Re: possible squid bugs [was: cacheability engine]

On Tue, 28 Mar 2000, Henrik Nordstrom wrote:

> Duane Wessels wrote:
>
> > putting such a change into the 2.x line violates our "promise"
> > that all 2.x versions would be store-wise compatible. People
> > would have to run ugly converters, etc.
>
>
> Have we promised forward compability (that files created with a newer
> version is equally usable by a older release) or only backward
> compability?

It probably wasn't that specific. I take it to mean backwards
compatibility. Specifically, when upgrading you won't lose your cache.

> There is actually no need to change the on-disk object format, only
> swap.state. I have always disliked how the "swap meta" information only
> is a raw dump from the StoreEntry. It should be a strucure of it's own.

Well, its not the whole StoreEntry. Its just the chunk that is
"static". It doesn't include pointers and such.

> The object meta information is not a big deal. The format is extensible
> and allows for changes (new meta format -> new identifier). The

yes, thats true.

> swap.state is a trickier issue as there is no room for version
> information.

I can think of a hack for that. I can add a header to the
swap.state file. A magic constant plus a version number. If
the header is missing, its known to be the old format.

Duane W.
Received on Mon Mar 27 2000 - 23:17:39 MST