stderr is piped to stdout

[erikeah]

How to replicate error:

This ain file should output connection headers to stderr and print a json to stdout.

[Host]
http://localhost:3000/places

[Headers]
Content-Type: application/json

[Backend]
curl

[BackendOptions]
-sS -D /dev/stderr

But what actually happen is that stderr (the headers) are printed on stdout.

You can check this by running this line and reviewing the content of headers.txt file:

ain the_code_provided.ain 2>headers.txt

Why I consider that this should not be like this?

1. Because as user I expect to be printed to stderr
2. Because stderr could be used for logging and will allow to do this with the provided example :

   ain the_code_provided.ain | jq

   And don't make crash the jq parser.

Thanks in advanced!

[jonaslu]

Thanks for reporting! I'll have a think of the impact of changing the behaviour (which was a conscious decision to capture all output in the order the backend prints it): https://github.com/jonaslu/ain/blob/main/internal/pkg/call/curl.go#L89

In the meantime, you can work around this by doing:

Host]
http://localhost:3000/places

[Headers]
Content-Type: application/json

[Backend]
curl

[BackendOptions]
-sS -D /proc/${PID}/fd/2

PID=$$ ain the_code_provided.ain

That will print headers to the shells stderr and output to stdout

[erikeah]

Thanks a lot for quick response! Once you have decided what to do let me know and may I can submit PR.

[jonaslu]

Hi again!

I agree with this. Output from the backend can even intermingle, so separating them solves that too: https://unix.stackexchange.com/questions/476080/what-prevents-stdout-stderr-from-interleaving

For the delay: I'm in the middle of a major overhaul of the internals and wanted to see what impact it had, but it looks like doing this won't collide that much so go ahead and do the PR.

[erikeah]

Great! Then once I have the code, I will open PR!

[erikeah]

Hi Jonas, today I had a bit of time and I did the task as you can see in my fork, however I find that I have introduced a lot of boiler plate on the runAsCmd functions.

What do you think if I create a new function share for all backends to run the command and contain the boiler plate?

Pseudo-code based on my fork (internal/pkg/call/call.go):

func CallBackend(ctx context.Context, callData *data.Call, leaveTmpFile, printCommand bool) (Output, error) {
       ...

    if printCommand {
        if command, err := backend.getAsString(); err != nil {
            return Output{StdErr: []byte(command)}, err
        } else {
            return Output{StdOut: []byte(command)}, nil
        }
    }

    cmd, err := backend.GetCmd(backendTimeoutContext) // May returns the exec.CommandContext
        output, err := runCmd(cmd)

        ...
    return output, nil
}

[jonaslu]

Hard to tell, better if you do it as a commit on top of your fork and we can look at it then. There's always git reset.

[erikeah]

I will give a try this weekend then.

[jonaslu]

Fixed in commit eb9fa6c, closing this