[GRASS-dev] Re: [GRASS GIS] #718: r.li forgets mask/illegal filename

[GRASS GIS]

#718: r.li forgets mask/illegal filename
------------------------+---------------------------------------------------
  Reporter:  kyngchaos  |       Owner:  grass-dev at lists.osgeo.org
      Type:  defect     |      Status:  new                      
  Priority:  normal     |   Milestone:  6.4.0                    
 Component:  Raster     |     Version:  6.4.0 RCs                
Resolution:             |    Keywords:  r.li                     
  Platform:  All        |         Cpu:  All                      
------------------------+---------------------------------------------------
Comment (by hamish):

 ok, I can reproduce this now. The conf file above doesn't trigger it
 because it sets up a moving window which covers the whole map. You need to
 use r.li.setup to make a subsampled window to have it happen. (???)

 NC sample dataset / 'g.region rast=landclass96'
 ~/.r.li/history/landclass96_conf2
 {{{
 SAMPLINGFRAME 0.311787072243|0.18776371308|0.341772151899|0.376425855513
 SAMPLEAREA 0.366920152091|0.259493670886|0.0443037974684|0.0361216730038
 SAMPLEAREA 0.587452471483|0.42194092827|0.0443037974684|0.0437262357414
 SAMPLEAREA 0.32319391635|0.42194092827|0.0886075949367|0.041825095057
 }}}

 causing
 {{{
 G65> r.li.shannon map=landclass96 conf=landclass96_conf2
 output=landclass96_shannon --o --v
 *** glibc detected *** malloc(): memory corruption: 0x09166940 ***
 D0/0: daemon while loop: toReceive.type=-1208818294
 Illegal filename. Character <   > not allowed.
 Illegal filename. Character <   > not allowed.
 Illegal filename. Character <   > not allowed.
 WARNING: Unable to open header file for raster map < @(null)>
 CHILD[pid = 20559]: unable to open < > mask ... continuing without!
 Aborted
 }}}

 (I added that debug msg locally)


 code updated in 6.5svn and 7svn to fail more gracefully if the type value
 is garbage. (in r38757 and r38758)


 hmmph but now that I try gdb that memory has something else in it which
 doesn't trigger the breakage. I'll have to come back to it later.


 ----

 I suspect there may be some bad inefficency in that worker.c receive()
 while loop, perhaps explaining the 50% of proc time spent in the kernel
 for at least the moving window conf file (found in comment:3).

 perhaps the alloc()s within each iteration of the loop? especially for the
 moving window mode which will cycle through the loop many, many times. And
 no G_free()?

 ----

 also I notice that the ~/.r.li/output/conf_filename file is created with
 the exe bit set for some reason?!:
 {{{
 daemon.c:138:   sprintf(out, "%s/.r.li/output/%s", getenv("HOME"),
 output);
 daemon.c-139:   res = open(out, O_WRONLY | O_CREAT | O_TRUNC, 0755);
 }}}

 should that be 0644 or better yet just left to the system? (how, by just
 removing the 0755?)


 Hamish

-- 
Ticket URL: <https://trac.osgeo.org/grass/ticket/718#comment:15>
GRASS GIS <http://grass.osgeo.org>