cliquer memory leaks

[jason]

See  http://ask.sagemath.org/question/1191/memory-blowup-2 for the initial report. Basically, Sage's interface to cliquer has a number of memory leaks:

1. in cliquer.pyx, graph_new is called several times without a corresponding graph_free.

2. The lists that are returned from the cliquer functions are not deallocated.

3. Inside the cliquer spkg, the cl.c file allocates lots of option structs through calls to sage_init_clique_opt(), but never deallocates them.

[jason]

Here are my changes to the cl.c file inside the cliquer spkg. Do you happen to have time to package these up, Nathann? I was a little confused since usually unmodified source is supposed to be in the src/ directory, but it looks like we have sage-specific stuff in the src/ directory in the cl.c file.

diff -Naur cliquer-1.2.p10/src/cl.c cliquer-1.2.p11/src/cl.c
--- cliquer-1.2.p10/src/cl.c	2010-02-15 22:26:56.000000000 -0600
+++ cliquer-1.2.p11/src/cl.c	2012-03-03 03:11:03.000000000 -0600
@@ -97,9 +97,11 @@
   number1=FALSE;
   set_t s;
   int i,l;
+  clique_options *opts = sage_init_clique_opt();
   s=clique_unweighted_find_single(g,min_weight,
 				  max_weight,maximal,
-				  sage_init_clique_opt());
+                                  opts);
+  free(opts);
 
   // Writing the answer into a int [] to be read by Sage
   int size=set_size(s);
@@ -132,9 +134,10 @@
   maximal=TRUE;
   number1=FALSE;
   int i,j,l;
-
+  clique_options *opts = sage_init_clique_opt();
   clique_unweighted_find_all(g,min_weight,max_weight,
-			     maximal,sage_init_clique_opt());
+			     maximal,opts);
+  free(opts);
 
   int size=set_size(clique_list[0]);
   *list=malloc(sizeof(int)*(size+1)*clique_count);
@@ -162,7 +165,9 @@
   number1=FALSE;
   clique_options *opts;
   opts=sage_init_clique_opt();
-  return clique_unweighted_max_weight(g,opts);
+  int result = clique_unweighted_max_weight(g,opts);
+  free(opts);
+  return result;
 }
 

[jason]

I'm not so sure about using sage_free to free memory allocated by cliquer's malloc. Could that cause problems?

Anyways, the attached patch and the above changes make a *huge* difference in memory usage when calculating cliques over a large range of graphs.

[dcoudert]

Apparently this patch is ready to be reviewed, but has still a needs_work status. I tried it and it is working correctly (install, tests, docbuild,...).

However, I don't know how to evaluate the memory improvements. Could you provide an example ?

[ncohen]

Duplicate of #12905

Nathann