I have tagged the subject line with [Post-release] because I think it is too late for for any but the most obvious and simple changes in this release cycle, but I hope some of the memory management issues that I am discovering right now for the tk device driver will be addressed right away in the next release cycle.
I have discovered through the comprehensive tests of epa_build results that I just implemented that examples/c/x01c -dev tk (and presumably the rest of the C examples) segfault under a certain combination of circumstances. (1) Must be using the version of Tcl/Tk(8.6) built by epa_build. (2) Must have -DUSE_TCL_TK_STUBS=ON I know I did tests for this exact combination of circumstances in the very recent past (via the test_interactive target which definitely runs examples/c/x01c -dev tk and a lot of other tests) without a segfault. However, that is the nature of memory management issues; some times you get segfaults and sometimes you don't. When I ran valgrind on examples/c/x01c -dev tk for the above combination of circumstances it showed an invalid read followed by a segfault occurred at line 712 of bindings/tcl/tclAPI.c; Tcl_SetVar( interp, "dir", TCL_DIR, TCL_GLOBAL_ONLY ); The TCL_DIR macro has not been clobbered according to the debug fprintf( stderr, "adding %s to auto_path\n", TCL_DIR ); that occurs (now) just before that statement, and everything else about this statement is pretty simple and benign so I don't understand at all what is causing this segfault. But it is a virtual certainty it is caused by something being overwritten either in the heap or stack. Since -DUSE_TCL_TK_STUBS=ON is a new innovation that is not as well tested as our traditional Tcl, etc., results which did not use the stubs Tcl/Tk libraries, I have decided to work around this issue for the default case by adopting -DUSE_TCL_TK_STUBS=OFF for that case. I then looked carefully at the -DUSE_TCL_TK_STUBS=OFF case with a valgrind run on examples/c/x01c -dev tk. For this non-stubs case there were no obvious issues like invalid reads, but there were a fair number of fouled up memory allocations by the time the example was done as you can see from the attached valgrind report. Some of those (such as at least the first two) are definitely due to PLplot code and should be addressed by us post release just in case one of those is the culprit that indirectly causes the invalid read and segfault issues for the stubs case. Of course, the issue might also be that PLplot can reliably use the Tcl/Tk stubs libraries for Tcl/Tk version 8.5, but not 8.6 because of our extensive use of of Tcl/Tk library API which has been deprecated for many releases of Tcl/Tk and which therefore might not be reliably maintained any more by the Tcl/Tk developers (say as of version 8.6). @Arjen: To avoid such Tcl/Tk potential internal bit-rot issues, I would strongly encourage you (post-release) to establish some minimum Tcl/Tk version we support (8.5?) and review all our Tcl/Tk code and systematically replace Tcl/Tk API calls that were deprecated as of that version with the recommended replacements. That's worth doing in its own right and may have some impact on the memory management issues I have discovered. I am bringing up these tk device driver memory management issues because they are fresh in my mind, but I emphasize again they should only be tackled post-release. 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 __________________________
valgrind.out.gz
Description: compressed valgrind report
------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk
_______________________________________________ Plplot-devel mailing list Plplot-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/plplot-devel
