Mailing List Archive

Date: Mon, 7 Aug 1995 00:32:19 -0700 (PDT)
From: Brian Behlendorf <brian@organic.com>
Precedence: bulk
Reply-To: new-httpd@hyperreal.com

The GXBITHACK functionality meant that a If-Modified-Since request on a SSI
document will have the timestamp checked against the SSI document's file
system LM time, if the group x bit is set, yes? If so, I think that's
not working in 0.8.8. An easy test - Hyperreal's home page is designed
to have random images next to the "raves" and "music" sections, and with
GXBITHACK the images were random the first time and the same upon
immediate reloads. Now, they're different every time, and the
permissions of that file are indeed rwxrwxr-x. Thoughts?

Brian

Hyperreal's home page isn't the best test case --- it's MultiViews
content-negotiated ("DirectoryIndex index"), which cuts off all forms
of caching or conditional GETs right there.

I did try non-negotiated test cases when I first wrote the code, and
they at least appeared to work when I tried it...

rst

On Mon, 7 Aug 1995, Robert S. Thau wrote:
> Hyperreal's home page isn't the best test case --- it's MultiViews
> content-negotiated ("DirectoryIndex index"), which cuts off all forms
> of caching or conditional GETs right there.
>
> I did try non-negotiated test cases when I first wrote the code, and
> they at least appeared to work when I tried it...

Ah, okay, I just tried http://hyperreal.com/index.html over and over, and
that worked as expected. Why does MultiViews prevent conditional GETs?

Brian

--=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=--
brian@organic.com brian@hyperreal.com http://www.[hyperreal,organic].com/

> Date: Mon, 7 Aug 1995 10:32:51 -0700 (PDT)
> From: Brian Behlendorf <brian@organic.com>

> Ah, okay, I just tried http://hyperreal.com/index.html over and over, and
> that worked as expected. Why does MultiViews prevent conditional GETs?
>
> Brian

> As with type-map files, Apache-current tries to keep caches from
> caching negotiated documents (unless CacheNegotiatedDocs is explicitly
> set in the config files). It used to do this by adding a Pragma: line
> to the response headers, but there are caches that don't actually
> respect this, so I switched it to withholding Last-modified:, which
> seems to inhibit caching more thoroughly. (NB conditional GETs only
> make sense if you already have a cached copy).

> Whether it makes sense to do this when only one variant exists (as in
> the typical "DirectoryIndex index" case) is, of course, open to
> debate.

It might be better to use "Expires: 0" (*) rather than withhold
a "Last-Modified".

(*) or however "now/immediately" is expressed.



Caches (including Netscape browser) take notice of "Expires".

Some proxy caches try to compute an "Expires" if they don't find one, based
on Last-modified. If they also have to guess L-m or assume it equals "now",
then it can get into a real mess and set "Expires" to something silly.


+1 for a better behaved Apache.


rob

Date: Mon, 7 Aug 1995 10:32:51 -0700 (PDT)
From: Brian Behlendorf <brian@organic.com>

Ah, okay, I just tried http://hyperreal.com/index.html over and over, and
that worked as expected. Why does MultiViews prevent conditional GETs?

Brian

As with type-map files, Apache-current tries to keep caches from
caching negotiated documents (unless CacheNegotiatedDocs is explicitly
set in the config files). It used to do this by adding a Pragma: line
to the response headers, but there are caches that don't actually
respect this, so I switched it to withholding Last-modified:, which
seems to inhibit caching more thoroughly. (NB conditional GETs only
make sense if you already have a cached copy).

Whether it makes sense to do this when only one variant exists (as in
the typical "DirectoryIndex index" case) is, of course, open to
debate.

rst