R: cannot call igraph function in callbacks

[gaborcsardi]

BEcause it frees all memory allocated by igraph.

[gaborcsardi]

I guess there is not much we can do at this point, but 1) document that this is not allowed and 2) check the finally stack at the beginning of each call from R, and if it is not empty, then quit with an error.

[gaborcsardi]

Since we don't have a hook that is called each time an R->C call happens, maybe a possible solution is to redefine .Call within the package. Unfortunately this will have some overhead, both the .Call redefinition and the check for the FINALLY stack being empty.

[gaborcsardi]

Redefining .Call is quite some overhead, actually:

> time_that("Redefining .Call does not have much overhead #1", replications=10,
+           init = { library(igraph) ; g <- graph.ring(100) },
+           { for (i in 1:20000) {
+             base::.Call("R_igraph_vcount", g, PACKAGE = "igraph")
+           }  })
replications    user.self     sys.self      elapsed   user.child    sys.child 
     10.0000       0.1515       0.0004       0.1518       0.0000       0.0000 
> 
> time_that("Redefining .Call does not have much overhead #1", replications=10,
+           init = { library(igraph) ; g <- graph.ring(100) },
+           { for (i in 1:20000) {
+             igraph:::.Call("R_igraph_vcount", g, PACKAGE = "igraph")
+           }  })
replications    user.self     sys.self      elapsed   user.child    sys.child 
     10.0000       0.2692       0.0003       0.2693       0.0000       0.0000 

And this is even without the extra .Call to check the FINALLY stack. And this is with the proper check:

> time_that("Redefining .Call does not have much overhead #1", replications=10,
+           init = { library(igraph) ; g <- graph.ring(100) },
+           { for (i in 1:20000) {
+             igraph:::.Call("R_igraph_vcount", g, PACKAGE = "igraph")
+           }  })
replications    user.self     sys.self      elapsed   user.child    sys.child 
     10.0000       0.4127       0.0011       0.4137       0.0000       0.0000 

[gaborcsardi]

OK, redefined .Call for now, until there will be a better solution.

[ludwikbukowski]

is it fixed? I got segmentation faults in simple scripts:

library('igraph')
g <- make_tree(10)
#g<- set_vertex_attr(g, "xx", value = 10:19)
f.in <- function(graph, data, extra) {
print(graph)
FALSE
}
tmp <- bfs(g, root=1, "out",callback=f.in)

Remove if not related to this issue

[ntamas]

Most likely it is fixed; the latest R interface of igraph is version 1.1.2, released 2017-07-21, which is several years later than the date of the above commit. Please file a separate bug report for your issue in https://github.com/igraph/rigraph/issues

[ludwikbukowski]

Thank you for the response.
I just did it here: igraph/rigraph#253