When writing programs which utilize network, you can see quite noticeable slowdown of compilation with CGO_ENABLED=0
.
For example, the simplest HTTP server:
package main
import (
"flag"
"fmt"
"log"
"net/http"
)
func handler(w http.ResponseWriter, r *http.Request) {
fmt.Fprintf(w, "Hi! glad you requested %s.\n", r.URL.Path[1:])
}
func main() {
port := flag.Int("port", 9000, "")
flag.Parse()
http.HandleFunc("/", handler)
err := http.ListenAndServe(fmt.Sprintf(":%d", *port), nil)
if err != nil {
log.Fatal(err)
}
}
the timings are:
% time go build
go build 0.46s user 0.06s system 131% cpu 0.396 total
% time CGO_ENABLED=0 go build
CGO_ENABLED=0 go build 3.93s user 0.15s system 143% cpu 2.849 total
So far I'm not using bindings to C so CGo handling or not seems irrelevant, what I'd like to do is to compile 100% static binary, but not if there's such slowdown.
What is the cause of such behavior?
The problem is that the standard library packages are built without flags. CGO_ENABLED
changes build flags and therefore it can't use the pre-built packages so most of the standard library needs to be rebuilt.
As the other answer mentioned, go build -i
will install the packages built with the new flags, but that won't really solve much because if you install packages built with CGO_ENABLED=0
, it will speed up all future builds with CGO_ENABLED=0
, but it will slow down all the builds without it.
Unfortunately the way pre-built packages are installed by default today is pretty inefficient because everything goes into the same directory under the same names regardless of how it's built. If you want to be able to have fast builds of go programs with different flags, besides doing go build -i
you also need to use the the -installsuffix
and/or the -pkgdir
flags. In the system I work in, we have a handful of different compilation modes, each mode has different flags (because of a lot of old C code we interface with) and each mode also has its own -pkgdir
.