I can also reproduce this, but this code has not been changed in several years. I consider graphviz_draw() to be largely deprecated (in favor of graph_draw()), since the graphviz interface is fairly buggy and inconvenient. I'm planning to remove this function altogether in new releases, and I'm not very inclined to debug it. Best, Tiago Am 17.08.20 um 17:44 schrieb Jeff Trull:

I can reproduce this.

On Mon, Aug 17, 2020 at 1:29 AM Rolf Sander <rolf.sander@mpic.de <mailto:rolf.sander@mpic.de>> wrote:

Hello,

I get a segmentation fault when trying to plot a loop with graphviz_draw(). Here is a minimal example to reproduce the error:

import graph_tool.all as gt g = gt.Graph() v0 = g.add_vertex() v1 = g.add_vertex() e01 = g.add_edge(v0,v1) e11 = g.add_edge(v1,v1) # loop gt.graphviz_draw(g)

I have version 2.33 of python3-graph-tool installed on my computer which runs on Linux Mint 19.2 Tina (based on Ubuntu 18.04 LTS, bionic). My graphviz version is 2.40.1-2.

It works fine if I don't add the loop e11 to the graph.

It works fine if I use gt.graph_draw() instead of gt.graphviz_draw()

It works fine if I use graphviz directly (via a .dot file).

It did work fine in earlier versions of graph_tool.

Can anyone reproduce this?

Best regards Rolf

-- -----------------------------------------------------------------------    Rolf Sander                              phone: [+49] 6131/305-4610    Max-Planck Institute of Chemistry        email: rolf.sander@mpic.de <mailto:rolf.sander@mpic.de>    PO Box 3060, 55020 Mainz, Germany     homepage: www.rolf-sander.net <http://www.rolf-sander.net> -----------------------------------------------------------------------                https://www.encyclopedia-of-geosciences.net              https://www.geoscientific-model-development.net ----------------------------------------------------------------------- _______________________________________________ graph-tool mailing list graph-tool@skewed.de <mailto:graph-tool@skewed.de> https://lists.skewed.de/mailman/listinfo/graph-tool

_______________________________________________ graph-tool mailing list graph-tool@skewed.de https://lists.skewed.de/mailman/listinfo/graph-tool

-- Tiago de Paula Peixoto <tiago@skewed.de>

Am 17.08.20 um 22:20 schrieb Jeff Trull:

I tracked this down and I believe the issue is related to the interpretation of the global variables Agdirected and Agundirected from libgvc. If you replace lines 75 and 76 of graphviz_draw.py from:

libgv_directed = libgv.Agdirected libgv_undirected = libgv.Agundirected 

to:

        libgv_directed = ctypes.c_int.in_dll(libgv, "Agdirected")         libgv_undirected = ctypes.c_int.in_dll(libgv, "Agundirected")

these values are interpreted correctly (as bit fields, not function pointers) and the test case works again.

Thanks for the fix, and the patch in gitlab, which has been merged. Best, Tiago -- Tiago de Paula Peixoto <tiago@skewed.de>

Am 18.08.20 um 12:38 schrieb Rolf Sander:

Tiago: I can understand very well that it is a lot of work to maintain the two similar functions graphviz_draw() and graph_draw() at the same time. Still, it would be a pity if graphviz_draw() disappears completely in future versions of graph-tool. I'm using graphviz_draw() a lot because it has some features that graph_draw() doesn't have. In particular, graphviz can put text nicely into oval-shaped vertices. Cairo, however, produces large circles that hide much of the graph. Using the vertex property "aspect" is very slow and does not improve the plots much. Examples are available here:

Indeed, we still need some form of non-overlapping heuristics in graph_draw(). I'll leave graphviz_draw() for the time being. Best, Tiago -- Tiago de Paula Peixoto <tiago@skewed.de>