[pacman-dev] [PATCH] Add events ALPM_EVENT_HOOK_{START,DONE}

[Andrew Gregory]

On 12/13/15 at 10:38pm, Allan McRae wrote:
> On 07/12/15 03:35, Olivier Brunel wrote:
> > Add events to let frontends know when hooks are being processed (and when it's
> > done), as that might be useful to update the UI.
> > 
> > It can also be useful to know when ALPM_EVENT_SCRIPTLET_INFO might be happening
> > in that regard (There's no specific event for running scriptlets, but they do
> > happen at specific times after/before ALPM_EVENT_PACKAGE_OPERATION_{START,DONE}).
> > 
> > Signed-off-by: Olivier Brunel <jjk at jjacky.com>
> > ---
> > On a not-quite-related topic, I wonder whether a flag should be added to skip
> > hook processing, much like ALPM_TRANS_FLAG_NOSCRIPTLET ?
> > I guess it might not be necessary, since frontends can simply set the hookdirs
> > to NULL to disable them. Thoughts?
> > 
> > I also wondered about the same in the POV of pacman, i.e. an option --nohooks,
> > because it seems currently one can only add hookdirs, and therefore only replace
> > the pacman default (/etc/pacman.d/hooks) (I assume setting it to /dev/null might
> > work to disable things?) but not the system/ALPM one (/usr/share/libalpm/hooks)?
> > 
> > Also unless I missed it, the documentation doesn't mention that last directory
> > anywhere?
> > 
> 
> Agreed - we need:
>  - document --hookdir in pacman.8.  It should be mentioned that this is
> what HookDir is overriding in pacman-conf.8 too.
>  - add a --nohooks option
> 
> 
> And now, onto patch review.
> 
> I think _alpm_hook_run needs to exit after reading the hooks directories
> if there is no hooks. Only if there are hooks should
> ALPM_EVENT_HOOK_START be called.  It would be even better if it was
> called if there were hooks for pre or post transaction.
> 
> That would require _alpm_hook_run to first loop over the hooks doing:
>   if(hook && hook->when == when && _alpm_hook_triggered(handle, hook)) {
> then if any hooks matched, do ALPM_EVENT_HOOK_START and run the hooks.
> I think that check is fast enough that we do no need a callback until we
> know some hooks will be run.  Andrew: would you agree?

Agreed.  It has been plenty fast in all of my tests so far.

apg