[PATCH] Review & Test: mod_include redesign

[André Malo <nd...@perlig.de>]

[CC to Ron Park and Glenn Nielsen to get this onto their radar]

Since it turned out that one segfault was hunting the next and mod_include
became more and more unmaintainable, Brian Pane and I started (independently
:) a rewrite of the filter parser. While Brian used the old design and
external API (mostly for backwards compat) I've tried to redesign it from
the scratch. However, we've finally decided to take the newly designed
version.

The new code consists of a buffer centric parser (as opposed to a brigade
centric one). There's just one main loop that knows about buckets and the like
and passes most of the actual parser work to functions that deal only with
buffers. The output is handled similar to mod_bucketeer; we no longer insert
buckets somewhere in the currently parsed brigade, we just create a new one
and append all new buckets at the end. From time to time this one will be
passed down the filter chain. This results even in a simpler -- though
incompatible -- external API. In order to be able to backport the
change(binary compatible), there are just some lines wrapper code necessary
which emulate the old API.

The new code passes all tests, except one (modules/include/#36), where I'd
say, we should change the expectation ;-)
(1.3 throws nothing in this test, the current mod_include throws two error
messages and the new module throws just one error message).

I've uploaded all the code for review and test.

- plain code (for 2.1 HEAD)
      <http://cvs.apache.org/~nd/mod_include.c> (123K)
      <http://cvs.apache.org/~nd/mod_include.h> (5K)

- full patch against 2.1 (about 200K)
  It contains also patches of mod_cgi and mod_cgid (because of the modified
  API) and some other neat things like namespace protection of the public
  parts etc. - but don't worry, once I'm committing it, it'll we splitted into
  several commits.
      <http://cvs.apache.org/~nd/ssi-redesign21.patch>

- full patch against 2.0 (about 70K)
  This patches only mod_include.c and mod_include.h and only the parts that
  are necessary to get it working. It also contains the wrapper code for
  binary compat.
      <http://cvs.apache.org/~nd/ssi-redesign20.patch>

Please have a look at it. Opinions, comments etc are highly appreciated.
Thanks.

nd

Re: [PATCH] Review & Test: mod_include redesign

[Glenn Nielsen <gl...@mail.more.net>]

I am glad to see that work continues on  fixing bugs in mod_include. :-)

The only place I have had problems is on two of our production servers.
I will have to wait until the mod_include patch is considered ready for
release (production) in the 2.0 branch before I would consider upgrading
and doing a production test.

Regards,

Glenn

André Malo wrote:
> [CC to Ron Park and Glenn Nielsen to get this onto their radar]
> 
> Since it turned out that one segfault was hunting the next and mod_include
> became more and more unmaintainable, Brian Pane and I started (independently
> :) a rewrite of the filter parser. While Brian used the old design and
> external API (mostly for backwards compat) I've tried to redesign it from
> the scratch. However, we've finally decided to take the newly designed
> version.
> 
> The new code consists of a buffer centric parser (as opposed to a brigade
> centric one). There's just one main loop that knows about buckets and the like
> and passes most of the actual parser work to functions that deal only with
> buffers. The output is handled similar to mod_bucketeer; we no longer insert
> buckets somewhere in the currently parsed brigade, we just create a new one
> and append all new buckets at the end. From time to time this one will be
> passed down the filter chain. This results even in a simpler -- though
> incompatible -- external API. In order to be able to backport the
> change(binary compatible), there are just some lines wrapper code necessary
> which emulate the old API.
> 
> The new code passes all tests, except one (modules/include/#36), where I'd
> say, we should change the expectation ;-)
> (1.3 throws nothing in this test, the current mod_include throws two error
> messages and the new module throws just one error message).
> 
> I've uploaded all the code for review and test.
> 
> - plain code (for 2.1 HEAD)
>       <http://cvs.apache.org/~nd/mod_include.c> (123K)
>       <http://cvs.apache.org/~nd/mod_include.h> (5K)
> 
> - full patch against 2.1 (about 200K)
>   It contains also patches of mod_cgi and mod_cgid (because of the modified
>   API) and some other neat things like namespace protection of the public
>   parts etc. - but don't worry, once I'm committing it, it'll we splitted into
>   several commits.
>       <http://cvs.apache.org/~nd/ssi-redesign21.patch>
> 
> - full patch against 2.0 (about 70K)
>   This patches only mod_include.c and mod_include.h and only the parts that
>   are necessary to get it working. It also contains the wrapper code for
>   binary compat.
>       <http://cvs.apache.org/~nd/ssi-redesign20.patch>
> 
> Please have a look at it. Opinions, comments etc are highly appreciated.
> Thanks.
> 
> nd