Graphviz Bug Report 210

Number: 210
Title: agerror() is not called reliably during agread()
Submitter: wifflefan
Date: Thu Oct 17 12:55:53 2002
Subsys: Libagraph
Version: 1.8.10
System: x86-Linux-Redhat 8.0, Redhat's 2.4.18-10 kernel
Severity: minor
Problem:
When I attempt to agread() a truncated dot file two consecutive times, agerror() is called during the first agread(), but it is not called during the second agread().

The failure to call agerror() a second time can produce two results:

When I use my own agerror() function (suggested in North's "Agraph Tutorial", page 12), the program segfaults.
When I rely upon the libagraph agerror() function, the program does not segfault.

The source below will reproduce this error:


#include <stdio.h>
#include <agraph.h>

int main()
{
   FILE* file = fopen("error.dot", "r");
   Agraph_t *dotGraph = 0;

   dotGraph = agread(file, 0);

   printf("the first read is complete.\n");

   dotGraph = agread(file, (Agdisc_t*)NULL);

   printf("the second read is complete.\n");

   return 0;
}

void agerror(int code, char *str)
{
   printf("my agerror : error code %d : %s\n", code, str);
}



Input: 


digraph G {
        subgraph cluster_c0 {a0 -> a1 -> a2 -> a3;}
        subgraph cluster_c1 {b0 -


Comments: 

The dot input causing the bug is a truncated version of the cluster1.dot shipped with graphviz-1.8.10.

I chose an arbitrary place to truncate it.

The same behavior is shown with files that have some pretty ugly syntax errors.


Fix: 

Sorry, I don't have a fix, but $SRC_DIR/agraph/grammar.c lines 979-1028
appear to be the ones failing to call agerror() the seoncd time.  Maybe
that'll help a little bit?

[erg] This is not a bug, but is part of the I/O and parsing model of the
library and stdio.


Owner: erg

Status: Fixed (18 December 2002)