From Octave
Jump to navigation Jump to search





  • implement SURF, integralImage is cumsum(cumsum(a,1),2)
corner/cornermetric, harris

First post on mailing list in 12.01.2013



chain matrix multiplication

a=[1 2; 3 4];
b=[4 2; 8 1];
c=[2 3; 1 -1];
s{1}=a;s{2}=b; s{3}=c;

Tracking octave bugs with hg bisect

I had a strange problem when loading gzip compressed ascii files in octave. It failed dependent on the integer values. After some stripping I made a minimalistic test script (min_testcase_fails.m) which always fails in a current dev (88616c872933):

fid = fopen (fn,"w");
fprintf(fid, "%i %i %i %i\n",639, 25, 160, 978160);
fprintf(fid, "%i %i %i %i\n",687, 25, 171, 978160);
fprintf(fid, "%i %i %i %i\n",663, 31, 173, 978161);
fprintf(fid, "%i %i %i %i\n",663, 15, 154, 978161);
fprintf(fid, "%i %i %i %i\n",655, 21, 151, 978161);
## gzip it!
fn_gz=strcat(fn, ".gz");
cmd=cstrcat("gzip ",fn," -c > ",fn_gz)
c=load("file2.txt.gz")  #this fails in newer versions

When run:

$ octave -q min_testcase_fails.m 
fn = file2.txt
cmd = gzip file2.txt -c > file2.txt.gz
error: value on right hand side of assignment is undefined
error: called from:
error:   /home/andy/src/octave-bugs/min_testcase_fails.m at line 17, column 2

If line 4 with "663, 15, 154, 978161" is changed to ""663, 16, 154, 978161", load works as expected. Strange, isn't it? I also had an old build (dev ab1c6e6d1be6 from Sun Oct 28 21:48:02 2012) and "load" worked in this version.

Running bisect

I decided to make a fresh clone in a separate directory:

cd ~/src
hg clone octave-src
cd octave-src
hg bisect -b 88616c872933
hg bisect -g ab1c6e6d1be6

After this hg bisect uses a binary search strategy to test revisions:

Teste Änderungssatz 16428:f016a5342e19 (1380 Änderungssätze verbleiben, ~10 Tests)

So after checkout I had to compile the source tree (used ccache in the hope of faster checkout-compile cycles).

cd .. && mkdir octave-build && cd octave-build
export CXX="ccache g++"
export CC="ccache gcc"
make -j 7

After compilation I tried my test min_testcase_fails.m:

~/src/octave-build$ ./run-octave -q ../min_testcase_fails.m
fn = file2.txt
cmd = gzip file2.txt -c > file2.txt.gz
c =

      639       25      160   978160
      687       25      171   978160
      663       31      173   978161
      663       15      154   978161
      655       21      151   978161

So this is apparently a good one. Tell it hg bisect!

cd ../octave-src && hg bisect -g

After this you try make again, bootstrap && configure if make fails, rerun the testscript, tell hg bisect if it's good or bad and repeat this until the revision which introduced the problem is found. Or you can use "hg bisect --command", see next point.

hg bisect --command

I used Jordis bisect script for bug 32818 and modified it to my needs:

~/src$ cat 
## Try a simple build first
if  ! make -j 7
   cd $SRCDIR
   ./bootstrap || exit 127
   #rm -rf $BUILDDIR
   #mkdir $BUILDDIR
   $SRCDIR/configure || exit 127
   make -j 7|| exit 127
$BUILDDIR/run-octave -q ~/src/min_testcase_fails.m || exit 1
exit 0

After makin it executable, let hg bisect use it

~/src/octave-src$ hg bisect -c ../

Now sit back and relax, this could last some hours... Finally:

Änderungssatz 16554:03a28487fa9d: good
Die erste fehlerhafte Revision ist:
Änderung:        16555:04fb96f4bea1
Nutzer:          John W. Eaton <>
Datum:           Tue Apr 23 12:57:16 2013 -0400
Zusammenfassung: allow double-click in file browser to load data files