os/exec: automatic encoding conversion for stdout/stderr on Windows

[lime2008]

Proposal: os/exec: Handle Windows Standard Streams Encoding

Currently, the os/exec package does not differentiate between the default behavior of different Windows versions regarding standard output and error encoding. This can lead to encoding issues when running commands that output non-UTF-8 characters in Windows with the "Use Unicode UTF-8 for worldwide language support" beta feature enabled.

Problem:

• Windows traditionally uses locale-specific encodings for console output (e.g., gbk in China).
• Windows introduced a beta feature to enforce UTF-8 encoding for all console applications.
• os/exec does not account for this difference while getting stdout, potentially leading to garbled output when this beta feature is not enabled.

Proposed Solution:

Introduce a mechanism in os/exec to handle Windows console encoding variations, specifically:

1. Detection: Automatically detect if the Windows UTF-8 beta feature is active.
2. Transparent Handling: Based on the chosen configuration, seamlessly handle encoding and decoding of output streams within os/exec.

Benefits:

• Code can be more consistent by removing the need for platform-specific encoding workarounds.
• Improved compatibility with applications relying on UTF-8 output.
• Enhanced user experience by avoiding garbled characters in console output.

This feature request aims to improve the reliability and user-friendliness of os/exec when interacting with console applications in diverse Windows environments. Thank you to everyone who reviews this request.

[qmuntal]

Thanks for reporting this issue @lime2008. I'm moving this out of the proposal process given that you are suggesting that it can be fixed without adding new APIs. Let's treat this a bug fix rather than a proposal for now.

Could you provide some more detailed steps for reproducing this issue on my PC?

[lime2008]

Below is a script that demonstrates the problem:

package main

import (
    "fmt"
    "log"
    "os/exec"
)

func main() {
    cmdRunner := exec.Command("cmd", "/C", "echo 中文字符测试 any chinese character test")
    output, err := cmdRunner.Output()
    if err != nil {
        log.Printf("Error executing command: %s", err)
        output = []byte(fmt.Sprintf("Error: %s", err.Error()))
    }
    //utf8Result, err := EnsureUTF8(output)
    utf8Result := output
    fmt.Print(string(utf8Result))

}

and it will make the output like:

[lime2008]

after enabling the settings(sorry my system interface is in Chinese.)
it's in Settings -> Time & language -> Language & region

and then it will output normal characters like

Thanks for the review.

[seankhliao]

I'm concerned that os/exec currently treats output as a stream of bytes, not necessarily text that has an encoding that needs to be translated. What happens when transparent handling tries to convert binary data?

[lime2008]

How about add an interface to explicitly convert []byte to str based on the system character set? Or give a param to decide convert or not
Consider in python there is also a mode r to auto decode using system config and rb for raw bytes.