5 ms·

Standards-compliant web browsers are impossibly complex to build and maintain, which is why the handful which exist are all under powerful commercial influence.

Standards-compliant web browsers are impossibly complex to build and maintain, which is why the handful which exist are all under powerful commercial influence. If you want to "fix" the internet, spend some time enriching open, federated protocols that are designed to be human-scale. The internet is here to stay, but we can do better than the web as it presently exists. Gemini is worth looking into: https://gemini.circumlunar.space https://gemini.circumlunar.space

catalogia 6y ago
Drew DeVault has a good blog article about this problem. Choice quote: > I used wget to download all 1,217 of the W3C specifications which have been published at the time of writing, of which web browsers need to implement a substantial subset in order to provide a modern web experience. I ran a word count on all of these specifications. How complex would you guess the web is? > The total word count of the W3C specification catalogue is 114 million words at the time of writing. If you added the combined word counts of the C11, C++17, UEFI, USB 3.2, and POSIX specifications, all 8,754 published RFCs, and the combined word counts of everything on Wikipedia’s list of longest novels, you would be 12 million words short of the W3C specifications. https://drewdevault.com/2020/03/18/Reckless-limitless-scope.html https://drewdevault.com/2020/03/18/Reckless-limitless-scope....
- captn3m0 6y ago
  Wow. Who would like a w1c specification set, with everything trimmed down like how k3s.
- gsnedders 6y ago
  This seems exceptionally misleading. Playing with the filters on https://www.w3.org/TR/ https://www.w3.org/TR/, we find out filtering to latest documents (to exclude cases where there are multiple versions of a single document) brings us down to 953, excluding non-normative notes brings us down to 573, we can likely exclude all semweb-related specs to bring us down to 480 specs. And that's still including 199 working drafts, not all of which have any implementation (and some are unlikely to ever have any at this point, essentially long-abandoned). The web platform arguably has problems with scale, but it's nowhere near that extreme.
  - bryanrasmussen 6y ago
    Also if you were starting a browser development nowadays you could probably safely strip out most of the XML specs as not that relevant. Keep XML, XML Namespaces, drop XSL-T (despite how much I love it, I would not implement if I were building a browser), XSL-FO (never had browser implementation tht I'm aware of), XML Schema and probably some other stuff I'm forgetting. Those are some wordy specs we would be dropping.
carapace 6y ago
I've been playing with Gemini and I love it. On the one hand it's so constrained it's almost a toy, but on the other it's usable and fun! As far as interacting with the WWW vs "Gemini space" it's night and day, or should I say nightmare and day.
- RodgerTheGreat 6y ago
  It's flexible enough that you could write a variety of interesting services in it. The primary limitation would be the amount of information you can feed in with URLs capped at 1024 bytes. Mostly-read-only APIs like fetching maps are still quite practical. With INPUT, REDIRECT, and the mechanism for creating sessions via transient certificates, field-at-a-time forms are possible. You could even build a pretty friendly experience around it, since the server can validate and re-prompt each field as you go. I've been thinking that a text/gemini browser could recognize link lines that end in .png/.gif/.jpeg and offer a button next to them to manually expand the imaages inline. It would be a nice balance between being able to see figures along with the rest of a document without bloating page fetches. Perhaps other data types would have sensible inline interpretations, too- CSV files become inline tables if you expand them, etc. Just a few ideas I'm knocking about.
  - z3t4 6y ago
    Once users get Tbit Internet broadband, web sites will be several GB big, or we would have a chicken vs egg problem.
  - TeMPOraL 6y ago
    I understand the protocol is simple, but there's no reason the client has to be. Taking your example: > CSV files become inline tables if you expand them, etc. Why not only make them inline tables, but make them searchable, pivotable and graphable? Embed a bunch of statistical tooling directly in the interface. It's something I dream would happen on the Web - if Google^H^H^H^H^H^Hweb standard bodies could agree on a new tag for tabular data, say <datatable>, that would - by default - come with aforementioned capabilities, and with enough styling and hooks to lure in the webdevs, then maybe all the tabular data that's on webpages could be finally made explorable in an efficient fashion...
