Good work! I've tagged this as semver-minor because it adds a new feature to http as well.

Very nice! Thank you for taking this on

It actually looks like this already needs a rebase? :/

@addaleax Yep, I will be doing the rebase in a bit. Note that the commits as of right now represent maybe 50% of the commits that will be part of this PR... more to come.

Update: I've backported commits that have to do with HTTP/2 all the way up to August 15. Based on how many come afterward I'm thinking of just opening a new PR for remaining commits.

I created list of commits based on any commit the all matched the following criteria:

• Has not been backported to 8.x before, and associated PR does not have the do-not-backport-v8.x label
• Is on the branch v10.x-staging (as of August 15)
• Satisfies at least one of the following:
  • Touches files that match at least one of these globs: lib/internal/http2, doc/api/http2.md, test/**/test-http2-*, src/node_http2*
  • Has http2 in the commit name
  • Is from a PR with the http2 label

I removed commits that I believed were too broad to be applied to a PR based around just backporting http2, which resulted in (1) the commits in the PR, and (2) a smaller list of commits that I did not add to this PR but are worth noting. I grouped these commits below based on what I think should be backported together in follow-up PRs:

• This seems to be part of a "StreamBase refactor" which I wasn't sure whether it belonged in v8.x
  • e85c20b net,http2: merge write error handling & property names (net,http2: merge write error handling & property names #19734)
• Changes to errors.js seem independent of H2 changes
  • d28211e errors: validate input arguments (errors: validate input arguments #19924)
  • b6fbe16 errors: alter ERR_HTTP2_INVALID_CONNECTION_HEADERS (errors: alter ERR_HTTP2_INVALID_CONNECTION_HEADERS #19807)
  • f086354 errors: alter ERR_HTTP2_PSEUDOHEADER_NOT_ALLOWED (errors: alter ERR_HTTP2_PSEUDOHEADER_NOT_ALLOWED #19958)
• Runtime-Controlled Debugging System -- this seems fairly independent despite the change being mostly in HTTP2
  • 182c73b http2: switch to new runtime-controlled debugging system (src,http2: refactor + generalize http2 debugging #20987)
• These are a dependent chain of commits that didn't seem to merge cleanly
  • 0b3c80c http2: fix issues with aborted respondWithFile()s (http2: fix issues with aborted respondWithFile()s #21561)
  • eea199b test: fix http2 connection abort test (test,doc: address http2 connection abort/network error handling #21861)
  • c3d9000 doc: document http2 network error behaviour (test,doc: address http2 connection abort/network error handling #21861)
  • 793a5bb test: don't fail http2 abort test if 'data' is called multiple times (test: don't fail http2 abort test if 'data' is called multiple times #21925)

@addaleax, it seems like you are the author of the majority of these changes, so any advice you might have on how to proceed with these would be appreciated.

Ping @MylesBorins @addaleax @nodejs/http2 -- Since Node 10 will be going to LTS soon, I want to make sure we have enough time to react to any requested changes/future PRs.

@kjin thanks for the hard work! We should plan for another 8.x semver minor to accommodate these changes. It looks like there are semver minor changes outside of http2, but if they scoped only to http2 we would be able to land as a semver-patch (as it is not stable in 8.x)

@MylesBorins With the exception of adding the ERR_TLS_RENEGOTIATE error, do you know what else would count as semver-minor that is outside of the H2 domain?

From a very quick look I'm thinking about #20094 which is a semver-minor which affects http.

TBH it probably doesn't make sense to just land those changes on http, we should eat the semver-minor on this

CI: https://ci.nodejs.org/job/node-test-pull-request/17592/

Update #2: I've added commits that were landed between 10.9 and 10.11. There were some in particular that I opted to change/skip, that might need additional attention:

• d3ceaa1 (http2: emit timeout on compat request and response #22252): I changed tests related to timeouts. More info in http2: emit timeout on compat request and response #22252 (comment).
• 2a88731 (lib: merge onread handlers for http2 streams & net.Socket #22449): This PR has a lot of non-http2 stuff going on.
• 24675a4 (http2: add origin frame support #22956): This change makes use of a V8 API that's not available in Node 8. I use the closest approximation of the API (StringOneByte overload) instead. (The corresponding commit in this PR has a similar note.)
• d1826fe (http2: close fd in doSendFileFD() #23047) This PR appears to fix an issue which might be the result of a PR that wasn't backported earlier (src,http2: use native-layer pipe mechanism from FileHandle instead of synchronous I/O #18936 -- contains do-not-backport-8.x tag).

I believe as it is, this PR passes tests. However, there are a few things I want to get confirmation for from these folks:

• @apapirovski http2: fix responses to long payload reqs #20084 (comment) (talked in person, this seems OK)
• @jasnell http2: emit timeout on compat request and response #22252 (comment) as well as -- does the third note in the list above make sense to you? (Edit -- I looked and it seems safe to me, the Isolate being used should be the same in both implementations)
• @addaleax [8.x] Backport http2 changes from 10.x #22850 (comment) as well as -- Does it make sense for me not to backport the second commit listed above?

@kjin It’s okay not to backport 2a88731, yes

Ping @nodejs/lts -- I think this is ready to move forward (let me know if you need anything from my end)

CI: https://ci.nodejs.org/job/node-test-pull-request/17883/

The linter is failing, it seems to be this commit that is failing:

commit 019e127df96e2a53fa5830446a7c206a4eb3d4e1
Author: James M Snell <jasnell@gmail.com>
Date:   Sun Sep 16 19:13:11 2018 -0700

    http2: add origin frame support
    
    v8.x Backport Note -- as V8 doesn't expose an overload of String::WriteOneByte
    in Node 8 that accepts an isolate argument, the Origins constructor has been
    changed to not accept an isolate.
    
    PR-URL: https://github.com/nodejs/node/pull/22956
    Reviewed-By: Matteo Collina <matteo.collina@gmail.com>

Landed on v8.x-staging in fe3836a...18a2b3d

Not sure if this is the right place to post this; please redirect me if needed.
I used 8.13.0 today and it seems that it behaves different compared to 10.12.0 regarding emitting error events.
e.g. if I pass a non existing file to respondWithFile() the error event is issued after the close event for 8.x; for 10.x error is emitted first.

@Flarna if you could open a new issue that documents how to recreate the observed behavior we can get to this ASAP

@MylesBorins done: #24559