Viewport Selector Segmentation Fault

   774   15   0
User Avatar
Member
138 posts
Joined: March 2015
Offline
Hi, I'm experiencing crashes with the viewport selector but before I submit a bug report, I'd like to see if it's repeatable for others. (To make sure it's not a driver issue or fixed in a Daily Build)

I've attached a hip file and the sequence for the crash is as follows.

1. Connect box2 to polybevel1 (instead of box1)
2. Select polybevel1 & click viewport selector
Crash (1996:Fatal Error: Segmentation fault)

The workaround is the following
1. Connect box2 to polybevel1 (instead of box1)
2. Select polybevel1, clear current selection, then click viewport selector.

This sample file is to represent a situation where the point order changes upstream and you need to change a selection downstream. If the selection is emptied before you click selector, it will work. If not, it will crash.

Houdini 16.5.323, Win10 & GTX1070 up-to-date.

Rob

Attachments:
Crash_Test.hiplc (61.7 KB)

User Avatar
Member
1101 posts
Joined: Sept. 2015
Offline
I don't get this problem.

Using 16.5.268
User Avatar
Member
138 posts
Joined: March 2015
Offline
Hi, thanks for your input. Before I start looking into my system/drivers, I just want to make sure I've been clear on the “viewport selector”. (I'm not sure what it's actually called) I've attached a screenshot of the button I'm referring to.

If it still works on yours, I'll have to look into what's causing it on mine.

If anyone on 16.5.323 can test too, it would be appreciated.

thanks
Rob
Edited by robsdesign - Jan. 1, 2018 12:52:04

Attachments:
Viewport_Selector.jpg (57.2 KB)

User Avatar
Member
1101 posts
Joined: Sept. 2015
Offline
Ah, ok…I did think you meant something else with ‘viewport’ selector.

And I do get the ‘crash’ as well. (16.5.268)
User Avatar
Member
102 posts
Joined: Oct. 2016
Offline
Hi, robsdesign, if it could be somehow helpful for you.
Crashes too for the first described situation, doesn't crash for the second one.
Writes > 10768: Fatal error: Segmentation fault
For somewhat I've installed 334 daily build. Trial to reinstall production build tells me it's already installed, so think I won't reinstall installed. Windows 10 too, haha understood I don't remember my gpu and don't know where to see it.
User Avatar
Member
130 posts
Joined: Feb. 2017
Offline
it makes sense to me…

An edge is not referred to as edge1, edge2, edge3….etc…rather an edge is denoted by a start point and an end point. So edge p2-6 means it is made up of point2 and point6.

so if you've made your edge selection, now you go back up and make changes to your geo, edge p2-6 may no longer exist because those 2 points may now be waaaaaay off on the opposite sides of your geo. (If you make ‘addition’ to your geo, say a simple extrude, then edge p2-6 would still be there AND in the same place…it would still work without first clearing the selection. If edge p2-6 still exists but is in a different place, I think that would be a worse problem than crashing coz now you ‘think’ it's worked fine…but really it hasn't)

In general computing terms, segmentation fault refers to you're trying to access something non-existent…ie. edge p2-6 no longer exists because of the changes you made

Nothing to do with graphics card, op system, version, tea leaves, sheep entrails…etc…etc..
Edited by vusta - Jan. 2, 2018 05:31:59
User Avatar
Member
138 posts
Joined: March 2015
Offline
Thanks for the help BabaJ & Ryu Ku, I've submitted a bug report and it has been accepted as Bug ID 87247.
User Avatar
Member
1101 posts
Joined: Sept. 2015
Offline
Did you read vustas' comment?

He's right.

If you look at your second box it is a polygon mesh rather than a polygon.

This changes the point number order of the second box compared to the first box.

The first box has a ‘real’ geometric connection (between point 2 and point 6) along the boxes corner.

In the second box the point numbers change with point 2 and point 6 diagonaly accross from each other along the top surface;

There is no geometric edge/line between these two points on box 2.

