Re: snapshot available

From: Dancer <dancer@dont-contact.us>
Date: Mon, 31 May 1999 00:27:47 +1000

Duane Wessels wrote:
>
> I put up a snapshot of the current source at
> ftp://tirana.nlanr.net/priv/prerelease/.
>
> This has a new store-filesystem interface, as documented (hopefully) in
> http://squid.nlanr.net/Squid/Prog-Guide/prog-guide-9.html. It also
> has internal DNS code.
>
> In the process of writing the store IO interface, I pretty much removed
> all the USE_ASYNC_IO bits in disk.c and some other places too. My
> intention is to put all that stuff in store_io_asyncufs.c. I got that
> started, but its not finished. If you try running with
> --enable-async-io you will certainly encounter bugs and assertions.
>
> The 'cache_dir' option now takes a filesystem type as the
> second keyword, i.e.:
>
> cache_dir ufs /cache1 5000 16 256
>
> Also recognized is 'asyncufs', but its broken as described above.
> This change to 'cache_dir' usage is not documented yet.
>
> NOTE: this is not a "DEVEL" release!
>
> Duane W.

Excellent. This gives us something to hack on without getting too far
out of synch.

I've been thinking about:

* DNS (the brief discussion we had about it, and Henrik's comments)
* IP protocol independance.
* Past discussions with Alex about modularity.

I definitely want to look at how you did the DNS stuff. If it's modular
enough, I may be able to drop in IP protocol independance as an
alternative source-module. Otherwise, it could be hacked on until it
_is_ modular.

I'm coming to agree more and more with Alex's standpoint on
modules-not-conditionals.

Curious: Have we thought about dynamic shared objects (ala apache)?
Cross-platform code (AIX, Win32, Solaris, Linux, at least) to implement
this would be a jiff to whip up. (Integrating it usefully with squid
would take a little longer than that, obviously) Then we could spend
less time dealing with configure and Makefile and encourage people to
produce simple modules for things like authentication, DNS, expiry, and
whatnot that could be runtime loaded rather than compile-time only.
Certainly the Red-hat and Debian package maintainers would probably much
prefer something of that order.

It's late, and I'm waffling,
D
Received on Tue Jul 29 2003 - 13:15:58 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:12:08 MST