Try to capture all filedescriptor output and err #630

This try to fix a long standing issue that stdout and stderr going directly to the filedescriptor are not shown in notebooks. This is annoying when using wrappers around c-libraries, or calling system commands as those will not be seen from within notebook. Here we redirect and split the filedescriptor and watch those in threads and redirect both to the original FD (terminal), and ZMQ (notebook). Thus output sent to fd 1 & 2 will be shown BOTH in terminal that launched the notebook server and in notebook themselves. One of the concern is that now logs and errors internal to ipykernel may appear in the notebook themselves, so may confuse user; though these should be limited to error and debug; and we can workaround this by setting the log handler to not be stdout/err. This still seem like a big hack to me, and I don't like thread. I did not manage to make reading the FD non-blocking; so this cannot be put in the io-thread – at least I'm not sure how. So adds 2 extra threads to the kernel. This might need to be turn off by default for now until further testing. Locally this seem to work with things like: - os.system("echo HELLO WORLD") - c-extensions writing directly to fd 1 and 2 (when properly flushed). I have no clue how filedescriptor work on windows, so this only change behavior on linux and mac.

Bypass the original fd 2 for stderr, and use the new piped one.

This is important for example for subprocess that will peak at the filedescriptor.

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Try to capture all filedescriptor output and err #630

Try to capture all filedescriptor output and err #630

Commits on Apr 13, 2021