Re: dynamic modules

----- Original Message -----
From: "Henrik Nordstrom" <hno@hem.passagen.se>
To: "Robert Collins" <robert.collins@itdomain.com.au>
Cc: <squid-dev@squid-cache.org>
Sent: Wednesday, January 31, 2001 8:43 AM
Subject: Re: dynamic modules

> Robert Collins wrote:
>
> > anyone think that dynamic modules are a good thing/bad thing?
>
> I am in favor of it, with interfaces for at least
>
> * object store
> * removal policy
> * reply filters
> * request filters (extrapolation of redirectors)
:-] Already partially there...
> * ACL types

Right. We're thinking along the same sort of lines. I am planning to put the module framework code in one place, with a defined API,
and each module calls one or more registration functions - one for each of the types above.

>
> To allow for this, a few APIs of Squid should be formalized:
>
> * how to register for filedescriptor events
> * how to open files/sockets/pipes
> * how to sleep (time events).

Are these not formalised at the moment? How generic can we be? I.E. is there any reason we can't simply specify that to sleep, use
add_event, to block use a callback approach combined with the current fd routines? Or Did you mean we just need to write
how-to-use-the-existing-routines down in the writing modules guide?
What I'm trying to say, is these issues already exist for the fs,repl,auth modules today : what do we need to do in a minimalistic
sense that is different?

>
> I am also in favor for using dynamic linking in bringing these modules
> in, but dynamic linking should be a optional feature, allowing for a
> static binary to be built with all the required modules linked in from
> start.

Agreed. For secure setups dynamic loading is bad IMO.

Rob
Received on Tue Jan 30 2001 - 15:23:40 MST