Performance

Memory Usage

A trade-off between memory usage and execution time can be controlled by GOGC environment variable. Less GOGC values trigger garbage collection more frequently and golangci-lint consumes less memory and more CPU. Below is the trade-off table for running on this repo:

Why golangci-lint is so fast

1. Work sharing The key difference with gometalinter is that golangci-lint shares work between specific linters (golint, govet, ...). We don't fork to call specific linter but use its API. For small and medium projects 50-90% of work between linters can be reused.

   • load []*packages.Package by go/packages once

     We load program (parsing all files and type-checking) only once for all linters. For the most of linters it's the most heavy operation: it takes 5 seconds on 8 kLoC repo and 11 seconds on $GOROOT/src.

   • build ssa.Program once

     Some linters (megacheck, interfacer, unparam) work on SSA representation. Building of this representation takes 1.5 seconds on 8 kLoC repo and 6 seconds on $GOROOT/src.

   • parse source code and build AST once

     Parsing one source file takes 200 us on average. Parsing of all files in $GOROOT/src takes 2 seconds. Currently we parse each file more than once because it's not the bottleneck. But we already save a lot of extra parsing. We're planning to parse each file only once.

   • walk files and directories once

     It takes 300-1000 ms for $GOROOT/src.

2. Smart linters scheduling

   We schedule linters by a special algorithm which takes estimated execution time into account. It allows to save 10-30% of time when one of heavy linters (megacheck etc) is enabled.

3. Don't fork to run shell commands

All linters has their version fixed with go modules, they are builtin and you don't need to install them separately.