Did you happen to look at the date on your calendar this morning?
:p
On 2015-04-01 11:17, Daniel Morissette wrote:
Good morning Even,
That sounds like an ambitious world (universe?) domination plan, but
before going too deep into the implementation details, can you comment
on the potential for real life widespread adoption of such a universal
format in a world where everyone and their dog wants to have their own
file format which is so much better than everybody else's?
In other words: GML failed in a similar attempt to replace all other
formats, so despite the benefits of BF over XML encoding, what would
give UFO more chances of succeeding?
Have a nice ... day.
Daniel
On 2015-04-01 8:16 AM, Even Rouault wrote:
Hi,
Since some time a few ideas came to my mind and I felt today was a
good one to
share them and get feedback.
Considering the never ending proliferation of GIS file formats,
currently 220
handled in GDAL trunk, it seems wise to put an end to it. Especially
since the
counter used to iterate over the drivers is a unsigned 8 bit, so we
will soon
be unable to add more, or at the expense of sacrificing our ports to
Intel 8008
or Motorola 6800, which would be pretty sad.
Therefore I'd like to propose the UFO format, which stands for Universal
Format Oh-yeaaah!
The basic idea of UFO is that it isn't a fixed format, but a varying
and self-
described one. XML (or perhaps EXI?) + XSD + XSLT + XPath +
Schematron could
probably do it, but for efficiency I thought to a byte-code
interpreted by
libgdal and whose interface with libgdal would match the GDAL driver
interface. So basically each dataset would contain its own driver.
The big
plus is that you could write image translators that would generate
binary
encodings optimized for the particular dataset being encoded: for
example, it
is kind of stupid to write the values of each pixel of a Mandelbrot
fractal
whereas its mathematical description fits into a few lines of code.
Furthermore, still pursuing with that example, we could even have
raster of
arbitrary resolution, since that's a characteristics of fractals. And
many GIS
datasets have indeed fractal charasterics, such as coastlines (
http://en.wikipedia.org/wiki/Coastline_paradox )
For security reason, we should aim at supporting only simple &
verifiable
languages, so Brainfuck (Brainf**k for the most puritans of us) seems
to be a
good fit : http://en.wikipedia.org/wiki/Brainfuck. Basically it is a
Turing
complete language with only 8 commands. So as much powerful as
needed, while
being very easy to learn and implement. To save some efforts, I'd humbly
suggest we adopt libbf ( http://savannah.nongnu.org/projects/libbf ),
an older
project of mine that also incorporates a on-the-fly optimizer &
compiler for
most popular architectures.
The plan would be to have an initial version of the UFO driver ready
for GDAL
2.0 and push strongly for its widespread adoption in all GIS, remote
sensing,
OSS & proprietary vendors, etc.... Perhaps we should establish a
dedicated
workgroup at OGC to make it a standard ? Then we could deprecate and
remove
all existing drivers and at the time of GDAL 3.0, UFO would be the
only one
remaining driver, making the Intel 8008 port very happy!
Happy to hear from your thoughts before formalizing that as a RFC,
Even
_______________________________________________
gdal-dev mailing list
gdal-dev@lists.osgeo.org
http://lists.osgeo.org/mailman/listinfo/gdal-dev
