"typecheck" takes ~5 times longer than "go build"

[alecthomas]

$ time golangci-lint run --disable-all --enable typecheck
Congrats! No issues were found.
golangci-lint run --disable-all --enable typecheck  4.61s user 1.51s system 391% cpu 1.562 total
$ time go build .
go build .  0.72s user 0.50s system 139% cpu 0.873 total

[golangci]

hi!

1. run with -v please
2. can you compare it with another linters? gometalinter, errcheck? It should not be slower.

Typecheck works fast, but program loader is slow. I have plans to improve its speed by caching and reusing compilation cache.

[alecthomas]

$ time golangci-lint -v run --disable-all --enable typecheck
INFO[0000] Concurrency: 8, machine cpus count: 8
INFO[0000] Active linters: [typecheck]
INFO[0000] set GOROOT="/usr/local/Cellar/go/1.10.2/libexec"
INFO[0000] Found paths for analysis for 201.447µs: [.]
INFO[0001] Program loading took 1.462691677s
INFO[0001] worker.6 took 4.96µs with stages: typecheck: 803ns
INFO[0001] worker.5 took 2.21µs
INFO[0001] worker.4 took 913ns
INFO[0001] worker.3 took 945ns
INFO[0001] worker.7 took 10.953µs
INFO[0001] worker.2 took 9.82µs
INFO[0001] worker.8 took 12.754µs
INFO[0001] worker.1 took 47.418µs
INFO[0001] Workers idle times: #2: 76.443µs, #3: 85.506µs, #4: 132.962µs, #5: 146.895µs, #6: 154.698µs, #7: 83.501µs, #8: 48.761µs
INFO[0001] processing took 15.629µs with stages: max_same_issues: 14.396µs, path_prettifier: 331ns, max_from_linter: 204ns, exclude: 200ns, cgo: 173ns, nolint: 130ns, max_per_file_from_linter: 116ns, uniq_by_line: 42ns, diff: 37ns
Congrats! No issues were found.
INFO[0001] Extracting issued lines took 0s
INFO[0001] Memory: 15 samples, avg is 188.9MB, max is 244.5MB
INFO[0001] Execution took 1.486011496s
golangci-lint -v run --disable-all --enable typecheck  4.48s user 1.54s system 378% cpu 1.591 total

Yes, they are all slow. I'm guessing go build is fast because of the new build cache. It'd be great if golangci-lint could utilise the cache.

[jirfag]

While investigating #87 (comment) I found that go build cache can be used to speed up loading. But it's not compatible with interfacer and megacheck: they build SSA representation and therefore need information about all functions of all dependencies. This information isn't stored in build cache.
Megacheck and interfacer are enabled at 80% of cases, it doesn't make sense to optimize 20%.

More perspective is to make incremental program loading to speedup 100% of not --fast analyzes. It will be done by #92.

[alecthomas]

Makes sense, thanks 👍