Re: [Plplot-devel] comprehensive testing results on Cygwin [POSSIBLE SOLUTION?]

[Alan W. I. <ir...@be...>]

On 2016-03-14 14:58-0700 Alan W. Irwin wrote:

> I am going to follow up by trying some valgrind runs on Linux for
> example 19 without shapelib and for both Tcl and C to see if valgrind
> can spot the fundamental memory management issue that is causing the
> segfault for your particular (Cygwin) combination of circumstances.

Hi Arjen:

In fact, for the -DHAVE_SHAPELIB:BOOL=OFF case where example 19
complains about shapelib being missing, if you build the test_c_psc and test_tcl_psc targets
on Linux (to build all required prerequisites) and then run

cd examples/tcl
env PATH=../../bindings/tcl:$PATH valgrind --num-callers=40 --trace-children=yes ./x19 -dev psc -o testt.psc

there are no memory management issues (0 errors) and some minor memory
leak issues.

Furthermore,

valgrind --num-callers=40 ../c/x19c -dev psc -o testc.psc

gives perfect valgrind results (0 errors, no leaks are possible) for
the C case, and the Tcl and C example PostScript results are identical for this
example.

So valgrind finds nothing seriously wrong with example 19 for
-DHAVE_SHAPELIB:BOOL=OFF for the Tcl and C cases on Linux.

However, I followed up further by mimicking what x19 does by hand.
That is, I ran the following:

../../bindings/tcl/pltcl -dev psc -o testt.psc

source x19.tcl
plinit
x19
plend

And after a lot of complaints about shapelib being missing I ran into
the following peculiar matrix operator message:

[...]
*** PLPLOT WARNING ***
plmaptex is a no-op because shapelib is not available.
No matrix operator named "62"

So from that last line I am pretty sure that the "62" in

$w cmd plmaptex {} "ss/ss64ne_General_Text" 1.0 0.0 0.5 "Caffyns\nHeanton\nDown" $minx $maxx $miny $maxy 62

in examples/tcl/x19.tcl is the cause of the above matrix operator
warning on Linux.

That warning message comes from line 382 of bindings/tcl/tclMatrix.c.
I notice that for all other error conditions that are handled
with Tcl_AppendResult, the returned value is TCL_ERROR, but in
this particular case the returned value is NULL.  Is that difference
in pattern significant?  In other words, if you changed that NULL
to TCL_ERROR would that solve the segfault on Cygwin?

In other words do we have two bugs here; one where some misuse of matrix
is causing the

No matrix operator named "62"

warning message, and then a second bug in the way that issue is handled?

Anyhow, I hope these speculations lead you to a definitive fix for
both bugs if it turns out you agree with me that there are actually
two bugs to deal with here, the NULL return value that likely should
be changed to TCL_ERROR to get rid of the segfault on Cygwin, and also
the actual condition that causes the above matrix operator message to
be emitted.

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
__________________________