Bisect points to http caching

[Augie Fackler]

On Oct 10, 2010, at 5:09 PM, Matt Mackall wrote:

> On Sun, 2010-10-10 at 22:23 +0200, Dirkjan Ochtman wrote:
>> On Sun, Oct 10, 2010 at 19:16, Matt Mackall <mpm at selenic.com> wrote:
>>> I just bisected some weird behavior in hgweb where I was getting
>>> partial/mangled pages and/or styling and found this:
>>> 
>>> The first bad revision is:
>>> changeset:   12183:f64b416b0ac8
>>> user:        Dirkjan Ochtman <dirkjan at ochtman.nl>
>>> date:        Wed Sep 08 15:23:48 2010 +0200
>>> summary:     hgweb: support very simple caching model (issue1845)
>>> 
>>> Sticking a return as the first line of caching() fixes the problem. The
>>> server is single-repo hgweb under Apache2 mod_wsgi and possibly under a
>>> WSGI server Augie's running called "spawning". But the bad behavior
>>> doesn't seem to appear under hg serve.
>> 
>> Hmm, sucks. Is this actually reproducible, or do you have a single
>> case where this fails?
> 
> Run Apache/"spawning" with hgweb.wsgi. On any hgweb page, hit reload
> several times in a row. Sometimes it'll hang, sometimes it'll return a
> mangled page. No error log output. Augie and I confirmed this on two
> separate setups.

I investigated this some today with spawning (get it on pypi) in single-thread single-worker mode. The bug still occurs. Safari claims it gets a 304 from the server, but spawning thinks it served a 200. I'll try and look again with wireshark to figure out what's up, but we may need to disable caching for 1.7.

> 
> -- 
> Mathematics is the supreme nostalgia of our time.
> 
> 
> _______________________________________________
> Mercurial-devel mailing list
> Mercurial-devel at selenic.com
> http://selenic.com/mailman/listinfo/mercurial-devel