Re: [RFC][PATCH] Reset ZLE hooks when changing prompt themes (was Re: [RFC][PATCH] `newuser` prompt theme)

[Lawrence Velázquez <larryv@xxxxxxx>]

Hi,

Is workers/48609 ready to commit, or should Marlon make further
changes, or...?

vq

On Wed, May 5, 2021, at 2:10 AM, Marlon Richert wrote:
> On Sun, May 2, 2021 at 8:59 PM Bart Schaefer <schaefer@xxxxxxxxxxxxxxxx> wrote:
> >
> > On Fri, Apr 16, 2021 at 3:38 PM Marlon Richert <marlon.richert@xxxxxxxxx> wrote:
> > >
> > > Attached is a new version of the patch that handles _all the things._ \o/
> >
> > This looks OK, with two questions:
> > 1.  Subshells have been introduced.  Those won't be treated as
> > interactive shells.
> 
> Are you sure about that? When I do `foo() { ( setopt ) }; foo`, it
> still lists 'interactive' (though not 'monitor' or 'zle').
> 
> > Does that matter?
> 
> It doesn't appear to matter. The prompt previews look the same as
> before. Try `prompt -p` (without further args). Doesn
> 
> > 2.  This bit in the restore style:
> > +        $( add-zsh-hook -L )
> > +        $( add-zle-hook-widget -L )
> > There may not be anything to be done about this, but unless I'm
> > mistaken that just re-adds all the previous hooks, it doesn't remove
> > any?  The code it is replacing clobbered the entire list of precmd and
> > preexec hooks, which I suppose also might also be wrong if anything
> > other than the prompt theme was manipulating them.
> 
> `add-zsh-hook -L` outputs statements in the form of `typeset -g -a
> <hook>_functions=( <func> ... )`, while `add-zle-hook-widget -L`
> outputs statements in the form of `zstyle zle-<hook> widgets
> <nr>:<widget> ...` (which is where add-zle-hook-widget stores the hook
> widgets to call internally). eval'ing either of these results in the
> same good, old-fashioned clobbering as in the old code. So, no
> functional change there. :)
> 
> I just realized, though, that these lines (new and old) don't actually
> do anything useful:
> * When you call `prompt`, it will not unhook anything not called
> `prompt_*_<hook>`. So, whatever hooks you've set up that don't follow
> the naming scheme will still be there when you call `prompt restore`.
> No need to restore them.
> * When you call `prompt`, it will unhook everything called
> `prompt_*_<hook>`. So, if a prompt theme correctly implements the
> prompt system contract, its hook functions will already get unhooked
> when switching themes. Again, no need to do anything special when
> calling `prompt restore`.
> 
> The only case I can think of that would need special handling is if
> you already had hooks installed that follow the `prompt_*_<hook>`
> naming convention before calling `prompt` for the first time.
> 
> Should we just get rid of this part of the "restore" logic? And
> instead just document clearly that `prompt` will auto-remove all hooks
> that follow its naming scheme?
> 
> 
> And speaking of prompt hooks and naming schemes:
> 
> On Fri, Apr 16, 2021 at 12:34 AM Daniel Shahaf <d.s@xxxxxxxxxxxxxxxxxx> wrote:
> > Marlon wrote on Thu, Apr 15, 2021 at 21:54:48 +0300:
> > > On 14 Apr 2021, at 15:05, Daniel Shahaf <d.s@xxxxxxxxxxxxxxxxxx> wrote:
> > > >> +++ b/Functions/Prompts/promptinit
> > > >> @@ -178,8 +177,13 @@ Use prompt -h <theme> for help on specific themes.'
> > > >>
> > > >>        # Reset some commonly altered bits to the default
> > > >>        local hook
> > > >> -       for hook in chpwd precmd preexec periodic zshaddhistory zshexit; do
> > > >> -         add-zsh-hook -D "${hook}" "prompt_*_${hook}"
> > > >> +       for hook in chpwd precmd preexec periodic zshaddhistory zshexit \
> > > >> +           zsh_directory_name; do
> > > >> +         add-zsh-hook -D "$hook" "prompt_*_$hook"
> > > >> +       done
> > > >> +       for hook in isearch-exit isearch-update line-pre-redraw line-init \
> > > >> +           line-finish history-line-set keymap-select; do
> > > >> +         add-zle-hook-widget -D "$hook" "prompt_*_$hook"
> > > >
> > > > Recommend to name these prompt_${foo}_bar-{isearch-exit,isearch-update,…,keymap-select}
> > > > for some fixed value of «bar» to avoid namespace issues (i.e., name
> > > > collisions between existing prompts and future hooks).
> > >
> > > Wouldn’t that be a breaking change in the API, though?
> >
> > No, it wouldn't, because those are add-zle-hook-widget hooks and the
> > current API only deals with add-zsh-hook hooks.
> >
> > We could even recommend that add-zsh-hook hooks be named
> > prompt_${foo}_baz_{chpwd,…} for some fixed value of baz.  We don't even
> > have to change the code to support this (because the asterisk will match
> > «${foo}_baz» just fine); we just need to document the recommendation and
> > deprecate the previous recommended naming pattern.
> >
> > > Also, isn’t the prefix prompt_${foo}_ already a namespace of sorts? Why change this?
> >
> > What if the foo theme declares a function called prompt_foo_lorem (or
> > even prompt_foo_ipsum_lorem) that _isn't_ an «add-*-hook lorem» hook
> > function, and later we start to to auto-register functions matching the
> > name pattern?
> >
> > Or if we were to make «add-zsh-hook bar lorem» and «add-zle-hook-widget
> > bar ipsum» both valid, for a single value of bar, and a prompt theme
> > wanted to install both hooks _and_ name the functions «lorem» and «ipsum»
> > conventionally?
> >
> > Or if someone wanted to quickly grep for add-zle-hook-widget callbacks?
> 
> If we're going to change the prompt function naming scheme anyway, can
> we change it to something that starts with . or _? Because then
> `zstyle ':completion:*' prefix-needed yes` will conveniently hide
> these.
> 
> How about .prompt.<theme>.<hook>? And then auto-register these
> functions as hooks?
> 
> And should all of this go into the same patch? If `prompt`
> automatically unregisters all hooks that follow a naming pattern
> (regardless of how they've been registered), then I think it would
> make sense for it to automatically register them, too.
> 
>