Re: JFFS2 using 'private' zlib header (was [RFC] LZO de/compression support - take 6)

[Daniel Hazelton <dhazelton@enter.net>]

On Tuesday 29 May 2007 11:15:41 Satyam Sharma wrote:
> [ Trimmed Cc list; added original zlib authors. ]
>
> > On Tuesday 29 May 2007 09:33:51 Michael-Luke Jones wrote:
> > > On 29 May 2007, at 12:27, Satyam Sharma wrote:
> > > > Right, actually, zlib could be switched over to [using a common
> > > > directory].
> > > > Because zlib_deflate/ and zlib_inflate/ too share a private header
> > > > zutil.h which has unfortunately been stuck into include/linux/ with
> > > > a big
> > > > /* WARNING: this file should *not* be used by applications. */
> > > > comment ...
> > >
> > > Well, unfortunately zutil.h is currently being used in-kernel by fs/
> > > jffs2/compr_zlib.c despite the "WARNING: this file should *not* be
> > > used by applications" notice.
>
> Hmmm, either jffs2 thinks the zlib interfaces exposed through
> include/linux/zlib.h are insufficient, or it isn't using zlib properly
> (or at least the way it was supposed to be used :-)
>
> Looking at fs/jffs2/compr_zlib.c, however, it just seems to be
> an optimization -- skipping some checksum calculation if some
> flag (PRESET_DICT) is absent from the input stream about to
> be decompressed ...
>
> This /looks/ like an optimization that might make sense for other
> users of zlib too, in which case that entire check and
> skip-the-checksum thing could actually be put into zlib proper. (?)

This *is* a good idea. If someone doesn't beat me to it I'll make sure I've 
got the latest git and do up a patch that does this.

DRH

> > > So moving this header to a truly private location isn't possible
> > > right now, unfortunately,
> > >
> > > Michael-Luke Jones
> > > [added David Whitehouse to cc]
>
> On 5/29/07, Daniel Hazelton <dhazelton@enter.net> wrote:
> > I've looked at that code and it seems to need that file for one constant.
> > Perhaps it'd be better for jffs2/compr_zlib.c to define that constant
> > itself (or use it as a "Magic Number") rather than include the zlib
> > private header. Another possibility would be to move that constant out of
> > zutil.h and into zconf.h or zlih.b - doing any of those would allow the
> > zlib private header to be moved such that zlib could be changed to use a
> > common directory *and* have said private header in that directory.
>
> Moving PRESET_DICT to zlib.h robs lib/zlib_deflate/deflate.c from
> being able to resolve that macro. FWIW, a dirty hack could be to
> simply duplicate it in zlib.h. But then, implementation and interface
> are decoupled for good reasons ...
>
> Satyam