Comparison

golangci-lint vs gometalinter

GolangCI-Lint was created to fix the following issues with gometalinter:

1. Slow work: gometalinter usually works for minutes in average projects. GolangCI-Lint works 2-7x times faster by reusing work.
2. Huge memory consumption: parallel linters don't share the same program representation and can consume n times more memory (n - concurrency). GolangCI-Lint fixes it by sharing representation and consumes 26% less memory.
3. Doesn't use real bounded concurrency: if you set it to n it can take up to n*n threads because of forced threads in specific linters. gometalinter can't do anything about it because it runs linters as black boxes in forked processes. In GolangCI-Lint we run all linters in one process and completely control them. Configured concurrency will be correctly bounded. This issue is important because you often want to set concurrency to the CPUs count minus one to ensure you do not freeze your PC and be able to work on it while analyzing code.
4. Lack of nice output. We like how the gcc and clang compilers format their warnings: using colors, printing warning lines and showing the position in line.
5. Too many issues. GolangCI-Lint cuts a lot of issues by using default exclude list of common false-positives. By default, it has enabled smart issues processing: merge multiple issues for one line, merge issues with the same text or from the same linter. All of these smart processors can be configured by the user.
6. Integration into large codebases. A good way to start using linters in a large project is not to fix a plethora of existing issues, but to set up CI and fix only issues in new commits. You can use revgrep for it, but it's yet another utility to install and configure. With golangci-lint it's much easier: revgrep is already built into golangci-lint and you can use it with one option (-n, --new or --new-from-rev).
7. Installation. With gometalinter, you need to run a linters installation step. It's easy to forget this step and end up with stale linters. It also complicates CI setup. GolangCI-Lint requires no installation of linters.
8. Yaml or toml config. Gometalinter's JSON isn't convenient for config files.

golangci-lint vs Running Linters Manually

1. It will be much slower because golangci-lint runs all linters in parallel and shares 50-80% of linters work.
2. It will have less control and more false-positives: some linters can't be properly configured without hacks.
3. It will take more time because of different usages and need of tracking of versions of n linters.

Performance

Benchmarks were executed on MacBook Pro (Retina, 13-inch, Late 2013), 2,4 GHz Intel Core i5, 8 GB 1600 MHz DDR3. It has 4 cores and concurrent linting as a default consuming all cores. Benchmark was run (and measured) automatically, see the code here (BenchmarkWithGometalinter).

We measure peak memory usage (RSS) by tracking of processes RSS every 5 ms.

Comparison with gometalinter

We compare golangci-lint and gometalinter in default mode, but explicitly enable all linters because of small differences in the default configuration.

Copy
$ golangci-lint run --no-config --issues-exit-code=0 --timeout=30m \
  --disable-all --enable=deadcode  --enable=gocyclo --enable=golint --enable=varcheck \
  --enable=structcheck --enable=maligned --enable=errcheck --enable=dupl --enable=ineffassign \
  --enable=interfacer --enable=unconvert --enable=goconst --enable=gosec --enable=megacheck
$ gometalinter --deadline=30m --vendor --cyclo-over=30 --dupl-threshold=150 \
  --exclude=<default golangci-lint excludes> --skip=testdata --skip=builtin \
  --disable-all --enable=deadcode  --enable=gocyclo --enable=golint --enable=varcheck \
  --enable=structcheck --enable=maligned --enable=errcheck --enable=dupl --enable=ineffassign \
  --enable=interfacer --enable=unconvert --enable=goconst --enable=gosec --enable=megacheck
  ./...

On average golangci-lint is 4.6 times faster than gometalinter. Maximum difference is in the self-repo: 7.5 times faster, minimum difference is in terraform source code repo: 1.8 times faster.

On average golangci-lint consumes 26% less memory.