[gdal-dev] gdal-autotest2

[Even Rouault]

On lundi 20 février 2017 10:56:16 CET Kurt Schwehr wrote:
> Only 370K more LOC left to cover with proper library level unittests.

Changing the topic line.

Kurt

IHMO if your aim is to have gdal-autotest2 being merged some day, the forked approach is 
going to be a nightmare for you to manage, unless you have a huge workforce to put behind. 
I'd rather attempt at having the current autotest suite running as a hack into the cleaner 
approach, so it can be merged into trunk and once done, migrate things progressively.

Even

> 
> On Mon, Feb 20, 2017 at 10:01 AM, Aaron Boxer <boxerab at gmail.com> wrote:
> > On Sat, Feb 18, 2017 at 8:14 AM, Kurt Schwehr <schwehr at gmail.com> wrote:
> >> Well,  I made a start as a part of gdal-autotest2...
> >> 
> >> 
> >> https://github.com/schwehr/gdal-autotest2/blob/master/cpp/
> >> third_party/kakadu/coresys/common/kdu_arch_test.cc
> > 
> > That is a good start.
> > 
> >> On Sat, Feb 18, 2017 at 5:03 AM, Aaron Boxer <boxerab at gmail.com> wrote:
> >>> ---------- Forwarded message ----------
> >>> From: "Aaron Boxer" <boxerab at gmail.com>
> >>> Date: Feb 18, 2017 8:03 AM
> >>> Subject: Re: [gdal-dev] GDAL unable to write 4-band RGBA jp2 file with
> >>> Kakadu 7.8
> >>> To: "Kurt Schwehr" <schwehr at gmail.com>
> >>> Cc:
> >>> 
> >>> 
> >>> 
> >>> On Feb 18, 2017 7:26 AM, "Kurt Schwehr" <schwehr at gmail.com> wrote:
> >>> 
> >>> Jason,
> >>> 
> >>> Turns out you are hitting a known bug in v7.8 that is fixed in v7.9.
> >>> From the kakadu v7.9 docs:
> >>> 
> >>> h. Corrected a tiny, yet very significant bug introduced into the
> >>> 
> >>>    handling of opacity channel descriptions in version 7.8.  The
> >>>    correction is in function `jp2_channels::set_opacity_mapping', where
> >>>    the opacity channel's component index accidentally overwrote the
> >>>    colour channel's component index.
> >>> 
> >>> Apparently Kakadu does no regression testing before a release 😁
> >>> 
> >>> 
> >>> 
> >>> It's pretty easy to backport the fix if you are currently stuck at 7.8.
> >>> It's just a change of a 0 to a 1.
> >>> 
> >>> --- kakadu/v7_8/apps/jp2/jp2.cpp
> >>> +++ kakadu/v7_9/apps/jp2/jp2.cpp
> >>> @@ -5976,7 +5976,7 @@
> >>> 
> >>>    if (lut_idx < 0)
> >>>    
> >>>      lut_idx = -1; // For consistency in comparisons later on.
> >>>    
> >>>    j2_channels::j2_channel *cp = state->channels+colour_idx;
> >>> 
> >>> -  cp->component_idx[0] = codestream_component;
> >>> +  cp->component_idx[1] = codestream_component; /* Was a bug in version
> >>> 7.8 */
> >>> 
> >>>    cp->lut_idx[1] = lut_idx;
> >>>    cp->codestream_idx[1] = codestream_idx;
> >>>    cp->data_format[1] = data_format;
> >>> 
> >>> The 0 can trigger an error that looks just like you are seeing:
> >>> 
> >>> Error: GdalIO: Error in Kakadu File Format Support: Attempting to create
> >>> a Component Mapping (cmap) box, one of whose channels refers to a
> >>> non-existent image component or palette lookup table. (code = 1)
> >>> 
> >>> On Mon, Feb 13, 2017 at 9:28 PM, Kurt Schwehr <schwehr at gmail.com> wrote:
> >>>> Hi Jason,
> >>>> 
> >>>> I've seen the same thing with an alpha channel.  I'm not sure what's
> >>>> up.  Using just RGB was successful.
> >>>> 
> >>>> On 7.3, I'm also seeing problems when using more than the main thread
> >>>> on linux.
> >>>> 
> >>>> With 7.3, I just noticed today that kdu_get_num_processors() only
> >>>> returns 0 on linux as kdu_arch.cpp is missing #include <unistd.h> so
> >>>> _SC_NPROCESSORS_ONLN and _SC_NPROCESSORS_CONF are undefined.  When I
> >>>> enabled _SC_NPROCESSORS_ONLN, I got kdu_get_num_processors() returning
> >>>> 74
> >>>> on what I thought was a 6 core (aka 12 hyperthread) box.   If I change
> >>>> the
> >>>> fall through from 0 to 2, I get all sorts of trouble from internal
> >>>> kakadu
> >>>> assertions and TSAN failures.
> >>>> 
> >>>> So I clearly have more investigating to do.
> >>>> 
> >>>> I've yet to pass along any of what I've found to Taubman and I haven't
> >>>> yet pushed any of my new tests to github, but I hope to do both soon.
> >>>> 
> >>>> And I've yet to see what, if any, discussion is happening in the yahoo
> >>>> group.
> >>>> 
> >>>> -kurt
> >>>> 
> >>>> On Mon, Feb 13, 2017 at 8:14 PM, jason.liu <jason.liu at spookfish.com>
> >>>> 
> >>>> wrote:
> >>>>> Hi gdal-dev,
> >>>>> 
> >>>>> I am having trouble using GDAL to write 4-band RGBA jp2 file with
> >>>>> Kakadu
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/gdal-dev/attachments/20170220/fb3e5548/attachment-0001.html>