#620 "APPLY [ [] [unknown] ] []" leaks memory

[David Costanzo]

If APPLY attempts to run a "named slot template" instruction list and it has an unknown procedure, FMSLogo leaks memory.

I have reproduced this as far back as FMSLogo 6.11.0. In FMSLogo 6.10.1, the leak is not reproducible. A leak is reproducible in MSWLogo 6.5b, but I suspect that's a different memory leak that was fixed before FMSLogo 6.10.0.

How Reproducible:
Every Time

Steps to Reproduce:

SHOW NODES
REPEAT 1000 [ CATCH "error (LIST "APPLY (LIST [] (LIST "unknown) ) [] ) ]
SHOW NODES

What Happens:
The first item in the second NODES printout is several thousand larger than the first item in the first NODES printout

Expected Result:
The first item in the second NODES printout is about the same (within a hundred) of the first item in the first NODES printout.

[David Costanzo]

The bug is not limited to an unknown procedure, it seems like any "named slot template" instruction will cause it. For example, this also reproduces the leak:

SHOW NODES
REPEAT 1000 [ APPLY (LIST [] (LIST "IGNORE 1) ) [] ]
SHOW NODES

At first glance of the leaked nodes, it looks the node representing the anonymous function is leaked because the "treeification" creates a circular reference that is only broken if the node is explicitly untreeified.

In looking at CHANGELOG.TXT for version 6.11.0, I don't see anything that could have introduced this.

[David Costanzo]

This is fixed by combination of [r6323], [r6326], [r6332], and [r6343]. The fix will be available in FMSLogo 8.6.0.

The problem was a cycle that was created in the treeification of a procedure body. Each line in the bodylist became a TREE, which is used to get the parse tree of the line. In turn, the first expression each TREE was made into a LINE, which has a reference back to the bodylist line. This was introduced in 6.11.0 by [r1570] in a fix for Bug #126 that made FMSLogo look more like UCBLogo. This change removed code from treeify_bodylist() that untreeified the line after treeifying it. The correponding code is commented in out in make_tree_from_body() in paren.c in UCBLogo, which is why I removed it from FMSLogo:

        if (generation__tree(car(body_ptr)) == UNBOUND)
        setgeneration__tree(body, UNBOUND);
/*      untreeify(car(body_ptr));   */
        while (cdr(tree) != NIL)
        tree = cdr(tree);
        end_ptr = tree;

The circular reference is not a problem in UCBLogo because it has a mark-and-sweep garbage collector. It's a problem in FMSLogo because it uses reference counting and so the circular reference must be broken by an explicit untreeify call. I couldn't find a place to untreeify anonymous procedures, so I thought it was better not to have a cycle in the first place. Overall, this made the code base simpler because it made the structure of a treeified procedure simpler and removed the need to untreeify when procedures are cleaned up.

It seems silly to treeify the line, then untreeify it, so I reworked the treeification process to be able to create the tree of a line without making it a TREE.