As the summary says. If you start ash, then resize the window it's in, ash acts as though you had typed a blank line. To reproduce this without resizing a window, you can start ash in one terminal then send SIGWINCH to it from another terminal; ash reacts by re-displaying the prompt on a new line as though an empty line had been typed. This was introduced some time between 1_35_stable and 1_36_stable, maybe by 12566e7f9b5e5c5d445bc4d36991d134b431dc6c.
Created attachment 9466 [details] proposed fix Hi, the attached patch seems to fix the issue for me. Cheers
This seems to introduce a regression: ====================== echo -ne '' >input echo -ne '33 3!4 3!!5 36 ' | xxd -p -r PASS: xxd -p -r ignores the nibble with 2nd char bad make[1]: *** [/home/maribu/Repos/software/aports/main/bu :/
I have to correct myself: The tests are passing in the CI of Alpine Linux https://gitlab.alpinelinux.org/alpine/aports/-/merge_requests/43365 They are also failing for me without the patch on my machine. So there seems to be something affecting the tests on my local setup, rather than a regression that the unit tests would catch being introduced by the proposed patch.
Created attachment 9471 [details] proposed fix v2 The previous version still occasionally leaked a SIGWINCH instead of transparently handling it. With this new version of the patch I can no longer trigger the issue.
Created attachment 9476 [details] proposed fix v3 The v2 added a compile_commands.json by accident. (I used that for better code insight with clangd.) This is identical to v2 except not adding an unrelated file by accident.
Fixed in git, please test.
Tested at 93e0898c663a533082b5f3c2e7dcce93ec47076d and the bug is fixed :)