[vtk-developers] Tcl rendering tests and hangs

[Goodwin Lawlor]

On Sat, Apr 14, 2012 at 5:32 PM, Marcus D. Hanwell <
marcus.hanwell at kitware.com> wrote:

> On Fri, Apr 13, 2012 at 7:31 PM, Goodwin Lawlor
> <goodwin.lawlor.lists at gmail.com> wrote:
> >
> >
> > On Fri, Apr 13, 2012 at 4:45 PM, Goodwin Lawlor
> > <goodwin.lawlor.lists at gmail.com> wrote:
> >>
> >>
> >>
> >> On Fri, Apr 13, 2012 at 4:13 PM, Marcus D. Hanwell
> >> <marcus.hanwell at kitware.com> wrote:
> >>>
> >>> On Fri, Apr 13, 2012 at 11:09 AM, Goodwin Lawlor
> >>> <goodwin.lawlor.lists at gmail.com> wrote:
> >>> >
> >>> >
> >>> > On Thu, Apr 12, 2012 at 6:35 PM, Marcus D. Hanwell
> >>> > <marcus.hanwell at kitware.com> wrote:
> >>> >>
> >>> >> On Wed, Apr 11, 2012 at 2:09 PM, Marcus D. Hanwell
> >>> >> <marcus.hanwell at kitware.com> wrote:
> >>> >> > Hi,
> >>> >> >
> >>> >> > I have a Gerrit topic up for review,
> >>> >> >
> >>> >> > http://review.source.kitware.com/#/t/374/
> >>> >> >
> >>> >> > This gets some of the rendering tests up and running again, but
> they
> >>> >> > hang until killed. Does anyone with deeper knowledge of Tcl and
> VTK
> >>> >> > have any idea what I might have missed. You can check out and
> build
> >>> >> > this topic, it depends on another topic that restores some Tcl
> code,
> >>> >> > and adjusts it for the new layout.
> >>> >> >
> >>> >> I have updated this topic, and have passing Tcl tests. The main
> point
> >>> >> I would love a little feedback on is why the additional require
> >>> >> package call was not necessary before modularization. I can confirm
> >>> >> that the image comparison is working, and I have to separate the
> tests
> >>> >> into two parts (Graphics and Rendering).
> >>> >>
> >>> >> Marcus
> >>> >
> >>> >
> >>> > Hi Marcus,
> >>> >
> >>> > Pre-modularization,  the vtk-tcl interpreter built with the vtk libs,
> >>> > initialized all the vtk commands on startup without a "package
> require
> >>> > vtk".
> >>> > That's the interpreter ctest uses to run rtImageTest.tcl.
> >>> >
> >>> > The question is, post-mod, what has changed that requires
> >>> > rtImageTest.tcl to
> >>> > have a "package require vtk".
> >>> >
> >>> > I'll have a look when I've got a new build.
> >>> >
> >>> Thanks for the explanation, I don't see where it was doing the package
> >>> require vtk before, and all the tests seem to have that as their first
> >>> line. For now we have merged the change, but I would welcome patches
> >>> if the old behavior is still desirable. We would also appreciate help
> >>> in restoring some of these tests if anyone still works with Tcl.
> >>>
> >>> Thanks,
> >>>
> >>> Marcus
> >>
> >>
> >> The "package require vtk" wasn't been done before for the interpreter
> >> built with the vtk libs - the vtk commands were been initialized by the
> >> interpreter itself (in c) - have a look at vtkTkAppInit.cxx (if you have
> >> time!)
> >>
> >> If you were running vtk-tcl scripts with the interpreter above, "package
> >> require vtk" was (almost) redundant - but if you used your
> own/activestate's
> >> interpreter and built dynamic vtk libs with tcl wrappers, then you
> needed
> >> "package require vtk".
> >>
> >> The merged change is fine - it's unlikely but someone may run the
> >> regression tests manually with a different interpreter and now they will
> >> work.
> >>
> >> I'll build from master today and have a look through the tcl stuff over
> >> the weekend. We still use vtk-tcl in our lab...
> >>
> >> Goodwin
> >>
> >
> > Just fyi, here's the diff that caused the above problems
> >
> >
> http://vtk.org/gitweb?p=VTK.git;a=blobdiff;f=Wrapping/Tcl/vtkTkAppInit.cxx;h=8a6001416a69cf1ede821bce50c53f627768cd80;hp=857af94651b9f5366ede9ca9f62dad14c58d3229;hb=0c1471f1da0ee8d76e9db0d95a0c434f0b38a7f5;hpb=cdd4d6fdec8b4b2c143224ca1d94699c96a8b58a
> >
> > The init code was chopped out of vtkTkAppInit.cxx since it reflected the
> old
> > structure, I guess.
> >
> > I think it's better off this way, anyway.
> >
> I made that change, thanks for pointing out the change. If people want
> to bring it back I will know where to add the code back in now. I feel
> like it is better, but if it breaks a lot of old code I could see some
> argument for bringing it back.
>
> Marcus
>

I discovered why the tcl interpreter was initialized like it was - it seems
to be for the benefit of static builds, to force the linkage of static vtk
libs.

Instead of re-instating the old code, it would be better to use the tcl
function "Tcl_StaticPackage" to link in the vtk static libs. This would
mean that the vtk commands only become available in a tcl interpreter after
a "package require vtk" regardless if the build is shared or static...
which is the way it should be.

BTW, the cmake variables:

VTK_USE_TK
VTK_TCL_TK_STATIC

were absent from the modular build. There were declared in the root
CMakeLists.txt file.

VTK_TK_RESOURCES_DIR

was no longer being set.

The Tk include directories weren't being included...

I've got a static build with Tk working now, so I'll push something to
gerrit to review.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/vtk-developers/attachments/20120414/749dda8a/attachment.html>