Mailing-List: contact dev-help@httpd.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@httpd.apache.org
Received-SPF: neutral (nike.apache.org: local policy)
Message-ID: <46FAB60F.9030906@rowe-clan.net>
Date: Wed, 26 Sep 2007 14:42:07 -0500
From: "William A. Rowe, Jr." <wrowe@rowe-clan.net>
User-Agent: Thunderbird 1.5.0.12 (X11/20070719)
MIME-Version: 1.0
To: dev@httpd.apache.org
Subject: Re: 2.2.7
References: <C69314BD-56AC-4A0D-866D-6AE8DDCD7269@jaguNET.com>
 <46FA8DEA.8030204@rowe-clan.net> <004601c80070$3e576070$0100a8c0@mother>
In-Reply-To: <004601c80070$3e576070$0100a8c0@mother>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit

Steffen wrote:
> There is the ASF statement/promise:
> 
> "Modules compiled for Apache 2.2 should continue to work for all 2.2.x
> releases."

***SHOULD*** is the operative word.  There are always exceptions.

I have a half dozen examples where I've abused the microsoft foundation
classes in my code elsewhere, which have been broken by internal changes.
Why?  Because I dug into the internals instead of staying at the API's
documented presentation layer.  Of course, this was the only way I could
use the MFC's functionality to accomplish what I was trying to do, so I'd
balanced the danger of leveraging the internals against the benefits of
tweaking the internals.

Exploiting a flaw doesn't make it part of the API contract.  ABI was not
broken, and in fact won't be broken again with my proposed correction.

In this case, we have two assumptions, that win32 handles are 'special
things' which should behave differently than unix stdin/out/err when
the APR refactoring had intended to make that not-the-case.  And that
you can always emit errors to stderr and record them somewhere, while
with the old piped logging code (on ANY platform!) this was not so.

Both flaws are corrected, the fallout, IMHO, is a tiny fraction of the
user base but must be addressed for those users promptly in another
iteration.  That's how open sources work.

FWIW there is another bit of fallout from keeping a usable stderr stream
available at all times; c.f. Bug 43491 which I'm also analyzing for a
solution by 2.2.7.

> So I expect that modules have *not* to be new compiled. It is quite
> confusing for our users en authors have to maintain in most
> cases up to 4 versions. What about the Magic Number ?

I agree that due to these corrections, the MMN must be bumped with 2.2.7,
thank you for raising that point!

> Btw.
> I guess (when I see the downloads)  already a few thousands (production)
> sites are running the 2.2.6 version from the ApacheLounge*. Not a single
> problem report received, all running well including mod_perl and other mods
> which are failing with the ASF build.

Equate downloads with production users?  A tiny fraction from the feedback
I've received, the majority of windows users who've spoken to me via email,
at conferences etc are using windows for two purposes, testing/staging, and
learning.  There are certainly very robust windows users in production, but
those don't correspond to the number of downloads you observe.

But that's a fraction of the hundreds of thousands of hits that /dist/httpd/
takes per release, and we can't even most mirror activity!  So I'm sure from
your downloaders' perspective all is well, but the vast majority looking to
httpd for the sources and binaries wouldn't be served by ignoring them.

> So for this moment the Windows Community has no need to hurry a new 2.2.7.
> We are also a little afraid that same things can happen as with the
> unusually closely 2.2.5==>2.2.6 process, a broken build.

s/the/your/ -- Good to hear, although as Tom suggests - the patches he
provided you are 'mysterious' :)  When he (and I) thoroughly understand
the corrections and the reason they are the 'right' thing, we'll both be
satisfied that it's a good thing for -all- users to adopt the patches.
This is the process of /developer/ community review, a major shortcoming
of many third party offerings (and some commercial products, for that matter).
Tom's done a great job of clarifying this to users downloading the package,
but I'm afraid your messages aren't so on-target.

AFAIK the build was not broken, do you know something I don't?  I've tested
./configure; make and nmake -f Makefile.win against a host of versions and
OS's and don't recall any failure.

Code developed under scrutiny of your peers (in the win32 case, such as Tom,
Jorge, Mladen, Randy, etc - even myself) is what the ASF offers, as opposed
to what AL has up for download mislabeled as 2.2.6.  I don't mean users'
testing and observation, I mean people who understand the internals.  When
your patch happens to work (and you aren't sure why) the first question is
to dig deeply, and consult your peers, which is what Tom is engaged in.