and received the dumbphones from MS, and eventually MS stopped doing cellphones completely

while once the relevant exclusivity agreements ended, HMD also got to make Nokia-brand smartphones

so far my guess is just that above: maybe it's because a label would have an input/UI element to go with it, and so maybe it isn't included in tab order

OK, the stupid page doesn't break right away

it briefly loads and shows a broken page, then more elements disappear, and finally the entire page goes away

I guess the thing is that webdesigners must realize doing things in chronological order helps

as in, you don't disappear the page, you appear the page

njsg: I don't know, but that should work with TAB or at least in the caret browsing mode. But it doesn't

initial load: broken page. 2 seconds later: some elements go away. ~5-6 seconds later: the rest of whatever was on screen vanishes away, including the scrollbar

it's... a very interesting failure mode

tomman: react and jquery have that idiotic default error handler for everything that it just deletes all elements on the page and that's it

"destroy all evidence"

'document.body.innerHTML = ""' or something

FLAWLESS VICTORY

OK, after checking the document tree after failure, let's see what I got left

It's almost like they are purposely doing things as hard to debug as possible

indeed it removed EVERYTHING, save for a bunch of script tags... and a <svg> that contains... sprites?

njsg: I think that a label for a checkbox SHOULD be selectable by navigating to it using TAB

hitting ESC many times don't stop the carnage

Sompi: the thing is, the checkbox itself is also reachable the same way, isn't it?

njsg: yes, but in this case the checkbox is hidden

what I'm getting at is that labels are, I think, meant to label visible form elements, so it's possible they're not focusable by design. but that's just a guess.

then why doesn't it work even in caret browsing mode?

that at least should work

proving a point

Sompi: no ideas so far, but using "opacity: 0" instead of "display: none" at least gives something that is in the tab order

alsp GTK4 is fucked and GTK5 will kill everything beware

I am forking GTK3 components for my self

(still causes the "hidden" checkbox to take empty space, though)

njsg: doesn't work either

Sompi: how so?

I can tab to it

The label element itself isn't tabbable for some reason

when the checkbox is hidden

oh wait, as in, I can tab to something, but really won't show feedback in the labels; it's possible caret browsing works now if there's a "visible" element for the label

yes

also when the checkbox is "opacity:0;" instead of "display:none;" tab browsing still doesn't work, and neither does caret browsing

works for me, maybe I did something else, need to check

hmm, actually tab browsing works but it doesn't show any visual feedback

so it is not very usable

and also the checkbox still consumes space from the page