tls: remove SLAB_BUFFER_SIZE

[apapirovski]

This constant has not been in use for many years now and the test alongside it is invalid, as well as flaky.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• commit message follows commit guidelines

[apapirovski]

CI: https://ci.nodejs.org/job/node-test-pull-request/15326/

[addaleax]

Would feel a teeny tiny bit more comfortable if the export was removed in a semver-major PR, but at the same time I find it hard to come up with a (once) valid use case that could actually be broken by this

[apapirovski]

@addaleax I'm fine with it being semver-major if that's what everyone decides. This test very infrequently fails with ECONNRESET (since it destroys the socket at a weird time) so I just want it gone from master... 😄

[addaleax]

@apapirovski If the test is flaky, we probably want it gone in older branches as well, right?

[apapirovski]

@addaleax I mean, it doesn't fail so often that it's a problem. I've seen it twice in six months. We don't have a thread for it. But yeah, would obviously be nice... we could prob backport just the test removal if this ends up semver-major.

(It's just that there are so many more test runs on master.)

[joyeecheung]

Should we mention it in deprecations.md as an EOL? It is not impossible that someone may expect this to be a number instead of undefined.

[lpinca]

@apapirovski

This test very infrequently fails with ECONNRESET (since it destroys the socket at a weird time)

Can you please elaborate a bit? I don't understand why. The socket is destroyed when the 'response' event is emitted. At this point no more data is written to the socket by any of the peers so I'm not sure how ECONNRESET could be triggered.

[apapirovski]

@lpinca I'm pretty sure the response event can trigger before the full \r\n\r\n bit makes it through, no? In general we tend to assume in our tests that chunks will make it through exactly as they're sent but it's not always true when the OS is busy.

[lpinca]

@apapirovski you sure? afaik the 'response' event is emitted only after receiving \r\n\r\n?

[apapirovski]

@lpinca ok, just checked and you're right on that point. I think there could still be final bits making it through though since we occasionally have empty writes in TLS for state maintenance? I would need to look in more detail but IMO something like that might be going on here.

I'll do some packet inspection and see exactly what's being sent before this lands, to make sure there's not a real bug hiding.

[lpinca]

@apapirovski yes I was actually wondering if this was caused by a deeper bug. Thanks.
Anyway changes look good to me.

[addaleax]

CI: https://ci.nodejs.org/job/node-test-pull-request/15936/

[maclover7]

New CI since old one 404s now: https://ci.nodejs.org/job/node-test-pull-request/16078/

[targos]

Resumed build: https://ci.nodejs.org/job/node-test-pull-request/16189/

[maclover7]

@apapirovski @addaleax I believe this should be ready to land, right? Just needs another CI run (since the old ones keep becoming stale)?

[jasnell]

CI: https://ci.nodejs.org/job/node-test-pull-request/16384/

[jasnell]

Landed in 0aae34f