[darcs-users] [patch310] Resolve issue 1599: automatically expire unused caches

[Petr Rockai]

Eric Kow <kowey at darcs.net> writes:
> If the exception is not a timeout, we test if the URL is reachable and
> whitelist it if so (the whitelist avoids future testing).  But why do we
> do this?  And doesn't that confuse Darcs into thinking it successfully
> fetched a file when it did not?  Have you tested the actual case where
> you get an exception but not a timeout?

I believe that we do this because a 404 could be quite legitimate: the
repository could be partially lazy but still have a lot of things we
want to use. We don't want to blacklist a source for this reason
alone. We also don't want to inspect a source each time it gives a 404,
since this is not for free, either.

> Also, are there not web server configurations where
> http://example.com/foo gives you a 404, but
> http://example.com/foo/_darcs/foo gives you a result?  What happens in
> those cases?  In other words, does the test of trying to fetch
> http://example.com/foo/ actually make sense?
>
> Also, why is it important to normalise the URL to have a trailing slash?
> Is it just to avoid duplicate entries in the whitelist?

I suspect that there are webservers that treat foo/ and foo
differently. But I am not saying one is better than the other. Anyway, I
think the right thing to look for is foo/_darcs/hashed_inventory (we
shouldn't be using caches for non-hashed repositories).

I wholeheartedly agree with the indentation issue Eric has pointed out,
too. Please try to fit your code in 80-90 columns if possible, although
100 is acceptable and 110 is stretching it but could be forgiven if
there's a good reason. : - ) Code that is wider is usually hard to read
for almost everyone (or so I guess).

Yours,
   Petr.