That’s just… no. I do not believe I have ever encountered any software which would parse it in that way, and I refuse to believe such software ever existed. It would be <http://example.com/>.
But the PDF matches the HTML. I dunno, something weird is going on. Look at the hyperlinks there, too, “http://xn--ivg but not the rest of the URL that follows, and how the -- has been changed to –. Something went wrong somewhere in the editing or publication.
http:// is not a typo for https://. There's still a fairly large amount of web servers that do not talk https, and you simply cannot assume that they do. That will leave you with a lot of dead links. Besides, most that accept both will auto-renegotiate to https.
> There's still a fairly large amount of web servers that do not talk https, and you simply cannot assume that they do.
OTOH I'm browsing since years forcing HTTPS only and life goes on fine. If the absolute worse comes to worse, I can use archive.is or archive.org but it's very rare that I need that.
Basically: if a link is HTTP to me it's not worth opening.
The one exception would be Debian packages URLs: but these are signed and the signatures are verified.
User _apt is the only one allowed to emit HTTP traffic.
This prevents my ISP or anyone else injecting nasty stuff.
Just because it is accessible to you does not mean it is accessible to everyone else. HTTPS has many failure modes which make it unreliable for essential access, such as time mismatches, certificate expirations, ssl version mismatches, etc. Security and privacy are important, and they are also not absolute. Sometimes the risk is outweighed by the importance of being able to access essential resources and reading material.
User preferences should not be encoded into parser behavior, that’s nuts. You wouldn’t just arbitrarily change an ftp:// link to an imap:// link, so why would you accept it here? That exists at a whole other layer of the stack.
This sort of work is something I wouldn't be able to do, but I can't help but point out at least one potential issue with the paper. It's a lot easier to find problems than solutions I guess.
Are the benchmarks comparing node versions valid to conclude a real world performance increase?
I had a lot of fun writing low latency parsers for various message standards C++. There are a lot of fun things you can do when you can take ownership of the read buffer and you can figure out how to parse in-situ (modifying the data in place as you move along)
There are some auto rewrite rules. Off the top of my head: numbers in the beginning are stripped, [pdf] or [video] can be added to the end, and one more I can't remember that gets stripped off beginning and can cause confusion.
A pdf link to "5 Reasons To Do Things" will be "Reasons To Do Things [pdf]" for example.
Yes. And the algorithm is really incredibly stupid, but dang is opposed to even small improvements (like showing the changed title on submission beforehand, like the „x characters to long“ message).
I think anybody can edit a title within a short time of posting something. Or if there is a karma threshold it is way less than 80k.
I caught that one manually but YOShInOn's tail end needs some love and could be updated so it that it fixes up titles that get mashed automatically or adds a comment sometimes to editorialize or provide an archive link.
The article explains optimizations to spend less cycles parsing URLs than other libraries. Very interesting work, there's no reason not to do things efficiently when it's possible.
Also, good luck using regex to write a RFC or WHATWG conformant URL parser.
I guess there are reasons to do things efficiently when possible, but a million URLs is not it, the adage about the root of all evil comes to mind. A billion URLs per second and it's almost interesting, but not really.
RFC 3986 is a lot simpler than WHATWG spec. You can literally write a zero copy 3986 parser whereas you can’t with WHATWG. (And Ada is still faster than 3986 parsers)
> For example, the input string http://xn--6qqa088eba.xn--3ds443g/./a/../b/./c should be normalized to the string https://xn--xn6qqa088eba-l19f.xn--xn3ds-zu3b/b/c
Why would normalization change http:// to https:// ?
There’s got to be some accidental mangling there. Somewhere. Because of that error, and still more because of the blatant error in the next sentence:
> For example, given the base string http://example.org/foo/bar, the relative string http://example.com/ leads to the final URL http://example.org/example.com/.
That’s just… no. I do not believe I have ever encountered any software which would parse it in that way, and I refuse to believe such software ever existed. It would be <http://example.com/>.
But the PDF matches the HTML. I dunno, something weird is going on. Look at the hyperlinks there, too, “http://xn--ivg but not the rest of the URL that follows, and how the -- has been changed to –. Something went wrong somewhere in the editing or publication.
My guess is that the html formatter changed the text "example.com" into "http://example.com" to make it a valid absolute URL.
Because its 2024
http:// is not a typo for https://. There's still a fairly large amount of web servers that do not talk https, and you simply cannot assume that they do. That will leave you with a lot of dead links. Besides, most that accept both will auto-renegotiate to https.
> There's still a fairly large amount of web servers that do not talk https, and you simply cannot assume that they do.
OTOH I'm browsing since years forcing HTTPS only and life goes on fine. If the absolute worse comes to worse, I can use archive.is or archive.org but it's very rare that I need that.
Basically: if a link is HTTP to me it's not worth opening.
