"Improve" root import filter for universal members

[som-snytt]

Instead of elaborate import selector machinery,
just filter at ImportInfo.

Follow-up to #8541 with JZ's clever suggestion.

He really doesn't like it when you mess up his benchmarks.

[som-snytt]

i.e. isUniversalMember, but using this method for uniformity with the main case above. allImportedSymbols is used for completions.

[som-snytt]

As we say in testing, "I must be doing something wrong."

I ran some tests during lunch.

I expected to see a number, then a number that got bigger, and then another number that got smaller.

Yes, I read the part where: "Do not assume the numbers tell you what you want them to tell."

[retronym]

I am seeing a speedup:

[info] HotScalacBenchmark.compile                     ../corpus          a8c43dc                    false  2.13.2-bin-f80f5b8-SNAPSHOT    scalap  sample  330   981.146 ± 1.920  ms/op                           z
[info] HotScalacBenchmark.compile                     ../corpus           latest                    false  2.13.2-bin-3d0d203-SNAPSHOT    scalap  sample  330   962.412 ± 3.381  ms/op

Your numbers look suspiciously slow (in the base line and in the patched version). Maybe there is another factor dominating the the performance? What hardware + OS did you use?

[som-snytt]

I'll get more experience with it and also not do any gaming while running tests (kidding).

Thanks again for the code hint.

[retronym]

All good with the benchmark run: https://scala-ci.typesafe.com/grafana/dashboard/db/scala-benchmark?orgId=1&from=1564683691856&to=1574312618704&var-branch=All&var-source=All&var-bench=HotScalacBenchmark.compile&var-host=scalabench%40scalabench%40

[som-snytt]

Running benchmark on home Dell i7 with SSD, Ubuntu subsystem under windows, instead of office Dell xeon Ubuntu with fidget spinner disks yields smaller numbers. Smaller is better, right?

[retronym]

Yep, smaller is better.