Am Sun, Jun 28, 2026 at 01:55:37PM +0200, schrieb Dr. Tobias Quathamer: > I've tried to look into the test logs, but wasn't able to figure out > what went wrong. Could you please have a look?
To reproduce a little more interactively you can run the autopkgtest locally and enter the chroot on failure (I do this with a autopkgtest-virt-unshare at least). # apt install vim-nox vim-youcompleteme # vim -c 'set encoding=utf-8' -c 'pack youcompleteme' ycmd/tests/go/go_module/unicode/unicode.go (press `c` if youcompleteme asks to load an ycm_extra_conf.py) `:call cursor(11,7)` (you should now be on the 'U' of 'Unicøde') `:YcmCompleter GoToType` You will see a "RuntimeError: Cannot jump to location" in the bottom. That is what the test expects to see. Now exit vim (`:q`) and lets be puzzled for a moment why the test fails seemingly with a successful jump if we can not reproduce this manually. Now, my chroot environment is extremely minimal. It doesn't even have HOME set, let's fix that: `export HOME=/root`, repeat the vim action from above and lo and behold: We are taken to a type definition now! Currently that is `/usr/lib/go-1.26/src/builtin/builtin.go` on line 77 saying something as witty as `type int int`. I will leave the judgement if that is useful to someone who can actually write Go. I don't. It seems at least dubious that this depends on HOME (or some other way of placing GOCACHE), but okay. It at least does not seem overly useful to block a CVE on such a negative test being 'broken' now, hence I will disable that one and wait for ycmd upstream to catch up (they are on 0.20). Sorry that this was blocking… ycmd (and vim-youcompleteme) have a bunch of these tests which check for specific bugs and missing features that I am playing wack-a-mole with as those upstream tests tend to be better at catching also the occasional real problem than the smoke tests that I could come up with on my own. Best regards David Kalnischkies
signature.asc
Description: PGP signature
