[Pkg-zsh-devel] Bug#898619: Bug#898619: Bug#898619: zsh autocompletion does not deal gracefully with upgrades in the background

[Philipp Kern]

Hi,

On 2018-05-15 00:09, Frank Terbeck wrote:
> Axel Beckert wrote:
>> Philipp Kern wrote:
> [...]
>>> My gut feeling is that this is because I did not autocomplete in the
>>> old shell yet but now the module is gone.
>> 
>> Yes, that's the same conclusion I came to.
>> 
>>> At least with a new shell complist.so is only in /proc/$$/maps after
>>> an attempted completion.
> 
> Yes, modules are loadable at run-time.
> 
>>> Would it be possible to make sure that the autocompletion system
>>> loads the module on startup rather than on-demand?
> 
> Zsh is  setup to  do lots  of things  on demand.  It would  certainly 
> be
> conceivable  to be  more gracious  about  the issue.  But when  
> versions
> upgrade, I don't  think it's unreasonable to restart a  shell, like 
> Axel
> mentioned via "exec zsh".
> 
> If you'd really like that module loaded all the time, you can load it 
> in
> your setup. Personally, I'm loading a bunch of modules at startup time.

I'm caring about a large setup rather than a personal installation, 
though, which sort of shifts as to what's reasonable and what isn't. ;-)

Let's say I'm using /etc/zsh/newuser.zshrc.recommended (which is 
actually what I'm using verbatim). Could we at least fix it for that 
configuration? That one does this:

| # Use modern completion system
| autoload -Uz compinit
| compinit

/usr/share/zsh/functions/Completion/compinit does not invoke "zmodload 
-i zsh/complist" whereas the first invocation of the completion menu 
would. How bad would it be to add the module loading to the 
initialization? Is there a benefit to not doing that specific call on 
startup? The module itself is a 63k .so file.

In a very non-scientific test I get a VM increase from 37076 kB -> 43968 
kB (+18%), anon pages from 1840 -> 2148 kB (+16%) and file pages from 
3912 -> 4508 kB (+15%) on amd64 when running a full completion for 
l<tab> on my system. And that loads even more than just the module, 
because it goes through the whole setup.

I mean sure, I could add a zmodload call to our system-wide config. But 
I'd rather we solve this problem for all of Debian's users rather than 
just us. (This is assuming that people are on some variant of a rolling 
testing and don't reboot their machines all the time after updates - 
which currently violates other Debian assumptions like the kernel's as 
well.)

Kind regards and thanks for your consideration
Philipp Kern