Public SourceMage Discussion List

[Andrew <afrayedknot AT thefrayedknot.armory.com>]

On Fri, Aug 20, 2004 at 03:44:49PM -0700, Eric Sandall wrote:
> Quoting Andrew <afrayedknot AT thefrayedknot.armory.com>:
> > Master site should be on the lists, if its not file a bug.
> >
> > Theres really very little we can do on the sorcery side to fix the problem
> > of only the master site having the source (how are we supposed to
> > know?). However when sorcery expands urls it keeps the requested urls
> > (the ones listed in the spell) first. So a spell could list the main site
> > explicitly (rather than GNU_URL), of course the order gets re-arranged
> > the next layer down if netselect is used *sigh*.
> >
> > Theres a very marginal benefit to adding a field in DETAILS like
> > "NO_URL_EXPAND=1" or something equally clever, then the spell could have
> > that bit set, and the SOURCE_URL set to a site known to have the source,
> > then once the source is reasonably distributed remove the NO_URL_EXPAND
> > tag. Of course I wonder how often it is the case that only one site in
> > the world has the source...  (just looking at cost/benefit analysis)
> 
> That's probably not worth it as (from my experience) the mirrors are only 
> up to
> a day behind (some SF mirrors are a little slower, but that's very few).

Exactly, it would require work for me (and we cant do that! :-) ), and
it would essentially punish gurus for updating spells early, they'd have
to modify the spell to force the use of the master site, then remember
to go remove that a day or two later. I think the better answer is to
use stable, or wait a few minutes for sorcery to try the master url
(or wait a day or two for the source to distribute). We do background
downloading for this reason, you can still do useful stuff with your cpu
while we scrub the internet for your sourcefile.

> I'd
> rather not explicitly list a URL if it's included in the mirrors list as 
> that
> just seems redundant to me. Could netselect and/or Sorcery do a check to 
> see if
> the file even exists before trying to download it or would that just add 
> more
> headaches for little to no gain?

Netselect doesnt do any checking for the file (to my knowledge), it knows
just enough to pull the hostname out of standard urls, and do icmp stuff
with it.

Assuming we're talking about the case of wget-able files, we actually do
a lookup first (see url_foo_verify) although I question the necessity
of it, theres really very little difference between checking if a file
is there and actually downloading it, its kinda silly to connect and
see if something is there, disconnect and reconnect again. Imagine if
we did this with phones:

*ring*
"Hello"
"Hi is eric there?"
"yea let me get him"
...
"hi this is eric"
"i just wanted to know if you were there, thats all, bye"
<click>
*ring*
"hello?"
"hi can i talk to eric (now that i know he's there)?"

I have the same sentiment about the whole FUZZ feature, where we try to
poke around for alternate extensions to the file (Seth has been lobbying
for its removal for a while now).

-Andrew