I'm not exactly sure what this feature will be used for. If you want to script something around a program expecting to run in a term there are already better tools for that like expect [1]. I'm not the one who added this goal.
On Wed, Sep 12, 2012 at 8:22 PM, Roberto E. Vargas Caballero <[email protected]> wrote: > I agree with you that it is a bit stupid, and maybe we should use directly > write. Look these new patches. I was thinking of using the dup2 syscall in the shell process (fork child) to make the the stderr/stdout redirect to the file. That way you don't even need to call write explicitely. But I'm not sure you can do this since I'm already redirecting them to the cmdfd. I don't know if it's faster either. Hence my previous comment in the form of a question :) Since we're dealing with a hdd, my guess would be that buffered io is faster given you buffer more than 1 character (which was not the case the first patch). It really all depends on what this feature is for. To me it just look like some kind debug tool for st. 1: http://en.wikipedia.org/wiki/Expect
