Usage with tracing

[anewton1998]

Based on #66 , there is an apparent issue with using the tracing and tracing-subscriber crates:

BTW The tracing call was blocking because rdap-cli sends data to minus before minus initializes the interface but after it has acquired the Stdout lock. So tracing just waits until stdout becomes free again (which it never gets).

To get tracing-subscriber to work without blocking, I had to use with_writer and pass in a std::io::Write trait object, such as this:

#[derive(Clone)]
pub(crate) struct PageWriter(pub(crate) Pager);

impl std::io::Write for PageWriter {
    fn write(&mut self, buf: &[u8]) -> std::io::Result<usize> {
        self.0
            .push_str(String::from_utf8_lossy(buf))
            .map_err(|e| std::io::Error::new(ErrorKind::Other, e))?;
        Ok(buf.len())
    }

    fn flush(&mut self) -> std::io::Result<()> {
        Ok(())
    }
}

Is there a way to avoid sending data to minus before initialization? Should a tokio task issue a yield_now() as the first thing it does?

[AMythicDev]

The fix that I can guess right now is to move the code that locks the Stdout to just above where minus initializes the interface.
But I want to know the rationale behind this... Like you won't get any logs printed out once the pager takes over. Those logs won't be printed even after the pager quits and to my guess those tracing logs will probably block the pager.
The best approach would be to use with_writer(). I don't think it is in any way a bad approach. But rather than manually creating a struct and implementing Write, you can use tracing-appender to log everything to a file.

The reason I encountered the same issue even after setting up tracing-appender was because I was setting up tracing just before where the pager starts but after Stdout has been locked. So the logs that occurred within the time period between locking of Stdout and setting up of tracing using tracing-appender caused the the halt.

EDIT:- Another option that is related to sending data only after the pager has been initialized is to use hooks(#95) which are planned. I can't give you any pointers right now as I haven't started working on it but once I get them landed you can hook a function to send data to the pager when the ui has been loaded.

[anewton1998]

Perhaps the tracing use case is causing confusion. But I had to do the same thing for termimad (a markdown renderer). Fortunately, it has an API where I can pass it a std::fmt::Write trait object. However, what would happen if a library crate doesn't have that, and just assume writing to stdout.

If moving the current logic causes backward compatibility issues, then I wouldn't want that. And I can live with the solutions I have now. But I was wondering if there was something my code could be doing to avoid the Write trait objects.

[AMythicDev]

Well I guess the only solution that I see is to move the code that locks stdout to where we draw the initial UI.

But I wanna know that if a program that only prints to stdout print after the UI is drawn (with Stdout locked) the program will cause everything to freeze just like what happened with tracing.

Even if we decide to the program's output via minus, how can we get the exact data that it sent to the stdout without having some sort of API.

You had to implement std::io::Write on a custom wrapper that wrapped the Pager struct in tracing's case but not in terminad's situation because Pager implements std::fmt::Write and not std::io::Write. Although it's quite possible to implement both but it may cause name confusion for some users (Wanna know your opinion on that). Lastly I think whatever the code sample that you posted above is absolutely OK in Rust; I have seen tons of cases where this pattern is heavily used. So this also adds on whether to add std::io::Write on Pager.

[anewton1998]

Termimad has two methods, write_text and write_text_on. I am using write_text_on, which takes a Write object because write_text will block trying to use stdout. I'll see about writing some demo code if that would be helpful.

[AMythicDev]

I didn't get what exactly you want. You didn't answer the questions that I raised in my previous comment.

[anewton1998]

Even if we decide to the program's output via minus, how can we get the exact data that it sent to the stdout without having some sort of API.

My apologies. I think you are right... there is no way to do this without an API. I don't know why, but I had it in my head that you already had some bit of magic somewhere that intercepted stdout. Sorry for the confusion and for wasting your time. Thanks for the discussion, and thanks for all the work you have done on this code.

[AMythicDev]

Ah! no thanks buddy.🙂. I am closing the discussion as I hope that you understand the problems with implementing what you want. Feel free to ask questions whenever you like.