kevin_thibedeau 6y ago
Those powerful commercial influences should use their muscle to force the use of protocols other than TCP rather than the silliness of piling everything onto port 80/443.
- pathseeker 6y ago
  I think you just mixed up HTTP and TCP. HTTP and it's secure friend HTTPS are on TCP port 80 and 443.
- snazz 6y ago
  They do. HTTP/3 is over QUIC, which uses UDP (on ports 80 and 443, as it should). That effort has been spearheaded by Google. Edit: Also, putting everything on port 443 is a major security and privacy boon. Internet service providers can no longer just block ports for services they don't like.
  - kevin_thibedeau 6y ago
    QUIC would never have been needed if it was actually possible to use SCTP without unnecessary roadblocks.
    - zozbot234 6y ago
      Is there an actual problem with routing SCTP over the Internet? ISTM that the main issue would be enterprise middleboxes being generally unable to deal with it, in which case QUIC may turn out to be more of a stopgap solution.
      - kevin_thibedeau 6y ago
        s/generally unable/not compliant with IP standards/
  - Reelin 6y ago
    This might be naive, but it bugs me that protocol ports are visible to intermediaries. HTTPS encrypts the URL path. Why don't TCP and UPD encrypt (or otherwise scramble) their port numbers? (Yes, I know, those protocols don't include cryptographic provisions. Maybe they should be updated?)
Taikonerd 6y ago
I love the idea behind Gemini, and on a similar note, I'm wondering: is there a name for the back-to-basics, web 1.0 philosophy? Like, a name for the "sites that look like https://motherfuckingwebsite.com https://motherfuckingwebsite.com " ?
- kixiQu 6y ago
  No, because the only traction it has ever had and will ever have is with techies who are capable of recognizing (and issuing) a description of "HTML documents with only very minimal client-side JS, often without any CSS whatsoever".
  - pbourke 6y ago
    or, you know, Craigslist and Wikipedia.
    - kixiQu 6y ago
      I don't think a case can be made for Wikipedia being on http://motherfuckingwebsite.com/ http://motherfuckingwebsite.com/ style level. Craigslist--maybe the listing pages, sure. And that's a fair point! It has more penetration than I was allowing. But I don't think it's going to get a designy sounding name because designy sorts of people don't feel the same nostalgic affection for fairly raw HTML.
- Reelin 6y ago
  Perhaps the idea of brutalist web design fits? (https://brutalist-web.design/ https://brutalist-web.design/) HN is a good example of it IMO.
  - clarry 6y ago
    It looks alright but to me the simple-web ideal is rather like.. no design. Like, you don't design an essay, or a comment, you just write one. And so was the web. Initially you did not do "web design", you wrote hyperlinked documents with some markup to support formatting (not design!) and embedding images; how it looks in the end is up to the user agent (and transitively to the user who could pick and configure their UA). Then web design happened and it all went to hell. Web "designers" started doing "cute" things and business owners started wanting more and more branding and fashionable stuff.. and here we are. I've said it before: https://news.ycombinator.com/item?id=22608293 https://news.ycombinator.com/item?id=22608293
dschuessler 6y ago
> Gemini is a new, collaboratively designed internet protocol, which explores the space inbetween gopher and the web, striving to address (perceived) limitations of one while avoiding the (undeniable) pitfalls of the other. I had to look up what gopher is and still don't get the pitch. What are the perceived limitations of gopher? What are the undeniable pitfalls of the web? If the author ever reads this: Why not write something along the lines of "Gemini is an internet protocol that avoids {{ insert pitfalls of the web here }} while being {{ insert opposite of perceived limitations of gopher here }}."?
- RodgerTheGreat 6y ago
  Gopher was an early competitor to HTTP. Gopher documents ("gophermaps") have a simple tabular presentation. The types of data and links which can be placed in a gophermap are limited and non-extensible, as Gopher predates MIME types. Gopher also has no facility for encrypted communication. The web (both HTTP(S) and its preferential presentation format, HTML) is built on standards which are obscenely complex, which have features that do not have user privacy at heart (cookies, user-agent strings, css lazy-loading resources, etc., etc.), and these standards are, as I noted above, heavily guided by profit-motivated actors. While I agree that the top-level pitch might benefit from speaking to a broader audience, the FAQ and standard document linked from the main page provide answers to many of your questions.

