Hacker News new | past | comments | ask | show | jobs | submit login

> I never suggested removing a language used by millions of people. Let me be more specific: Pruning support for old formats and protocols is a feature, not a bug.

You're contradicting yourself here; Hebrew is a language used by millions of people (about ten million, so 99.9% of the world's population does not use it) and also an old format (about 3000 years old). "Pruning" support for old formats makes history inaccessible; "pruning" support for old protocols requires constant effort to keep your servers compatible with whatever is fashionable with today's cascade of attention-deficit teenagers [CADT].

And if we follow your originally stated reasoning, "Continuing to support $x means continuing to defend attack surface that's implemented as $y year-old code, to deliver a feature that in $z the majority of people do not use," we ineluctably arrive at the conclusion that we should remove support for languages used by millions of people.

Taken literally, we should remove support for all languages, since no language is used by more than 50% of the world population, but in keeping with the principle of charity, I interpreted your "majority" as "vast majority". I'm not sure where exactly the vast-majority cutoff lies: a feature that 90% of people do not use? That would include all natural languages except English and Chinese. 95%? All languages except those, Hindi, and Spanish. (In particular, it leaves out all those RTL languages that cause so much complication in text rendering, like Arabic.) 99%? That leaves 20 languages, but not, for example, Persian, Swahili, Italian, or Thai. Even a cutoff of 99.9% might leave out Hebrew, Uighur, and Greek.

What percentage of users do you think use View Source? The web inspector? Printing?

That's not a strawman argument; it's a slippery-slope argument. And, I think, it's a valid slippery-slope argument. If we are going to avoid removing support for these things, we need a better basis on which to make the decision than, "Continuing to support $x means continuing to defend attack surface that's implemented as $y year-old code, to deliver a feature that in $z the [vast] majority of people do not use."

I do agree that there needs to be some kind of cutoff. Gopher is probably below it; WAIS and XBM certainly are. But FTP?

XBM was never very widely used in web pages because it didn't support color, grayscale, or compression, although for a little while it was the only image format supported by browsers that supported transparency. I did put it on a few of my web pages, but as soon as Netscape added support for transparent pixels in GIFs, I switched over and never looked back. This would have been about 01994.

By contrast, there are about 1.1 million anonymous FTP servers today, one out of every 4000 public IP addresses, and about one for every 30 HTTP(S) servers: https://zakird.com/papers/dsn-ftp.pdf That's more than the number of HTTP servers that existed for the first seven years of the Web, up to 01997: https://news.netcraft.com/archives/2021/05/31/may-2021-web-s...

You can be sure that there's millions of people using them. Probably more people than speak Hebrew, in fact.




We are at an impasse that can only be solved by data: how many unique users downloaded a document via an FTP URI in a web browser in a month? If I were a betting man, I'd wager it's less than the unique number of users who visited a webpage rendered using Hebrew, but we don't know.

Hopefully the browser makers will be transparent with this telemetry.


Precisely what I'm arguing is that we should stop destroying access to all information except the most popular—I'm astounded to find that you disagree! That is not an impasse that can be resolved by data.

Popularity is not a valid measure of value. The July 02021 issue of People magazine sold 3 million copies, in a single month, and is almost completely devoid of value. (Maybe in 02068 it will provide valuable insights into vapid 02020s US popular culture.) Amazon tells me Claude Shannon's Mathematical Theory of Communication is outsold by 185,210 other books at present, so perhaps it has sold 100 copies this month, but it is the foundation of data compression, error correction, and significant amounts of artificial intelligence work. One Hundred Years of Solitude has sold about 50 million copies—over the past 54 years, so perhaps it sells 80,000 copies a month, 40 times less than the July 02021 issue of People. (But probably less; it probably doesn't sell as much as it did 30 years ago.)

40:1 is more than the ratio between the number of HTTPS servers and the number of anonymous FTP servers.




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

Search: