To Andrew and Orion: I am handing off the octave-4 segfault issue that is described below to you guys because I have run out of ideas and you both have more octave/swig knowledge than I do. There is a small bit at the end where I solved all the valgrind reporting problems that were stopping valgrind analysis, but the extremely simple run-time test of our octave4 bindings still segfaults (presumably because the octave API has changed significantly between octave 3 and octave 4), and I don't know where to start to fix it because the full valgrind report (attached as plplot_valgrind.out.gz) still isn't helping me very much. But maybe one of you can figure it out or come up with an additional octave option to make the valgrind results more useful for diagnosing whatever the problem is that is causing the segfault.
More below. On 2015-07-14 00:28-0700 Alan W. Irwin wrote:
On 2015-07-11 19:27-0700 Alan W. Irwin wrote:[...Y]our build success motivates me to expand the epa_build implementation to see how far I can get with an epa_build of Octave-4.0.0 and patched versions of swig-2 and swig-3. Assuming I can epa_build all of those, then that would allow me to do the above run-time check here for patched versions of both swig-2 and swig-3, and thus help build the case to get your swig patch (or two variants of that if your patch has to be modified for swig-2) into the next official releases of both swig-2 and swig-3.Hi Orion: The current status of this effort is I am done with all the relevant epa_build configuration, but the result segfaults at run time. Here is what I have done so far. 1. Implemented an epa_build of lapack and blas. ctest of that build passed 100 per cent of all the many tests. 2. Upgraded the epa_build of swig-2.0.11 to swig-3.0.6 with your patch applied. The result works well with PLplot for the non-octave case. For example, the swig-generated Python, Lua, and Java bindings all work well. 3. Implemented an epa_build of octave_lite which is octave-4.0.0 with all optional components disabled. This version passes the simple test octave:1> x=-pi:0.1:pi; octave:2> plot(x,sin(x)) octave:3> exit with the plot done with the default gnuplot. However, valgrind of that simple test loses control of octave_lite (see below). 4. epa_built plplot_lite against this octave-lite version with no issues. 5. However, when I try the above simple test in a directory which contains the following file (to enable the PLplot version of plotting) wine@raven> cat .octaverc warning("off","Octave:shadowed-function"); addpath("/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/build_plplot_lite/bindings/octave/PLplot","/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/build_plplot_lite/bindings/octave/PLplot/support","/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/build_plplot_lite/bindings/octave/misc","/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/build_plplot_lite/examples/octave","/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/comprehensive_test_disposeable/shared/build_tree/bindings/octave","/home/wine/newstart/build_script/build_dir-linux/epa_build/Source/comprehensive_test_disposeable/shared/build_tree/bindings/octave/PLplot"); global pl_automatic_replot pl_automatic_replot=1; where the form of this file has been suggested by bindings/octave/USAGE the above simple plot test segfaults (where normally with PLplot built against octave-3.x.y, this simple test generates a plot using the xwin device). I would like to do further debugging of this segfault with valgrind, but somehow valgrind is losing control of octave_lite so never generates any messages after the normal burst of initial messages and the commands go the same speed as usual (not the slow speed you normally get with valgrind) regardless of whether I use gnuplot (without segfault) or the .octaverc trick above to attempt to use plplot (with segfault).
That issue had a simple solution; for valgrind to trace octave4 requires the valgrind option --trace-children=yes. However, that in turn shows a very extensive call stack where the number of callers is much larger than the typical limit of 50 for --num-callers used for older valgrind versions like provided by Debian wheezy. Thus, to access a recent valgrind version that allows --num-callers=500 at maximum, I epa_built valgrind-3.10.1 (as opposed to Debian wheezy's 3.7.0). And I generated valgrind results with the following (wrapped) command: /home/wine/newstart/build_script/install-linux_buildtools/bin/valgrind --num-callers=500 --trace-children=yes octave 2>| plplot_valgrind.out where I did the above simple test of the epa_built octave-4.0.0 in the directory where the above .octaverc existed. The segfault occurs and valgrind (as you can see from the attached plplot_valgrind.out.gz) detects a couple of invalid reads that presumably caused that segfault. If I do exactly the same test (except for output file named gnuplot_valgrind.out in a directory without that .octaverc, then no segfault occurs, gnuplot plots the plot, and the valgrind result (attached as gnuplot_valgrind.out.gz) is clean However, I cannot proceed further than that because plplot_valgrind.out.gz appears to contain no references to PLplot at all. So I am completely stuck and handing off to you guys. In any case, getting PLplot ready for octave4 is obviously post-release material so my plan for the forthcoming release of 5.11.1 is the -DTRY_OCTAVE4=ON cmake option will remain highly experimental, and unless the user sets that option our build system will either find and use octave3 or disable octave altogether. Alan __________________________ Alan W. Irwin Astronomical research affiliation with Department of Physics and Astronomy, University of Victoria (astrowww.phys.uvic.ca). Programming affiliations with the FreeEOS equation-of-state implementation for stellar interiors (freeeos.sf.net); the Time Ephemerides project (timeephem.sf.net); PLplot scientific plotting software package (plplot.sf.net); the libLASi project (unifont.org/lasi); the Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project (lbproject.sf.net). __________________________ Linux-powered Science __________________________
plplot_valgrind.out.gz
Description: valgrind output for simple test of octave4 with PLplot
gnuplot_valgrind.out.gz
Description: valgrind output for simple test of octave4 with gnuplot
------------------------------------------------------------------------------
_______________________________________________ Plplot-devel mailing list Plplot-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/plplot-devel
