Library ssreflect

This file is the Gallina part of the ssreflect plugin implementation.   
 Files that use the ssreflect plugin should always Require ssreflect and 
 either Import ssreflect or Import ssreflect.SsrSyntax.                  
   The contents of this file is quite technical and should only interest 
 advanced developers; features not covered by the Ssreflect reference    
 manual, such as the Unlockable interface, phantom types, and the        
 [the struct of T] construct, are covered by specific comments below.    

Declare Ssr keywords: 'is' 'by' 'of' '//' '/=' and '//='.                
 We also declare the parsing level 8, as a workaround for a notation      
 grammar factoring problem. Arguments of application-style notations      
 (at level 10) should be declared at level 8 rather than 9 or the camlp4  
 grammar will not factor properly.                                        

Make the general "if" into a notation, so that we can override it below 
 The notations are "only parsing" because the Coq decompiler won't       
 recognize the expansion of the boolean if; using the default printer    
 avoids a spurrious trailing %GEN_IF.                                    

Force boolean interpretation of simple if expressions.          

To allow a wider variety of notations without reserving a large number 
 of identifiers, the ssreflect library systematically uses "forms" to   
 enclose complex mixfix syntax. A "form" is simply a mixfix expression  
 enclosed in square brackets and introduced by a keyword:               
      [keyword ... ]                                                    
 Because the keyword follows a bracket it does not need to be reserved. 
 Non-ssreflect libraries that do not respect the form syntax (e.g., the 
 Coq Lists library) should be loaded before ssreflect so that their     
 notations do not mask all ssreflect forms.                             

Allow the casual use of notations like nat * nat for explicit Type 
 declarations. