5 GC Mutator Language, 2

The GC Mutator Scheme language is used to test garbage collectors written with the GC Collector Scheme language. Since collectors support a subset of Racket’s values, the GC Mutator Scheme language supports a subset of procedures and syntax. In addition, many procedures that can be written in the mutator are omitted as they make good test cases. Therefore, the mutator language provides only primitive procedures, such as +, cons, etc.

5.1 Building Mutators for GC2

The first expression of a mutator must be:

The rest of a mutator module is a sequence of definitions, expressions and test cases. The GC Mutator Scheme language transforms these definitions and statements to use the collector specified in allocator-setup. In particular, many of the primitive forms, such as cons map directly to procedures such as gc:cons, written in the collector.

5.2 Mutator API for GC2

The GC Mutator Scheme language supports the following procedures and syntactic forms:

Calls to this function are allowed only in syntactic positions that would discard its result, e.g., at the top-level or inside a begin expression (but not in the last expression in a begin). Also, this function appear only in the function position of an application expression.

So, in order to pass around a version of this function, you must write something like this (λ (c v) (begin (set-first! c v) 42)), perhaps picking a different value than 42 as the result.

Other common procedures are left undefined as they can be defined in terms of the primitives and may be used to test collectors.

Additional procedures from scheme may be imported with:

For example, the GC Mutator Scheme language does not define modulo:

(import-primitives modulo)

(test/value=? (modulo 5 3) 2)

5.3 Testing Mutators with GC2

GC Mutator Scheme provides two forms for testing mutators:

5.4 Generating Random Mutators for GC2

This PLAI library provides a facility for generating random mutators, in order to test your garbage collection implementation.

The mutator is created by first making a random graph whose nodes either have no outgoing edges, two outgoing edges, or some random number of outgoing edges and then picking a random path in the graph that ends at one of the nodes with no edges.

This graph and path are then turned into a PLAI program by creating a let expression that binds one variable per node in the graph. If the node has no outgoing edges, it is bound to a heap-value?. If the node has two outgoing edges, it is bound to a pair and the two edges are put into the first and rest fields. Otherwise, the node is represented as a procedure that accepts an integer index and returns the destination node of the corresponding edge.

Once the let expression has been created, the program creates a bunch of garbage and then traverses the graph, according to the randomly created path. If the result of the path is the expected heap value, the program does this again, up to iterations times. If the result of the path is not the expected heap value, the program terminates with an error.

If input is a port, its contents are assumed to be a well-formed PLAI program. If input is a file, the contents of the file are used.