On Tue, Aug 14, 2012 at 3:21 PM, Tamás Nepusz
<address@hidden> wrote:
> Thanks! In the meantime, since I can use igraph to write the correct format, and pipe it to sys.stdout, what would be the best hack for me to pipe stdout to a file? It feels like another line of code could get the functionality I need for the moment.
Well, it's a bit complicated. Since igraph's core is written in C and the functions that export a graph into a file require a FILE* object in the low-level C layer, it is not possible to use the standard Python trick, which would involve swapping sys.stdout with a file object created with the open() function and then swapping them back when igraph has saved everything. The only reason why sys.stdout works is because there is a standard protocol which tells igraph how to 'extract' a file handle from sys.stdout that we can pass on to igraph_write_graph_pajek in the C layer. The whole magic is in the igraphmodule_filehandle_init function in fileobject.c of the Python interface source code, and to be honest, I have no idea yet why it does not work on RHEL. The function essentially has two code paths: one that handles plain file names passed in strings, and one that handles Python's file-like objects such as sys.stdout. You have shown already that plain file names do not work for you, and
it was the same with passing in an object that was created with Python's open() function. Theoretically, sys.stdout is no different from any other file object created with open(), so that's why I think that this is pretty weird. The only difference I can think of right now is that sys.stdout auto-flushes the output to the screen after every newline character, while Python file objects do not necessarily do that.
Can you please try this code snippet:
g.write_pajek(f)
f.flush()
f.close()
--
T.