[Libexif-devel] MakerNote corrupted... and output file smaller than the input

[Roberto C. <rob...@en...>]

Hello,
I read a message posted in this mailing list, in which a Canon user
claims that files processed by libexif shrink.
I notice the same behavior with the files produced by my camera (a
Canon, but I think what I'm going to say is more general), so I
investigated in order to understand what libexif leaves behind.

Padding bytes, which are not reproduced by libexif, may be an
explanation, but only a partial one. The real and most serious reason
why a file shrinks is that maker note data that is referenced within
MakerNote entry but is allocated outside is lost whenever libexif tries
to write the file!
Think about the format of an IFD: each entry whose data size is bigger
than 4 bytes is listed indeed within the IFD, but its contents is stored
outside!
Several camera manifacturers organize the MakerNote entry like a IFD,
sad but true.

What currently happens is that, at least for some cameras, libexif
silently corrupts MakerNote entry.
This unfortunately isn't libexif's fault (unfortunately, because
eventual libexif's faults can be corrected!), I see it as a flaw in
exif's specs: since nobody else is supposed to know what the maker note
means, it should have been enforced that its data do not spread beyond
its "visible" borders.

However, I think this can't be an excuse.
Only two behaviors are acceptable to my eyes for files written by
libexif:
a) libexif drops the entire MakerNote entry and warns the user that it
   has done so;
b) libexif successfully reproduces MakerNote entry, even without knowing
   what its contents means.
I don't like the first, so I'm trying to come out with an idea to get
the second. I already have something in mind.

Suppose that libexif:
-) builds a sort of map of the apparently unused regions of the exif
   section of the original file;
-) keeps all those regions unchanged in the output file (for unchanged I
   mean same values at the same offsets).
Shouldn't this work out?
Of course it does not if a maker note references datas external to
itself with relative offsets, but we must make some common-sense
assumptions. In order to preserve maker notes that use relative offsets
to external datas, we would need to make sure that the output-file
offset of the maker note itself is the same of the input file, which I
see it as a much harder constraint.

What do you think about it? Does it make sense?
I'm looking forward to hear your opinions about all that, mine are just
a bunch of ideas from someone who is relatively new with exif
information.

Cheers,
Roberto