On Fri, Mar 13, 2015 at 10:26:12AM -0700, Alan Irwin wrote: > On 2015-03-13 10:19-0000 Andrew Ross wrote: > > >[....] I've been running into further problems even after > >fixing the dependencies as you suggested. > > > >1) The cairo build fails (seems to be due to a bug in cairo build scripts > >exposed in gcc-4.9.0 - see http://sourceforge.net/p/mingw-w64/bugs/396/) > > > >2) The libharu build succeeds, but the hpdf_pdfa.h include file is not > >installed so the plplot build of the pdf driver fails. I don't have > >time to investigate further. Just installing the debian packages is a > >workaround. > > Hi Andrew: > > I will take a quick look at (1) and (2) above to see if there is > anything simple I can do today to fix these other epa_build issues > that you have found, but if they are difficult I will leave > them to post-release. > > > > >I did get Qt5 to build under epa (this is before your latest commit) > >and so I did a separate build, put pointing cmake to this epa version. > >This worked and the valgrind test results are below. I did not see any > >invalid read errors. I did get a warning about xbd though, and return > >did not seem to work to exit the plot page. I had to use the right > >mouse button instead. This may well be related to my pbuilder > >environment, though I didn't get the message with the debian package. > > Hmm. That means there is a real difference between the epa_built > Qt5.3.2 and Debian unstable Qt5.3.2 libraries which is presumably > caused by the Debian patches to Qt 5.3.2 or the compiler optimization > difference (the only real differences between the two results that I > am aware of). Also, I just double-checked, and I get the invalid read > for the epa_built Qt5.3.2 on Debian stable regardless of whether I use > return or right click to exit the plot. And if that invalid read > manages (by chance) not to trigger a segfault, both those actions exit > the plot page fine. When you combine these variety of > platform-dependent symptoms for 5.3.2 with previous terrible 5.3.0 > results reported by one of our users, and the on-going character > offset issues (which are mostly compensated by an empirical vertical > adjustment we apply for the Qt5 case), I think the inevitable > conclusion is that Qt5.3.x is still flaky/immature. Thus, I am going > to mention in the release notes that PLplot users should use the > -DPLPLOT_USE_QT5=ON option with a lot of caution and if they run into > any difficulties because of Qt5 flakiness, they should stick with Qt4 > instead. > > > > >Anyway, the key message is that even with an epa_build of Qt5 I do > >not see the error so it is not an intrinsic Qt5 / plplot problem, > >but something different between our setups is causing problems. > > Thanks very much for all your hard work to arrive at this important > conclusion (since memory management issues are a fundamental bane to > run-time repeatibility/stability) that the invalid read for Qt5 > appears to be limited to Debian stable and thus has nothing to do with > epa_build or PLplot issues.
Alan, If I have time (may be post release) I will repeat my tests with Debian stable to see whether it is the Debian stable / unstable that is the difference, or whether it is something about our specific hardware / build environment. Andrew ------------------------------------------------------------------------------ Dive into the World of Parallel Programming The Go Parallel Website, sponsored by Intel and developed in partnership with Slashdot Media, is your hub for all things parallel software development, from weekly thought leadership blogs to news, videos, case studies, tutorials and more. Take a look and join the conversation now. http://goparallel.sourceforge.net/ _______________________________________________ Plplot-devel mailing list Plplot-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/plplot-devel
