This section describes the details on how instruction sequences are represented and combined. Recall from section 5.5.1 that an instruction sequence is represented as a list of the registers needed, the registers modified, and the actual instructions. We will also consider a label (symbol)(string) to be a degenerate case of an instruction sequence, which doesn't need or modify any registers. So to determine the registers needed and modified by instruction sequences we use the selectors

The basic combiner is append-instruction-sequences. append_instruction_sequences. This takes as arguments an arbitrary number of two instruction sequences that are to be executed sequentially and returns an instruction sequence whose statements are the statements of all the the two sequences appended together. The subtle point is to determine the registers that are needed and modified by the resulting sequence. It modifies those registers that are modified by any of the sequences; are modified by either sequence; it needs those registers that must be initialized before the first sequence can be run (the registers needed by the first sequence), together with those registers needed by any of the other sequences that are not initialized (modified) by sequences preceding it. the second sequence that are not initialized (modified) by the first sequence.

The sequences are appended two at a time by append-2-sequences. This The function append_instruction_sequences is given two instruction sequences seq1 and seq2 and returns the instruction sequence whose statements instructions are the statements instructions of seq1 followed by the statements instructions of seq2, whose modified registers are those registers that are modified by either seq1 or seq2, and whose needed registers are the registers needed by seq1 together with those registers needed by seq2 that are not modified by seq1. (In terms of set operations, the new set of needed registers is the union of the set of registers needed by seq1 with the set difference of the registers needed by seq2 and the registers modified by seq1.) Thus, append-instruction-sequences append_instruction_sequences is implemented as follows:

This procedure function uses some simple operations for manipulating sets represented as lists, similar to the (unordered) set representation described in section 2.3.3:

Preserving, The function preserving, the second major instruction sequence combiner, takes a list of registers regs and two instruction sequences seq1 and seq2 that are to be executed sequentially. It returns an instruction sequence whose statements instructions are the statements instructions of seq1 followed by the statements instructions of seq2, with appropriate save and restore instructions around seq1 to protect the registers in regs that are modified by seq1 but needed by seq2. To accomplish this, preserving first creates a sequence that has the required saves followed by the statements instructions of seq1 followed by the required restores. This sequence needs the registers being saved and restored in addition to the registers needed by seq1, and modifies the registers modified by seq1 except for the ones being saved and restored. This augmented sequence and seq2 are then appended in the usual way. The following procedure function implements this strategy recursively, walking down the list of registers to be preserved:[1]

Another sequence combiner, tack-on-instruction-sequence, tack_on_instruction_sequence, is used by compile-lambda compile_lambda_expression to append a procedure function body to another sequence. Because the procedure function body is not in line to be executed as part of the combined sequence, its register use has no impact on the register use of the sequence in which it is embedded. We thus ignore the procedure function body's sets of needed and modified registers when we tack it onto the other sequence.

Compile-if The functions compile_conditional and compile-procedure-call compile_function_call use a special combiner called parallel-instruction-sequences parallel_instruction_sequences to append the two alternative branches that follow a test. The two branches will never be executed sequentially; for any particular evaluation of the test, one branch or the other will be entered. Because of this, the registers needed by the second branch are still needed by the combined sequence, even if these are modified by the first branch.