The one exception would be Debian packages URLs: but these are signed and the signatures are verified.
User _apt is the only one allowed to emit HTTP traffic.
This prevents my ISP or anyone else injecting nasty stuff.
Just because it is accessible to you does not mean it is accessible to everyone else. HTTPS has many failure modes which make it unreliable for essential access, such as time mismatches, certificate expirations, ssl version mismatches, etc. Security and privacy are important, and they are also not absolute. Sometimes the risk is outweighed by the importance of being able to access essential resources and reading material.
User preferences should not be encoded into parser behavior, that’s nuts. You wouldn’t just arbitrarily change an ftp:// link to an imap:// link, so why would you accept it here? That exists at a whole other layer of the stack.
They would arbitrarily change an ftp:// link to an sftp:// link and then complain that it didn't work.
This sort of work is something I wouldn't be able to do, but I can't help but point out at least one potential issue with the paper. It's a lot easier to find problems than solutions I guess.
Are the benchmarks comparing node versions valid to conclude a real world performance increase?
one possible confounder is the version of V8.
https://github.com/nodejs/node/blob/v18.x/deps/v8/include/v8... https://github.com/nodejs/node/blob/v20.x/deps/v8/include/v8...
ideally, they would've patched Node 18.15 with their changes directly and test their patch against 18.15.
I wonder how much time was spent promoting this parser, vs time spent on writing it? I've seen a lot of spam for this one, and I'm not the only one.
https://daniel.haxx.se/blog/2023/11/21/url-parser-performanc...
I had a lot of fun writing low latency parsers for various message standards C++. There are a lot of fun things you can do when you can take ownership of the read buffer and you can figure out how to parse in-situ (modifying the data in place as you move along)
Found the easier to read/download from Arxiv link
https://arxiv.org/abs/2311.10533
The title seems to have a few words missing. Original title:
> Parsing millions of URLs per second
HN’s stupid/arrogant automatic title rewriter strikes again
I've never noticed a title being rewritten automatically when posting an article. Are you sure that's really a thing?
There are some auto rewrite rules. Off the top of my head: numbers in the beginning are stripped, [pdf] or [video] can be added to the end, and one more I can't remember that gets stripped off beginning and can cause confusion.
A pdf link to "5 Reasons To Do Things" will be "Reasons To Do Things [pdf]" for example.
„How“ at the beginning is stripped, leading to all these strange sounding „I <verb>“ submissions.
Yes. And the algorithm is really incredibly stupid, but dang is opposed to even small improvements (like showing the changed title on submission beforehand, like the „x characters to long“ message).
Fixed
So, suprassing 80k karma, one gets title edit rights?
I think anybody can edit a title within a short time of posting something. Or if there is a karma threshold it is way less than 80k.
I caught that one manually but YOShInOn's tail end needs some love and could be updated so it that it fixes up titles that get mashed automatically or adds a comment sometimes to editorialize or provide an archive link.
Lemire’s blog is well worth a read if you’re interested in this sort of thing https://lemire.me/blog/
The title doesn't sound impressive at all to make me want to read this
a regex can parse millions of URLS on a home computer (say 4GHz, you get 4000 cycles per URL.)
These things were desgined to be parsed.
Maybe you should’ve spent 2 minutes reading the article instead of arrogantly dismissing it with layman knowledge.
The article explains optimizations to spend less cycles parsing URLs than other libraries. Very interesting work, there's no reason not to do things efficiently when it's possible.
Also, good luck using regex to write a RFC or WHATWG conformant URL parser.
2 minutes reading an rfc about uris and I find a regex literally used in the specs:
https://www.rfc-editor.org/rfc/rfc3986#page-7
>> The following line is the regular expression for breaking-down a well-formed URI reference into its components.
>> ^(([^:/?#]+):)?(//([^/?#]))?([^?#])(\?([^#]))?(#(.))?
Did Chomsky die for nothing?
I guess there are reasons to do things efficiently when possible, but a million URLs is not it, the adage about the root of all evil comes to mind. A billion URLs per second and it's almost interesting, but not really.
RFC 3986 is a lot simpler than WHATWG spec. You can literally write a zero copy 3986 parser whereas you can’t with WHATWG. (And Ada is still faster than 3986 parsers)
Last I heard Noam Chomsky was still alive, and a quick Google doesn’t contradict that. Or is this some kind of high brow joke that went over my head?
The spec provides an expression for "well-formed" URIs. Good luck with real-world input.
That will get you only one million per second.
And depending on the length of your URL, 4000 will not be trivial, depending on the output format.
Correct, URLs can be something like 4,000 characters long in 15 year old Firefox. I wonder what the current maximum length is?
Today, Chrome supports 32,768 characters.. good luck processing that in 4,000 cycles! It'd require SIMD or some other fanciness.
> That will get you only one million per second.
But per core, right?