-
Notifications
You must be signed in to change notification settings - Fork 29.6k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
http: make parser choice a runtime flag #24739
Conversation
CI: https://ci.nodejs.org/job/node-test-pull-request/19078/ @nodejs/http @nodejs/http-parser |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Nice!
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Does lib/internal/print_help.js
also need to be updated?
@cjihrig No, it takes its data from the options list itself. :) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM. Very clever!
Might have merge conflicts with #24738 |
Continuing from this comment I'm -1 on this flag. Switching from Removing this flag, OTOH, will be semver major. Unless the plan is to remove this flag before Node 12 is released, we will have to wait 2 semver majors to remove this flag (and hence remove |
@ofrobots Thanks for weighing in – If this really becomes problematic (i.e. http_parser becomes too much of a burden to maintain), we could make the flag a no-op, and use llhttp regardless of what was passed on the command line, if we’re sufficiently satisfied with the results? On the other hand, introducing this flag now would give us another 4 – 5 months of allowing (easier) ecosystem testing as an opt-in switch? |
Maybe instead of making the legacy parser switch a noop in the future, we only introduce If we keep it as a build-time flag, I doubt how we can reach to a point where it's safe enough to ship it without enough experimentation in the user land. |
@joyeecheung I think that would reduce our ability to provide the legacy parser for as long as it makes sense... I guess that would be okay, but I'm not quite convinced we need to do that? |
<!-- YAML | ||
added: REPLACEME | ||
--> | ||
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Or maybe we should mark this flag itself experimental?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Do you see any kind of actual breaking changes that we could make to it? I wouldn’t consider making it a no-op a breaking change. I’d be happy to add a warning about that (possibly no-op-ing it) here, though?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
My reasoning is that the part being experimental is "the ability to choose your HTTPParser". So, while this flag is experimental, the user cannot always expect that they have the ability to choose that, and the flag can be a noop in a particular release (without going semver-major) because of security reasons, which I think doesn't matter that much for people who do experiment with it, and they will be given proper warning about this fact.
When we are confident enough we will just remove the old parser along with this flag, or make the flag a noop and deprecate it. Since in our stability contract, being experimental means the feature itself ("the ability to choose your HTTPParser") may be removed completely. The ability itself will not actually graduate into stable - it will be first experimental and then straight into deprecation.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
That sounds a bit complex, imo? Wouldn't something "This flag is likely to become a no-op and later be removed at some point in the future" be explicit enough + do the trick?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@addaleax What's the difference between that and making this experimental? In my understanding, being experimental means: a flag + a warning when it's used + an orange banner in the documentation + freedom to break it (including making it a noop) in a release line, which I think is what we want for this? (Or maybe my understanding of the experimental status is wrong, or my understanding of your reasoning is different from what you have in mind?)
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@joyeecheung The difference is that experimental
usually implies that people should be careful with using it in production, which doesn’t apply here, I think.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
@addaleax I see. I am tagging this tsc-review
(I don't think tagging it tsc-agenda
and discussing it in a meeting would help?) so we can fully discuss about the status of this flag, just to make sure we don't overlook something before pushing it out to users..
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. Refs: nodejs#24730
traditional → legacy
c54665a
to
ff9d9cc
Compare
I’ve rebased this. CI: https://ci.nodejs.org/job/node-test-pull-request/19179/ @ofrobots Could you take another look, esp. at the added warning in the CLI docs (ff9d9cc)? |
I've tagged this @addaleax 's take is that we don't need to discourage users to use it in production so it doesn't need to be experimental. I am currently inclined to making this flag experimental first just so that it fits better into our stability model, but I don't feel very strongly about that. See #24739 (review) for discussion. |
Fwiw, what I have in mind is:
|
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM
Had a linter failure after addressing a comment, sorry. New CI: https://ci.nodejs.org/job/node-test-pull-request/19227/ |
Resume CI: https://ci.nodejs.org/job/node-test-pull-request/19233/ ✔️ |
Resume Build CI: https://ci.nodejs.org/job/node-test-pull-request/19236/ ✔️ |
@Trott There was already a resume CI scheduled, it seems to be blocked on only 1 AIX machine being available I think? Did I miss something? |
@addaleax No, all my fault. I saw the failure in the CI interface and resumed without looking to see if one was already running. I'll try to do better. |
Landed in aa943d0. |
