[WIP] Add auto-complete suggestions

[pete-murphy]

trafficstars

Description of the change

Fixes #32.

Clearly and concisely describe the purpose of the pull request. If this PR relates to an existing issue or change proposal, please link to it. Include any other background context that would help reviewers understand the motivation for this PR.

---

Checklist:

• [ ] Added the change to the changelog's "Unreleased" section with a reference to this PR (e.g. "- Made a change (#0 by @)")
• [ ] Linked any existing issues or proposals that this pull request should close
• [ ] Updated or added relevant documentation
• [ ] Added a test for the contribution (if applicable)

[thomashoneyman]

@nwolverson -- I'm curious if you have any thoughts on this from the language server side of things; I'm not really sure what is and isn't possible with psc-ide to make this work.

[nwolverson]

As per discussion on discord I'd suggest avoiding proxying the purs ide requests directly simply for the reason that it likely allows read/write access to the entire filesystem as the current user, and probably allows more significant DoS potential than a more directed endpoint. However returning an unparsed blob of json to be parsed on the client side seems like a legitimate option if it avoids re-encoding results & can reuse code.

It is surprising that the request would take 2s, it would seem the only extra overhead is an HTTP request, you are spawning the purs binary from the shell as well as the Haskell code. Are you actually looking at the same request in the same workspace? I see groupReexports and maxResults on the left, I'd also prefer to see a better timing (eg just directly using time). Certainly making the socket request that purs ide client makes directly could be faster but that's something to shave off milliseconds I'd assume.

Happy to discuss the details of the completion results themselves later - but that request looks sane at least

[pete-murphy]

This is surprising to me—these are the results from this commit https://github.com/purescript/trypurescript/pull/270/commits/0bdd1fd65505807b0f8fd3dabf5048aac5a1bec2 (which is back to being a GET request, encoding the JSON on the server)

❯ hyperfine --warmup 1 "curl -Gs --data-urlencode 'q=app' localhost:8081/complete"
Benchmark #1: curl -Gs --data-urlencode 'q=app' localhost:8081/complete
  Time (mean ± σ):     173.3 ms ±  10.2 ms    [User: 7.3 ms, System: 11.3 ms]
  Range (min … max):   155.2 ms … 188.8 ms    16 runs

so, pretty snappy, but removing this line https://github.com/purescript/trypurescript/blob/0bdd1fd65505807b0f8fd3dabf5048aac5a1bec2/server/Main.hs#L191 seems to result in the 2s slowdown I was seeing before

❯ hyperfine --warmup 1 "curl -Gs --data-urlencode 'q=app' localhost:8081/complete"
Benchmark #1: curl -Gs --data-urlencode 'q=app' localhost:8081/complete
  Time (mean ± σ):      2.339 s ±  0.118 s    [User: 5.0 ms, System: 8.3 ms]
  Range (min … max):    2.209 s …  2.590 s    10 runs

I added that line because the previous iteration with the POST request was hanging and that seemed to fix it, but I don't fully understand the behavior there.

Other results for completeness :)

From 7496700 POST request (not much difference)

Benchmark #1: curl -s 'http://localhost:8081/complete' -X POST --data-binary '{ "command": "complete", "params": { "filters": [{ "filter": "prefix", "params": { "search": "app" } }] } }' -H 'content-type: application/json'
  Time (mean ± σ):     117.7 ms ±  10.5 ms    [User: 4.7 ms, System: 7.0 ms]
  Range (min … max):   100.6 ms … 137.3 ms    21 runs

Running purs ide client directly

Benchmark #1: echo '{ "command": "complete", "params": { "filters": [{ "filter": "prefix", "params": { "search": "app" } }] } }' | purs ide client -p 8082
  Time (mean ± σ):      69.8 ms ±   3.2 ms    [User: 7.3 ms, System: 16.2 ms]
  Range (min … max):    65.1 ms …  81.3 ms    33 runs