feature: ws in core

[eljefedelrodeodeljefe]

ws in core

I know this has been discussed before several times, but as far as I know, not with a PR or on a very technical, code-level basis. As I am about to make a rather big commitment to ws as technology, I was reviewing it and concluded this should better be under the core teams umbrella, even if it might be a longer road.

TL;DR

websockets are the primary reason for companies, which have not decided to make a commitment towards node to use node anyways. For a wider adoption of the language this is a critical point.

In order to gain momentum as a language, the experience then should be as performant, secure and well documented as possible. IMO, userland development cannot have all of those, as much as core could. Especially issues like Buffer(number), however dangerous they are, can only be effectively secured via backporting code through the language's LTS.

Even though I am a big small core supporter, this seems necessary to me. Small core could be held up by extracting a lot of boilerplate, option parsing and interfaces out of ws though. There is definitely room to improve.

Conclusion

My personal commitment on this was just time. If this doesn't lead anywhere, I am not too opinionated.

---

Development Efforts so far

• do options parsing natively (extract dependency)
• included Ultron for buffer weakness(?) (needs extraction of Ultron)
• no cpp dependency
• lots of style changes
• const requires, let local variable copies
• ported file and class structure for core
• ported tests to run without dependencies -> disabled 1/4 of them in order to run
• subtree merge

Known Issues

• some tests (<10) are not running
• a multitude of tests is not running in parallel after extraction of mocha
• some tests will fail because of not being able to require('internal/${dep}')
• linting will fail badly
• code can be simplified a lot.
• doc not not adapted yet.
• licensing needs to be checked and adapted.
• the whole NativeModule integration is a little bit shaky, atm, since I couldn't run the tests completely.
• Tests need to be run with NODE_TLS_REJECT_UNAUTHORIZED=0, therefore make test will fail automatically.
• benchmark: untouched

Stats

sloc <dir> | <file>

lib/internal/websockets
---------- Result ------------

            Physical :  2162
              Source :  1755
             Comment :  201
 Single-line comment :  62
       Block comment :  139
               Mixed :  4
               Empty :  210

Number of files read :  8

lib/websockets
---------- Result ------------

            Physical :  925
              Source :  714
             Comment :  72
 Single-line comment :  66
       Block comment :  6
               Mixed :  4
               Empty :  143

Number of files read :  1

[eljefedelrodeodeljefe]

After reviewing this, this would out-of-box introduce a lot of liability. I'll defer this to further research, namely prototyping a minimal ws implementation. Since I still find this to be be extremely useful, I'll keep nodejs/node-eps/pull/6 open.