catalogia 6y ago

Drew DeVault has a good blog article about this problem. Choice quote: > I used wget to download all 1,217 of the W3C specifications which have been published at the time of writing, of which web browsers need to implement a substantial subset in order to provide a modern web experience. I ran a word count on all of these specifications. How complex would you guess the web is? > The total word count of the W3C specification catalogue is 114 million words at the time of writing. If you added the combined word counts of the C11, C++17, UEFI, USB 3.2, and POSIX specifications, all 8,754 published RFCs, and the combined word counts of everything on Wikipedia’s list of longest novels, you would be 12 million words short of the W3C specifications. https://drewdevault.com/2020/03/18/Reckless-limitless-scope.html https://drewdevault.com/2020/03/18/Reckless-limitless-scope....

captn3m0 6y ago
Wow. Who would like a w1c specification set, with everything trimmed down like how k3s.
gsnedders 6y ago
This seems exceptionally misleading. Playing with the filters on https://www.w3.org/TR/ https://www.w3.org/TR/, we find out filtering to latest documents (to exclude cases where there are multiple versions of a single document) brings us down to 953, excluding non-normative notes brings us down to 573, we can likely exclude all semweb-related specs to bring us down to 480 specs. And that's still including 199 working drafts, not all of which have any implementation (and some are unlikely to ever have any at this point, essentially long-abandoned). The web platform arguably has problems with scale, but it's nowhere near that extreme.
- bryanrasmussen 6y ago
  Also if you were starting a browser development nowadays you could probably safely strip out most of the XML specs as not that relevant. Keep XML, XML Namespaces, drop XSL-T (despite how much I love it, I would not implement if I were building a browser), XSL-FO (never had browser implementation tht I'm aware of), XML Schema and probably some other stuff I'm forgetting. Those are some wordy specs we would be dropping.

carapace 6y ago

I've been playing with Gemini and I love it. On the one hand it's so constrained it's almost a toy, but on the other it's usable and fun! As far as interacting with the WWW vs "Gemini space" it's night and day, or should I say nightmare and day.

RodgerTheGreat 6y ago
It's flexible enough that you could write a variety of interesting services in it. The primary limitation would be the amount of information you can feed in with URLs capped at 1024 bytes. Mostly-read-only APIs like fetching maps are still quite practical. With INPUT, REDIRECT, and the mechanism for creating sessions via transient certificates, field-at-a-time forms are possible. You could even build a pretty friendly experience around it, since the server can validate and re-prompt each field as you go. I've been thinking that a text/gemini browser could recognize link lines that end in .png/.gif/.jpeg and offer a button next to them to manually expand the imaages inline. It would be a nice balance between being able to see figures along with the rest of a document without bloating page fetches. Perhaps other data types would have sensible inline interpretations, too- CSV files become inline tables if you expand them, etc. Just a few ideas I'm knocking about.
- z3t4 6y ago
  Once users get Tbit Internet broadband, web sites will be several GB big, or we would have a chicken vs egg problem.
- TeMPOraL 6y ago
  I understand the protocol is simple, but there's no reason the client has to be. Taking your example: > CSV files become inline tables if you expand them, etc. Why not only make them inline tables, but make them searchable, pivotable and graphable? Embed a bunch of statistical tooling directly in the interface. It's something I dream would happen on the Web - if Google^H^H^H^H^H^Hweb standard bodies could agree on a new tag for tabular data, say <datatable>, that would - by default - come with aforementioned capabilities, and with enough styling and hooks to lure in the webdevs, then maybe all the tabular data that's on webpages could be finally made explorable in an efficient fashion...

kevin_thibedeau 6y ago

Those powerful commercial influences should use their muscle to force the use of protocols other than TCP rather than the silliness of piling everything onto port 80/443.

