[1069] in BarnOwl Developers
Re: Rethinking ::Hook
daemon@ATHENA.MIT.EDU (David Glasser)
Thu Oct 29 18:12:42 2009
Resent-From: nelhage@mit.edu
Resent-To: barnowl-dev-mtg@charon.mit.edu
X-Original-To: nelhage@nelhage.com
Date: Sat, 17 May 2008 12:45:15 -0700
From: "David Glasser" <glasser@davidglasser.net>
To: "Nelson Elhage" <nelhage@mit.edu>
Cc: barnowl-dev@mit.edu
In-Reply-To: <20080517193559.GP8761@mit.edu>
On Sat, May 17, 2008 at 12:35 PM, Nelson Elhage <nelhage@mit.edu> wrote:
> Hey,
>
> I'd like to think about rethinking how BarnOwl::Hook works.
>
> Currently, for each hook (startup, shutdown, new message, main loop),
> we clear all watchers of the hook before a reload, and rely on
> reloading the modules to reinstall them.
>
> This has the unfortunate side-effect that we can't reload single
> modules, only all-or-nothing. And if a module fails to load because of
> e.g. a syntax error, it potentially gets broken, rather than the old
> version continuing to run. I think this is on the whole poor.
>
> The reason we have this behavior is to prevent a module's handler from
> ending up in the hook list twice. We have to do this because modules
> only give us a coderef when registering with a hook, and if the module
> has reloaded, the coderef is _different_, so we can't just check if
> it's already in the list.
>
> I see at least four different options to allow better behavior. Some
> of them can be used together:
>
> 1. Have modules register a function name as a string, rather than a
> subref. This has the property (not always good) that if the code
> changes but the hook doesn't get re-registered, the new hook takes
> effect.
>
> 2. Use Sub::Identify or similar to ask perl what function the
> subroutine represents, and filter the hook list to remove any other
> function of that name.
>
> 3. Use caller() to figure out which module is registering this hook,
> and only de-register hooks for a specific module before reloading
> that module
>
> 4. Have modules provide a string "name" or identifier for each hook
> they register, and remove handlers sharing the same name.
>
>
> I think, on the whole, that I prefer 2. It does have the property that
> anonymous subrefs (i.e. sub{...}) will get registered twice anyways.
I'd say that a good design would be something that either makes
sub{...} work correctly, or makes it obvious that they can't work (eg,
option 1). Making it look like any function will work but secretly
only supporting named subs seems like trouble.
--dave
--
David Glasser | glasser@davidglasser.net | http://www.davidglasser.net/
