Prompt is printed before command output
TatriX opened this issue · comments
Welcome to Crush
Type "help" for... help.
crush# ls
crush# file
.git .gitignore Cargo.lock Cargo.toml LICENSE README.md build.rs docs example_data ideas ordered_map signature src target tests todo
As you can see prompt after running ls is printed before command output.
After reading the code I found that this happens because command output is send to a printer thread, i.e. readline loop doesn't wait for command to finish.
Hi. So this is a problem that I haven't quite fond the right solution to. Commands that send streams of data exit as soon as they have sent all their output, even if the next pipeline step (or the pretty printer) hasn't processed them. In general, I think this is the way it should work.
As you point out, printing output in Crush is done by only one thread to make sure that output is still nicely formatted even if multiple background jobs are printing data to the screen at the same time. I still believe this is a good design, but it does lead to the problem you describe.
I guess the Printer struct would need a method that would wait until there is no output pending to be printed. The underlying channel::Sender has a is_empty method, but no wait_until_empty.
still nicely formatted even if multiple background jobs are printing data to the screen at the same time
Note that println! uses stdout which is protected by a mutex. So it's safe to print from multiple threads. And you may want to lock it in the printer thread and use that handle for performance reasons.
Didn't know that, thanks for the tip!
Down the line, I'd like to make multiplexing two streams fully readable, sort of like if you do tail -f *.log, at which point it's not enough to lock by row, you have to take the lock for the duration of printing a "segment" of a stream. I'd say that the whole pretty-printer is rather naive right now.
To be honest, making an amazing pretty-printer doesn't sound like the hardest part of writing a shell with the largest number of pitfalls, so I'm sort of punting that until later and trying to see if I can flesh out the trickiest bits of the design.
Right, any fix would do for now.
I believe 0d4ac9e should fix the issue. At least, I can no longer reproduce the problem.
Let me know if you're still seeing the problem.
Works like a charm, thanks!