[Proj] Virtual File Access

Greg Troxel gdt at ir.bbn.com
Fri Jul 26 06:00:16 PDT 2013


Frank Warmerdam <warmerdam at pobox.com> writes:

Perhaps I am totally misunderstanding what you are doing.   Is this
about letting IO done on behalf of a user to read and write the user's
data (not grid shift files, which are part of the library) to come from
different places?  Or is it about grid shift files, but the goal is
because the user has them in RAM rather than on the filesystem?   I had
the impression that this was about a way to bundle the standard grid
shift/etc. files into the library so that one could have just a single
library file and not the ancilliary files, because managing mulitple
files was somehow too hard.

> Quite possibly there are many things broken about the environment I work
> in, but running things on windows is certainly not one of them.

I'm glad to hear that :-).

So what motivates you to add this code, rather than just installing the
files in the right place?  What is the essence of the problem?

> But the point I'd like to stress is that I believe it is generally a good
> practice for low level libraries to provide a way to hook IO so that
> applications can do a variety of things around file access.  Libraries like
> libtiff, GDAL, shapelib and libpng have done this for a long time.  I'm
> extending it to PROJ.4.

I find that notion unusual and surprising.  I was completely unaware of
those libraries adding what feels like a custom VFS layer.  I looked at
  http://www.libtiff.org/libtiff.html
and found no mention of it.

> I'm not clear exactly what information you are looking for.  Normal
> practice with applications installers using PROJ.4 would be to install the
> files somewhere in the file system, and to call pj_set_searchpath()  to
> look in the directory in which things were installed from their application
> initialization.  The pj_set_searchpath() and pj_set_finder() were intended
> to make finding the data files tractable in cases where applications didn't
> mind writing them to disk.

Where I'm coming from is that I install proj via pkgsrc (and am the
package maintainer).  So it's "pkg_add proj-4.8.0.tar.gz" and then the
library is in /usr/pkg/lib, and the files in /usr/pkg/share/proj (plus
programs in /usr/pkg/bin, man pages...).  I expect it's the same for all
other packaging systems, modulo prefix.  The path to the files is set
via PROJ_LIB at build time to $(pkgdatadir).  So I view it as a bug to
have to set search paths.  There is a single correct place for the grid
files, and they are part of the package (or a depending package if one
wants to split things).

Also, I'd suggest that you consider putting this facility on a branch
instead of in a release until there's both a useful extension
(presumably reading files that were crunched into the library) and
broader consensus that it's a good addition (on the benefit/complexity
tradeoff).

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: not available
URL: <http://lists.osgeo.org/pipermail/proj/attachments/20130726/4c421e01/attachment.sig>


More information about the Proj mailing list