After reading vustas' comment yesterday I was going to add, but decided not too because I felt I might be getting too trivial, that…

I would agree that it is not a bug, but maybe could use an RFE.

Initially in your file you have the polybevel flagged and connected to box one, and everything is fine.

If you then connect it to box 2. Everything seems to go fine, because the bevel on box one is removed. However, you no longer have a bevel as expected with box 2 connected; You just get the box.

And at this point there is no error flags on the node.

I think (again, it's trivial), but for such a situation I think an error flag could be introduced in the code for that node if it encounters invalid point numbers (that indicate non-existent edges).

Would be helpfull for the less experienced.

Although if you manually re-type the point numbers in the bevel nodes ‘Group Type’ paramter text box from p2-6 to p0-2, you get the same as originally had with box 1.

It's just a bit ‘odd’ that, say even if you knew you had to change the points specified, you can't just use the ‘viewport selector’ and do that in the scene view - because it crashes, not letting you do so.

I've done some vex work where I create sets of points and use them in other parts of my scene. In those other parts of the scene where I am referencing those points, I sometimes mistakenly go out of range and my code trys to do geometry operations with non-existent points.

In those cases though I get error message flags on the nodes going back to the originating wrangle node that is not ‘supplying’ the ‘proper’ set of points.

Don't see why an error flag could not be introduced in here in the same way instead of letting it crash.
User Avatar
Member
138 posts
Joined: March 2015
Offline
Yes, I did read Vusta's comment but I don't need an explanation of what is happening, I can see it for myself as I created the sample file to show it. Also, I have no idea what tea leaves and sheep entrails have to do with this topic.

I used a polygon mesh for the 2nd box on purpose as it's a simple way to replicate the issue.

Anyway, it's already submitted and logged as a bug so I'll leave it up to SideFX to decide.
User Avatar
Member
130 posts
Joined: Feb. 2017
Offline
robsdesign
Also, I have no idea what tea leaves and sheep entrails have to do with this topic.

exactly…neither does which op system, version, nightly build, graphics card….etc…etc
User Avatar
Staff
4185 posts
Joined: July 2005
Offline
FWIW, crashes are always bugs (though some are 3rd party issues that can't be fixed by us). This looks like a legit bug though, it should handle garbage input properly.
User Avatar
Member
138 posts
Joined: March 2015
Offline
vusta
robsdesign
Also, I have no idea what tea leaves and sheep entrails have to do with this topic.

exactly…neither does which op system, version, nightly build, graphics card….etc…etc

Not sure about you but when I think I've found a bug, I search the forum first. When I did, I found plenty of references regarding segmentation faults that mentioned each of those things. Besides that, it's fairly common when diagnosing a bug to include this information.

Thanks for confirming twod, hopefully this thread can end now.
User Avatar
Member
130 posts
Joined: Feb. 2017
Offline
ah got it, I've got a measly 3 orange dots, you've got FIVE, therefore i'm not worthy of taking part in your intelligent threads, apologies for this post. Will never interrupt you again.
Edited by vusta - Jan. 3, 2018 04:38:03
User Avatar
Member
138 posts
Joined: March 2015
Offline
vusta
ah got it, I've got a measly 3 orange dots, you've got FIVE, therefore i'm not worthy of taking part in your intelligent threads, apologies for this post. Will never interrupt you again.

I wasn't implying that at all. I'm trying to say that my post has been answered, it's logged as a bug & now confirmed by a staff member. Can't we just leave it at that?
User Avatar
Member
229 posts
Joined: July 2005
Offline
I'm getting A LOT of polybevel seg fault crashes as well (16.5.323).
User Avatar
Member
138 posts
Joined: March 2015
Offline
Hi Stu, it should be fixed in build 16.5.349 but I'm waiting on the next production build so I haven't seen the fix myself.

Rob
  • Quick Links