100% cpu usage, polling the file tl is looking at

[nealmcb]

Whenever I run tl, it uses 100% of a cpu. Running strace -fp shows three lines over and over, perhaps 100,000 times a second.

[pid 219465] lseek(7, 0, SEEK_CUR)      = 423
[pid 219465] read(7, "", 65536)         = 0
[pid 219465] poll([{fd=7, events=POLLIN}], 1, 100) = 1 ([{fd=7, revents=POLLIN}])
[pid 219465] lseek(7, 0, SEEK_CUR)      = 423
[pid 219465] read(7, "", 65536)         = 0
[pid 219465] poll([{fd=7, events=POLLIN}], 1, 100) = 1 ([{fd=7, revents=POLLIN}])

File descriptor 7 is the file I'm running tl on. 423 is the length of the file in bytes.

This is on Ubuntu 22.04.3. tl was installed via pipx, version 1.0.3

SInce efficiently watching a file for changes is a feature, something seems broken.

[tstack]

On linux, poll() is used in here:

toolong/src/toolong/watcher.py

Lines 14 to 15 in 9fe54a6

	# DefaultSelector returns epoll on Linux, but that gives a permission error on Ubuntu
	from selectors import PollSelector as DefaultSelector

However, poll() / select() do not work with regular files (and neither does epoll(), which is why it was returning EPERM). These functions will always return that a file is ready to read/write. So, the loop in this file will spin and use 100% CPU.

The code likely works on something like MacOS/FreeBSD because they use the kqueue selector which has support for files.

[nealmcb]

Many thanks, @tstack.
I'm sure this is overly crude, but if I don't care about breaking the tail feature, it seems I can avoid the cpu loop by simply commenting out this line:

toolong/src/toolong/watcher.py

Line 77 in 9fe54a6

self._selector.register(fileno, EVENT_READ)

A quick search suggests an approach that should be a more reasonable option:

How can I tail a log file in Python? ... Purely pythonic solution using non-blocking readline()

It requires polling, but the current code also seems to set a timeout and thus poll every tenth of a second.

[willmcgugan]

Thanks. Should be fixed in 1.1.1

I feel cheated by the Python docs that implied files would work everywhere.