Graphviz Issue Tracker
Mantis Bug Tracker

View Issue Details Jump to Notes ] Issue History ] Print ]
IDProjectCategoryView StatusDate SubmittedLast Update
0001674graphvizNeatopublic2009-07-24 13:552011-04-28 04:03
ReporterOleg 
Assigned Togviz 
PrioritynormalSeveritycriticalReproducibilityalways
StatusacknowledgedResolutionopen 
PlatformOS*-*-*OS Version
Summary0001674: Objects placement bug
Description



input: (neato.exe 111.txt)
<CD>
graph { node [shape=box]; overlap=false;
NG1218 [width=1.547935, height=0.694444];
NG1222 [width=1.856385, height=0.694444];
NG1218 -- NG1222;
NG1226 [width=2.856385, height=0.694444];
NG1218 -- NG1226;
NG1230 [width=3.547935, height=0.694444];
NG1226 -- NG1230;
}
</CD>



output:



<CD>
graph 1 9.8802 0.72222
node NG1218 3.2175 0.36111 1.5417 0.69444 NG1218 solid box black lightgrey
node NG1222 -0.93219 0.36111 1.8611 0.69444 NG1222 solid box black lightgrey
node NG1226 -1.#IND -1.#IND 2.8611 0.69444 NG1226 solid box black lightgrey
node NG1230 8.0955 0.36111 3.5417 0.69444 NG1230 solid box black lightgrey
edge NG1218 NG1222 4 2.4515 0.36111 1.7595 0.36111 0.73917 0.36111 0 0.36111 solid black
edge NG1218 NG1226 4 -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND solid black
edge NG1226 NG1230 4 -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND -1.#IND solid black
stop
</CD>






What this: "-1.#IND"???
Steps To Reproduce

graph { node [shape=box]; overlap=false;
NG1218 [width=1.547935, height=0.694444];
NG1222 [width=1.856385, height=0.694444];
NG1218 -- NG1222;
NG1226 [width=2.856385, height=0.694444];
NG1218 -- NG1226;
NG1230 [width=3.547935, height=0.694444];
NG1226 -- NG1230;
}
Additional Information

[north]
Maybe we should just drop the Windows version!
We have so many problems with this particular platform.

It is well known that #IND means "indeterminate" (a.k.a. NAN in
other operating systems) so there is some (apparently new)
floating point discrepancy in how the Windows executables
do something. e.g. #IND can be an overflow (divide by 0).

[erg] This problem is arising from some obscure bug in our Voronoi code used for overlap removal. It goes away if you
even use a different initial layout using the start attribute.

We really don't recommend Voronoi overlap removal algorithm any more. Even though it usually works, the final
layout is not as satisfactory as other algorithms. Especially with 2.24, we recommend setting overlap=prism.

We should also think about possibly making overlap=false correspond to prism. We like to keep things backward
compatible, but this is now a case where the "default" value is not a good choice.

TagsNo tags attached.
AUXILLARY-FILES
DATE-FIXED
FIX-COMMENT
FORMER-ID1725
INPUT-FILE
OUTPUT-FILE
STATUS-COMMENT*
VERSION     2.24
Attached Files

- Relationships

-  Notes
There are no notes attached to this issue.

- Issue History
Date Modified Username Field Change
2011-04-28 04:03 user1 New Issue
2011-04-28 04:03 user1 Assigned To => user1


MantisBT 1.2.5[^]
Copyright © 2000 - 2011 MantisBT Group
Powered by Mantis Bugtracker