pathseeker 6y ago
I think you just mixed up HTTP and TCP. HTTP and it's secure friend HTTPS are on TCP port 80 and 443.
snazz 6y ago
They do. HTTP/3 is over QUIC, which uses UDP (on ports 80 and 443, as it should). That effort has been spearheaded by Google. Edit: Also, putting everything on port 443 is a major security and privacy boon. Internet service providers can no longer just block ports for services they don't like.
- kevin_thibedeau 6y ago
  QUIC would never have been needed if it was actually possible to use SCTP without unnecessary roadblocks.
  - zozbot234 6y ago
    Is there an actual problem with routing SCTP over the Internet? ISTM that the main issue would be enterprise middleboxes being generally unable to deal with it, in which case QUIC may turn out to be more of a stopgap solution.
    - kevin_thibedeau 6y ago
      s/generally unable/not compliant with IP standards/
- Reelin 6y ago
  This might be naive, but it bugs me that protocol ports are visible to intermediaries. HTTPS encrypts the URL path. Why don't TCP and UPD encrypt (or otherwise scramble) their port numbers? (Yes, I know, those protocols don't include cryptographic provisions. Maybe they should be updated?)

Taikonerd 6y ago

I love the idea behind Gemini, and on a similar note, I'm wondering: is there a name for the back-to-basics, web 1.0 philosophy? Like, a name for the "sites that look like https://motherfuckingwebsite.com https://motherfuckingwebsite.com " ?

kixiQu 6y ago
No, because the only traction it has ever had and will ever have is with techies who are capable of recognizing (and issuing) a description of "HTML documents with only very minimal client-side JS, often without any CSS whatsoever".
- pbourke 6y ago
  or, you know, Craigslist and Wikipedia.
  - kixiQu 6y ago
    I don't think a case can be made for Wikipedia being on http://motherfuckingwebsite.com/ http://motherfuckingwebsite.com/ style level. Craigslist--maybe the listing pages, sure. And that's a fair point! It has more penetration than I was allowing. But I don't think it's going to get a designy sounding name because designy sorts of people don't feel the same nostalgic affection for fairly raw HTML.
Reelin 6y ago
Perhaps the idea of brutalist web design fits? (https://brutalist-web.design/ https://brutalist-web.design/) HN is a good example of it IMO.
- clarry 6y ago
  It looks alright but to me the simple-web ideal is rather like.. no design. Like, you don't design an essay, or a comment, you just write one. And so was the web. Initially you did not do "web design", you wrote hyperlinked documents with some markup to support formatting (not design!) and embedding images; how it looks in the end is up to the user agent (and transitively to the user who could pick and configure their UA). Then web design happened and it all went to hell. Web "designers" started doing "cute" things and business owners started wanting more and more branding and fashionable stuff.. and here we are. I've said it before: https://news.ycombinator.com/item?id=22608293 https://news.ycombinator.com/item?id=22608293

dschuessler 6y ago

> Gemini is a new, collaboratively designed internet protocol, which explores the space inbetween gopher and the web, striving to address (perceived) limitations of one while avoiding the (undeniable) pitfalls of the other. I had to look up what gopher is and still don't get the pitch. What are the perceived limitations of gopher? What are the undeniable pitfalls of the web? If the author ever reads this: Why not write something along the lines of "Gemini is an internet protocol that avoids {{ insert pitfalls of the web here }} while being {{ insert opposite of perceived limitations of gopher here }}."?

RodgerTheGreat 6y ago
Gopher was an early competitor to HTTP. Gopher documents ("gophermaps") have a simple tabular presentation. The types of data and links which can be placed in a gophermap are limited and non-extensible, as Gopher predates MIME types. Gopher also has no facility for encrypted communication. The web (both HTTP(S) and its preferential presentation format, HTML) is built on standards which are obscenely complex, which have features that do not have user privacy at heart (cookies, user-agent strings, css lazy-loading resources, etc., etc.), and these standards are, as I noted above, heavily guided by profit-motivated actors. While I agree that the top-level pitch might benefit from speaking to a broader audience, the FAQ and standard document linked from the main page provide answers to many of your questions.