httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <>
Subject Re: [PATCH] PR37145: data loss with httpd-2.0.55 reverse proxy method=post
Date Fri, 21 Oct 2005 17:43:20 GMT
Ruediger Pluem wrote:
> On 10/19/2005 08:25 PM, William A. Rowe, Jr. wrote:
> [..cut..]
>>Researching as well.
> Any new results from your research? Otherwise I would like to commit the latest version
> of my patch to trunk if you have no objections.

Yes ... and from the breadth of your other patch, and other non-BRIGADE_CONCAT
cases in the code, I'm taking a deeper look at the semantics of transient
buckets.  As originally conceived, the contract was pretty clear, they existed
on the stack (or similar) and could be trusted only until the callee returned.

In this case, obviously, they exist until brigade get is called again.  But
we had never spelled out how transient a transient bucket is, except in the
case of pushing output buckets down pass brigade.

Go ahead and commit, it's obviously an effective solution for this specific
set of cases.  I'm not certain, however, that it's the entire fix :-/

We might be better off using this fix (and documenting the usage of all get
brigade calls w.r.t. transient buckets), while in 2.0.x we might want to
return an allocated bucket in mod_ssl to ensure third party 2.0 modules, with
this very same mis-assumptions, don't trip over this effect.  Thoughts?


View raw message