Hash tables are a fast, powerful mechanism for storing large numbers of associations. MIT/GNU Scheme's hash tables feature automatic resizing, customizable growth parameters, customizable hash procedures, and many options for weak references to keys or data.
The average times for the insertion, deletion, and lookup operations on a hash table are bounded by a constant. The space required by the table is proportional to the number of associations in the table; the constant of proportionality is described below (see Resizing of Hash Tables).
In addition to the hash table interface described in the following, MIT Scheme implements SRFI 69: “Basic hash tables”. The reason for supporting two interfaces is partly historical—MIT Scheme supported hash tables prior to the existence of SRFI 69—and partly technical—SFRI 69 fails to specify certain optimization-enabling exceptions to its semantics, forcing a correct implementation to pay the non-negligible performance cost of completely safe behavior. 1 The MIT Scheme native hash table interface, in contrast, specifies the minor exceptions it needs, and is therefore implemented more efficiently. We do not describe the SRFI-69-compliant interface here, as that would be redundant with the SRFI document.
(Previously, the hash-table implementation was a run-time-loadable
option, but as of release 7.7.0 it is loaded by default. It's no longer
necessary to call
load-option prior to using hash tables.)
 SRFI 69 does not give hash functions the flexibility to return new hash values after a garbage collection, which prevents a system whose garbage collector may relocate objects from hashing based on the addresses of objects in memory (see Address Hashing). SRFI 69 also does not specify circumstances when procedures passed as arguments to hash table operations may not themselves modify the hash table, which requires defensive copying and defensive repetitions of lookups.