given that __attribute__((constructor)) is 1. a gcc compiler extension 2. only run when it is loaded, I don't see why it should run before main in go

@seankhliao I don't think the problem is with __attribute__((constructor)) not running before Go main (or C main). Here's a slightly different code that illustrates the problem better. Notice that Go reports the new value forEXISTING_VAR but old value for NEW_VAR while C/libc reports new values for both.

package main
/*
#cgo CFLAGS: -Werror -I/usr/include
#cgo LDFLAGS: -L/usr/lib64/ -L/usr/lib/x86_64-linux-gnu/ -ldl
#include <stdio.h>
#include <stdlib.h>
__attribute__((constructor)) void setup_env(void) {
	setenv("EXISTING_VAR", "newValue", 1);
	setenv("NEW_VAR", "newValue", 1);
}
void print_env(void) {
	printf("C> EXISTING_VAR='%s'\n", getenv("EXISTING_VAR"));
	printf("C> NEW_VAR='%s'\n", getenv("NEW_VAR"));
}
*/
import "C"
import (
	"fmt"
	"os"
)
func main() {
	fmt.Printf("GO> EXISTING_VAR='%v'\n", os.Getenv("EXISTING_VAR"))
	fmt.Printf("GO> NEW_VAR='%v'\n", os.Getenv("NEW_VAR"))
	C.print_env()
}

Here’s how to run the code to repro the issue. Notice the missing value for NEW_VAR when accessed via Go’s os package.

$ EXISTING_VAR="oldValue" ./main
GO> EXISTING_VAR='newValue'
GO> NEW_VAR=''
C> EXISTING_VAR='newValue'
C> NEW_VAR='newValue'

I see, I think it's related to #27693 but not sure if it's the same issue

I agree, this is similar to #27693. The only difference is that in #27693, C.setenv after Go runtime is setup while in this case it is being called before Go runtime is setup. It is ok to not support the former because C.setenv isn't thread-safe.

Looking at the Go runtime setup code, it gets the environment from the arguments to C main. If instead Go runtime setup code get the environment from char **environ in libc, it should see the changes made to the environment by any preceding C code.

I think it's OK for us to just say that the behavior here is implementation defined. Especially when we start considering the behavior with -buildmode=c-archive, there is no perfect solution here.

For what it's worth, this is what I see for the first program:

GO> EXISTING_VAR=''
GO> NEW_VAR=''
C> PATH='(null)'
C> EXISTING_VAR='newValue'
C> NEW_VAR='newValue'

Thanks @ianlancetaylor.
Yes, I see the same output as you from the first program.

Closing as working as intended