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. PR-URL: #24739 Refs: #24730 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Gus Caplan <me@gus.host> Reviewed-By: Matheus Marchini <mat@mmarchini.me> Reviewed-By: Matteo Collina <matteo.collina@gmail.com> Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. PR-URL: #24739 Refs: #24730 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Gus Caplan <me@gus.host> Reviewed-By: Matheus Marchini <mat@mmarchini.me> Reviewed-By: Matteo Collina <matteo.collina@gmail.com> Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Notable Changes: * console,util: * `console` functions now handle symbols as defined in the spec. #23708 * The inspection `depth` default is now back at 2. #24326 * dgram,net: * Added ipv6Only option for `net` and `dgram`. #23798 * http: * Chosing between the http parser is now possible per runtime flag. #24739 * readline: * The `readline` module now supports async iterators. #23916 * repl: * The multiline history feature is removed. #24804 * tls: * Added min/max protocol version options. #24405 * The X.509 public key info now includes the RSA bit size and the elliptic curve. #24358 * url: * `pathToFileURL()` now supports LF, CR and TAB. #23720 * Windows: * Tools are not installed using Boxstarter anymore. #24677 * The install-tools scripts or now included in the dist. #24233 * Added new collaborator: * [antsmartian](https://github.com/antsmartian) - Anto Aravinth. #24655 PR-URL: #24854
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. PR-URL: #24739 Refs: #24730 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Gus Caplan <me@gus.host> Reviewed-By: Matheus Marchini <mat@mmarchini.me> Reviewed-By: Matteo Collina <matteo.collina@gmail.com> Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. PR-URL: #24739 Refs: #24730 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Gus Caplan <me@gus.host> Reviewed-By: Matheus Marchini <mat@mmarchini.me> Reviewed-By: Matteo Collina <matteo.collina@gmail.com> Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Notable Changes: * console,util: * `console` functions now handle symbols as defined in the spec. #23708 * The inspection `depth` default is now back at 2. #24326 * dgram,net: * Added ipv6Only option for `net` and `dgram`. #23798 * http: * Chosing between the http parser is now possible per runtime flag. #24739 * readline: * The `readline` module now supports async iterators. #23916 * repl: * The multiline history feature is removed. #24804 * tls: * Added min/max protocol version options. #24405 * The X.509 public key info now includes the RSA bit size and the elliptic curve. #24358 * url: * `pathToFileURL()` now supports LF, CR and TAB. #23720 * Windows: * Tools are not installed using Boxstarter anymore. #24677 * The install-tools scripts or now included in the dist. #24233 * Added new collaborator: * [antsmartian](https://github.com/antsmartian) - Anto Aravinth. #24655 PR-URL: #24854
Notable Changes: * console,util: * `console` functions now handle symbols as defined in the spec. #23708 * The inspection `depth` default is now back at 2. #24326 * dgram,net: * Added ipv6Only option for `net` and `dgram`. #23798 * http: * Chosing between the http parser is now possible per runtime flag. #24739 * readline: * The `readline` module now supports async iterators. #23916 * repl: * The multiline history feature is removed. #24804 * tls: * Added min/max protocol version options. #24405 * The X.509 public key info now includes the RSA bit size and the elliptic curve. #24358 * url: * `pathToFileURL()` now supports LF, CR and TAB. #23720 * Windows: * Tools are not installed using Boxstarter anymore. #24677 * The install-tools scripts or now included in the dist. #24233 * Added new collaborator: * [antsmartian](https://github.com/antsmartian) - Anto Aravinth. #24655 PR-URL: #24854
Add a `--http-parser=llhttp` vs `--http-parser=traditional` command line switch, to make testing and comparing the new llhttp-based implementation easier. PR-URL: nodejs#24739 Refs: nodejs#24730 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Gus Caplan <me@gus.host> Reviewed-By: Matheus Marchini <mat@mmarchini.me> Reviewed-By: Matteo Collina <matteo.collina@gmail.com> Reviewed-By: Ali Ijaz Sheikh <ofrobots@google.com>
Notable Changes: * console,util: * `console` functions now handle symbols as defined in the spec. nodejs#23708 * The inspection `depth` default is now back at 2. nodejs#24326 * dgram,net: * Added ipv6Only option for `net` and `dgram`. nodejs#23798 * http: * Chosing between the http parser is now possible per runtime flag. nodejs#24739 * readline: * The `readline` module now supports async iterators. nodejs#23916 * repl: * The multiline history feature is removed. nodejs#24804 * tls: * Added min/max protocol version options. nodejs#24405 * The X.509 public key info now includes the RSA bit size and the elliptic curve. nodejs#24358 * url: * `pathToFileURL()` now supports LF, CR and TAB. nodejs#23720 * Windows: * Tools are not installed using Boxstarter anymore. nodejs#24677 * The install-tools scripts or now included in the dist. nodejs#24233 * Added new collaborator: * [antsmartian](https://github.com/antsmartian) - Anto Aravinth. nodejs#24655 PR-URL: nodejs#24854
Refs: nodejs#24739 Fixes: nodejs#24730 PR-URL: nodejs#24870 Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com> Reviewed-By: Colin Ihrig <cjihrig@gmail.com> Reviewed-By: Fedor Indutny <fedor.indutny@gmail.com> Reviewed-By: Trivikram Kamat <trivikr.dev@gmail.com>
Add a
--http-parser=llhttp
vs--http-parser=legacy
command line switch, to make testing and comparing the new
llhttp-based implementation easier.
Refs: #24730
Checklist
make -j4 test
(UNIX), orvcbuild test
(Windows) passes