`wildtrigger()` might silently set `v:errmsg`

[echasnovski]

Steps to reproduce

1. Create an 'init.vim' file with the variation of content from :h cmdline-autocompletion:

   let g:log = []
   autocmd CmdlineChanged [:\/\?] call wildtrigger() | let g:log = g:log + [[getcmdline(), v:errmsg]]
   set wildmode=noselect:lastused,full
   set wildoptions=pum

2. vim --clean -u init.vim

3. Type :tag<Space><Space>. There is no error shown, but the v:errmsg is set to E433: No tags file. It can be verified later by :echo g:log which contains an entry ['tag ', 'E433: No tags file'].

   Pressing <Esc> shows an error E73: Tag stack empty, but it is the same behavior as without wildtrigger() autocompletion.

Expected behaviour

Calling wildtrigger() never silently sets v:errmsg. Only if that message is actually shown.

Version of Vim

9.1.1908

Environment

OS: EndeavourOS Linux x86_64, 6.17.8-arch1-1
Terminal: Ghostty 1.2.3-arch2
Value of $TERM: xterm-ghostty
Shell: nushell 0.108.0

Logs and stack traces

[zeertzjq]

Only if that message is actually shown.

This is not how v:errmsg works. The example at :h v:errmsg even says that you can use it to get the error message of a command run with :silent!:

:let v:errmsg = ""
:silent! next
:if v:errmsg != ""
:  ... handle error

[echasnovski]

Only if that message is actually shown.

This is not how v:errmsg works. The example at :h v:errmsg even says that you can use it to get the error message of a command run with :silent!:

Yes, I've seen that. Maybe got confused with Last given error message. as to me "given" ~~ "shown".

But nevertheless, I don't know if wildtrigger() setting v:errmsg at all is justified. It might, but in practice I've encountered it as an undocumented behavior. If it is justified, then at least documenting it explicitly on top of "... generally operates more quietly" makes the behavior more transparent.