User Feedback

Dean Dobranich - October 2006

24 October, 2006 John Greenfield and I met with Dean Dobranich today to demo our latest cut of ParaView 3. As expected, we ran into suprize problems. Here are the bugs we ran into as well as some feature requests.

• The Exodus reader did not give any status whatsoever when loading. For a while I thought we immediately ran into a problem that froze up ParaView, but we were really waiting for the load to finish. This is probably the same as bug 3426.
• In order to show the color bar, we had to pull up the "Edit Color Map" dialog box. This checkbox should really be on the GUI under the "Display" tab. It was not until I got back to my office when I realized that there is a toolbar button to toggle the color bar, too. The button says "SbV", which means nothing, but I presume this will be replaced with an icon soon. However, the toolbar toggle status is not updated when the color bar is enabled with the dialog box.
• Lots of times the rendering went all screwy. It predictably happened when enabling the color bar through the "Edit Color Map" dialog box. It also happened when selecting cells through the rubber band select. The 3D rendering artifacts polluted the surface with black streaks. It may have been Z-buffer fighting. It may also have been the position, normal, and connectivity array information getting mixed up to generate oblong polygons and/or incorrect normals. Another issue that happened at the same time was that the red border around the active view disappeared. I cannot replicate this behavior on my desktop. John predicts that this issue is caused by a 32-bit library accidentally linked into the 64-bit build.
• Dean has a feature request very important to him: the ability to select and cull an Exodus part by selecting it with the mouse. Dean visualizes Exodus data with, literally, hundreds of blocks. He has little notion of which blocks correspond to the parts on the screen. This feature could mean the difference between him using ParaView and him using something else.
• Selecting cells did not fill the element inspector. I cannot replicate this problem on my Windows desktop.
• There needs to be a way to add annotation that gives the current time. This is captured in bug 3058.
• I noticed that the image save option in the File menu is named "Save Screenshot." It should probably be named "Save View Image" to be consistent with ParaView 2.
• The "Save Screenshot" function failed to make a valid .jpg or .png (or at least, one that can be read by xv). It works fine on my Windows desktop (when read back in with IrfanView).
• When "Save Screenshot" finally did save a valid image (to a .bmp file), it captured the file browser dialog box.
• The block selection widget is not very friendly when trying to select a contiguous range of blocks (that is, do a shift click). I also noticed (on my Windows machine) that to select one thing you have to position right over the checkbox as opposed to clicking anywhere on the line. The same is true for the other Exodus reader lists (variables and node/side sets).
• It is not evident what the "visible" option means for 3D widgets.
• When a new filter is created, the new data does not have the same coloring as the input data (bug 3864).
• The threshold filter does not turn off the visibility of its input (bug 3896).

ParaView was able to run 45 minutes before it crashed. That was the only crash we encounted in about an hour of demoing.

Mike Wong - May 2006

8 June, 2006 I sat with Mike for about 1.5 hrs. today, and took him through the new ParaView. Overall it was a very positive session - we even loaded some of Mike's data (1.9 gigs) in addition to the teensy disk_out_ref dataset, and things worked great. Only one crash (see below), and it performed flawlessly for an hour after we restarted it.

