[Zope-dev] Re: [Zope] IE5 / Medusa bug?

[seb bacon]

Firstly, apologies for cross-posting - this should probably move over
to zope-dev henceforth.

Secondly, thanks for the replies.  In my small sample, it appears that
3 out of 7 IE browsers got the bug - no other flavours had a problem
at all.  They were mostly versions 5 and 5.5.  If there was any
pattern, it's possible that 56-bit versions were the baddies, but
that's not clear yet.

As Martijn correctly pointed out, there seems to be a Content-length
problem going on here.  I did a lot more tests on two different 2.2
servers, and have come closer to isolating the bug.

Contrary to what I said, it *is* triggered by bad html: the </head> 
tag was missing.  In documents of a certain length (at least 2000
bytes), which contain the <base href> tag, but don't have a </head>
tag, the Content-length is consistently reported as 14 bytes too
short.  This is regardless of the actual length of the document.

Why this should have a knockon effect on a small subset of MS
browsers, I don't know.  Especially confusingly random is the
requirement for a certain amount of data to break these browsers.

Even more mysteriously, if I replicate the data which zope is sending
to the browser using netcat (bad content-length, same content), IE
doesn't complain.  Things like Keep-alive perhaps spring to mind?

Anyway, I have fixed the problem in terms of my users by adding the
</head> tag, but this looks suspiciously like a ZServer problem as
well as an IE rendering problem.  I'm off on holiday tomorrow evening 
and I don't have any more time to investigate.  I haven't checked
against other versions of Zope, but the one I'm using is only patched
in DT_Util.py.

This is such a dumb problem it's just *got* to be my fault somewhere
down the line :)  I'll check against other servers when I get back
(mid Jan - long holiday :-)

cheers,
seb