SSFRAG : { ac     : (thm * thm) list,
           congs  : thm list,
           convs  : {conv  : (term list -> conv) -> term list -> conv,
                     key   : (term list * term) option,
                     name  : string,
                     trace : int} list,
           dprocs : Traverse.reducer list,
           filter : (controlled_thm -> controlled_thm list) option,
           name   : string option,
           rewrs  : thm list } -> ssfrag
STRUCTURE
SYNOPSIS
Constructs ssfrag values.
LIBRARY
simpLib
DESCRIPTION
The ssfrag type is the way in which simplification components are packaged up and made available to the simplifier (though ssfrag values must first be turned into simpsets, either by addition to an existing simpset, or with the mk_simpset function).

The big record type passed to SSFRAG as an argument has seven fields. Here we describe each in turn.

The ac field is a list of “AC theorem” pairs. Each such pair is the pair of theorems stating that a given binary function is associative and commutative. The theorems can be given in either order, can present the associativity theorem in either direction, and can be generalised to any extent.

The congs field is a list of congruence theorems justifying the addition of theorems to simplification contexts. For example, the congruence theorem for implication is

   |- (P = P') ==> (P' ==> (Q = Q')) ==> (P ==> Q = P' ==> Q')
This theorem encodes a rewriting strategy. The consequent of the chain of implications is the form of term in question, where the appropriate components have been rewritten. Then, in left-to-right order, the various antecedents of the implication specify the rewriting strategy which gives rise to the consequent. In this example, P is first simplified to P' without any additional context, then, using P' as additional context, simplification of Q proceeds, producing Q'. Another example is a rule for conjunction:
   |- (P ==> (Q = Q')) ==> (Q' ==> (P = P')) ==> ((P /\ Q) = (P' /\ Q'))
Here P is assumed while Q is simplified to Q'. Then, Q' is assumed while P is simplified to P'. If an antecedent doesn’t involve the relation in question (here, equality) then it is treated as a side-condition, and the simplifier will be recursively invoked to try and solve it.

Higher-order congruence rules are also possible. These provide a method for dealing with bound variables. The following is a rule for the restricted universal quantifier, for example:

   |- (P = Q) ==> (!v. v IN Q ==> (f v = g v)) ==>
      (RES_FORALL P f = RES_FORALL Q g)

(If f is an abstraction, \x. t, then RES_FORALL P f is pretty-printed as !x::P. t) Terms in the conclusions of higher-order congruence rules that might be abstractions (such as f above) should be kept as variables, rather than written out as abstractions. In other words, the conclusion of the congruence rule above should not be written as

   RES_FORALL P (\v. f v) = RES_FORALL Q (\v. g v)

The convs field is a list of conversions that the simplifier will apply. Each conversion added to an ssfrag value is done so in a record consisting of four fields.

The conv field of this subsidiary record type includes the value of the conversion itself. When the simplifier applies the conversion it is actually passed two extra arguments (as the type indicates). The first is a solver function that can be used to recursively do side-condition solving, and the second is a stack of side-conditions that have been accumulated to date. Many conversions will typically ignore these arguments (as in the example below).

The key field of the subsidiary record type is an optional pattern, specifying the places where the conversion should be applied. If the value is NONE, then the conversion will be applied to all sub-terms. If the value is SOME(lcs, t), then the term t is used as a pattern specifying those terms to which the conversion should be applied. Further, the list lcs (which must be a list of variables), specifies those variables in t which shouldn’t be generalised against; they are effectively local constants. Note, however, that the types in the pattern term t will not be used to eliminate possible matches, so that if a match is desired with a particular type instantiation of a term, then the conversion will need to reject the input itself. The name and trace fields are only relevant to the debugging facilities of the simplifier.

The dprocs field of the record passed to SSFRAG is where decision procedures can be specified. Documentation describing the construction and use of values of type reducer is available in the DESCRIPTION.

The filter field of the record is an optional function, which, if present, is composed with the standard simplifier’s function for generating rewrites from theorems, and replaces that function. The version of this present in bool_ss and its descendents will, for example, turn |- P /\ Q into |- P and |- Q, and |- ~(t1 = t2) into |- (t1 = t2) = F and |- (t2 = t1) = F.

The controlled_thm type is defined in the module BoundedRewrites, and pairs a theorem with a bound, which is either the value UNBOUNDED or the constructor BOUNDED applied to an integer reference. The reference is used to limit the number of times a rewrite may be applied. The filter gets information as to whether and how a rewrite is bounded as this can affect its decision on whether or not to include a rewrite at all (if it looks as if it will loop, and the bound is UNBOUNDED, it should be dropped, but it may choose to keep it if there is bound present).

The rewrs field of the record is a list of rewrite theorems that are to be applied.

The name field of the record is an optional name for the simpset fragment that is used when it, and simpsets that it becomes part of are pretty-printed.

FAILURE
Never fails. Failure to provide theorems of just the right form may cause later application of simplification functions to fail, documentation to the contrary notwithstanding.
EXAMPLE
Given a conversion MUL_CONV to calculate multiplications, the following illustrates how this can be added to a simpset:
   - val ssd = SSFRAG {ac = [], congs = [],
                        convs = [{conv = K (K MUL_CONV),
                                  key= SOME ([], Term`x * y`),
                                  name = "MUL_CONV",
                                  trace = 2}],
                        dprocs = [], filter = NONE, rewrs = []};
   > val ssd =
       SSFRAG{ac = [], congs = [],
               convs =
                 [{conv = fn, key = SOME([], `x * y`), name = "MUL_CONV",
                   trace = 2}], dprocs = [], filter = NONE, rewrs = []}
       : ssfrag
   - SIMP_CONV bool_ss [] (Term`3 * 4`);
   > val it = |- 3 * 4 = 3 * 4 : thm
   - SIMP_CONV (bool_ss ++ ssd) [] (Term`3 * 4`);
   > val it = |- 3 * 4 = 12 : thm
Given the theorems ADD_SYM and ADD_ASSOC from arithmeticTheory, we can construct a normaliser for additive terms.
   - val ssd2 = SSFRAG {ac = [(SPEC_ALL ADD_ASSOC, SPEC_ALL ADD_SYM)],
                         congs = [], convs = [], dprocs = [],
                         filter = NONE, rewrs = []};
   > val ssd2 =
       SSFRAG{ac = [(|- m + n + p = (m + n) + p, |- m + n = n + m)],
               congs = [], convs = [], dprocs = [], filter = NONE,
               rewrs = []}
       : ssfrag
   - SIMP_CONV (bool_ss ++ ssd2) [] (Term`(y + 3) + x + 4`);
     (* note that the printing of + in this example is that of a
        right associative operator.*)
   > val it = |- (y + 3) + x + 4 = 3 + 4 + x + y : thm
SEEALSO
HOL  Trindemossen-1