Issues

• Need Linux build soon - lots of users have data on Linux systems.
• Need support for opening parallel files. The current implementation is incorrect - the application opens one Exodus reader for every file selected in the file dialogue.
• Cut/Clip filters:
  • 3404 Need on/off switch for widget.
    • We could show/hide the 3d widget if the associated properties in the object inspector have focus. --Mark Richardson 16:23, 9 Jun 2006 (EDT)
    • I think a checkbox is more clear. The users may never use the properties or the widget may not have any properties Berk 11:01, 13 Jun 2006 (EDT)
    • I agree in spirit with Mark - I'd like to do away with the checkbox, but I've done a lot of thinking and user-watching, and I don't see a simple, understandable way to do it. We need the checkbox for now.--Hollywood 13:28, 15 Jun 2006 (EDT)
  • 3406 When the user stops dragging the widget around, the transparent plane should disappear, like it does now in ParaView. This eliminates the annoying z-fighting that makes the cut surface shimmer.
  • 3409 Undo on a cut operation does not update the position of the plane to register with the cut.
• 3410 When adding a new filter, the new filter's current variable should be the same as the input.
• Lots of keyboard shortcuts needed:
  • Check current Shift, Ctrl, etc. so they match ParaView. Currently, they do not.
    • Are we talking about the interactor modifiers here? ParaView allows customization of these. Are we going to implement that this month? Berk 11:04, 13 Jun 2006 (EDT)
  • Pipeline Browser Shortcuts:
    • Ctrl-D brings up display properties.
      • Sounds good. I don't think ctrl+d is already used. I can add it this month. --Mark Richardson 16:23, 9 Jun 2006 (EDT)
  • Geometry View Shortcuts:
    • Left arrow, right arrow should go back/forward one timestep.
    • When a widget is selected, that widget should get the keyboard shortcut. An example where this makes sense is the Cut filter. When its cut function is a plane, pressing x, y, or z on the keyboard should make that axis the normal.
      • We have some keyboard focus issues to deal with here. Some shortcuts do not work unless the widget has focus. For example, Delete will not delete a source unless the pipeline browser has focus. Are all shortcuts going to be global (i.e. associated with the whole GUI)? Also, I don't think there is a concept of a "selected 3D widget". We would have to add that to VTK and that might require some work. Berk 11:07, 13 Jun 2006 (EDT)
  • Object inspector.
    • Hitting 'spacebar' should be the same as clicking the 'Accept' button. This removes the need for the user to click an item (such as a side set), then move the mouse all the way up to the 'Accept' button. It was excruciating to watch, believe me.
      • Hitting the spacebar only activates a button if it has focus. When the user is done entering properties they can tab to the accept button and hit the space bar. This should work already. To make it easier for the user, we could make the accept button get clicked when they hit the enter key. We could even give the button a shortcut (alt+a?). --Mark Richardson 16:23, 9 Jun 2006 (EDT)]
      • After some arm twisting, I added a shortcut (ctrl+enter) to accept in ParaView. I didn't want to do enter or space bar because Tk binds those to special things automatically. One problem was that when someone edits a text widget, the result may not get propagated unless the text widget looses focus (if there are any actions/signals bound to the widget, they may not get fired). Users tend to use Enter or Tab for accepting text entry values. Are there similar problems with ParaView/Qt? If we could get enter to work, users would be happy. Berk 11:13, 13 Jun 2006 (EDT)
  • 3407 Ctrl-Z = Undo, Ctrl-Shift-Z = Redo
• Toolbar for common view shortcuts. There is a bar in ParaView now that has common operations like 'Zoom to data', etc. We need this toolbar in PV3
• 3411 Feature: Recent files list.
• Exodus Obejct Inspector
  • Right mouse click on a variable should give 'Set to current' option, which sets the current variable of this Exodus reader to that variable.
  • Variable name should be expanded fully to the longest variable name.
  • Filename widgets can be hidden entirely, or accessible through an 'expert' mode.
    • I think we should hide the filename from all readers, even in expert mode. Some readers can change their output type based on file and if there are filters connected to them, it is possible to get errors even seg fault. There are ways of getting around this but they require some work. Berk 11:16, 13 Jun 2006 (EDT)
  • Variables should be orderable by each part of the data in the list box: loaded/unloaded checkbox, type (cell/point), name, range.
  • Variable list should be mulit-selectable. User should be able to do normal selection operations (box select, single add/remove from selected set) and then to On/Off and other appropriate operations from the right mouse menu.
  • Variable loading is not covered in undo/redo.
  • Double-clicking on a variable should check/uncheck its loaded checkbox.
    • These sound like a decent amount of work. Are we to do all of these this month? Berk 11:16, 13 Jun 2006 (EDT)
• Pipeline Browser
  • Right Mouse menu on an item should include 'Zoom to Data' menu item.
  • Should be able to 'Delete' any item - not just the leaf items. Deleteing an item is a choice between 'Delete this item' and 'Delete this and all childe items'
    • I agree. There will have to be a bit of work to turn off all the affected displays before deleting the object. We should be able to use the pipeline graph to do this. --Mark Richardson 16:23, 9 Jun 2006 (EDT)
  • Items should be renamable in the 'normal' windows way of clicking on them, then keeping the mouse over the item to make it editable in place.
    • I agree. I put this off since we probably want to re-register the object. That way, the name the user chose can be in the state file. The server manager model could watch for the re-registration and send a notification to update the name everywhere. --Mark Richardson 16:23, 9 Jun 2006 (EDT)
    • Sounds good. We have to be careful here though. There are observers on register/unregister, such as the pipeline browser. Let's add a ReRegister(oldName, newName) to the proxy manager that would launch a ReRegister event. This way, the pipeline browser can be notified of a name change instead of a register/unregister pair. This would make undo easier too. Berk 11:47, 13 Jun 2006 (EDT)
• Need the eyeball control in the Pipeline Browser to control visibility.
  • This should be done by the end of June. --Mark Richardson 16:23, 9 Jun 2006 (EDT)
• Need 'Loading ...' dialogue box when a file is loading.
  • Utkarsh is looking into getting progress to work in ParaView III. Berk 11:47, 13 Jun 2006 (EDT)

Would it be possible to have a ParaView icon (the three parallel bars) in the right hand side of the menu bar, like Internet Explorer does? The IE 'flag' waves when something's happening, indicating to the user that the application's alive. Does this make sense for us, and could we update an icon like that in a way that makes sense. For example, when the render is going on, or when something's loading. If we have the bottom status bar, like the web browser, the little icon makes sense to me. We don't want to go too far with the 'it's a browser' thing, but a few of the things like the waving icon and the status bar make sense to me. --Hollywood 17:10, 13 Jun 2006 (EDT)

Comments

• Mike does 'topology' operations, then 'variable' operations. Block, side set, and node set operations are generally done together, then variable operations separate. Doesn't generally mix the two, so potentially the widgets could be totally separate (on separate panels/tabs)
• Is there a way to make the side sets and node sets better? One option would be to make the side sets different colors, several could be on, and the user could see them. (we'd have to color code the icons, as well - that would be cool). Same with Node sets, which should be glyphed in some way.
• Material boundary issues MORE TO COME.
• We'll need actor control per window - the Pipeline Browser right mouse menu option 'Display Settings ...' will have to bring up the controls for the current item in the current window.
• After watching Mike work, I think we should move the 'Accept' and 'Reset' buttons to the bottom of the object inspector. They're exactly like the Accept/Cancel buttons in a modal dialogue box, and to me that means they belong at the bottom. Also, They're fairly distracting where they are - they break the natural relationship between the Pipeline Browser and the Object inspector (assuming they're laid out like we have them).

Crashes

• Crashed only once (in almost 1 hr. of continuous running - goldstar!). We were dragging the Pipeline Browser and the Obejct Inspector around, and something we did caused a crash. Can't duplicate it, but it should be noted.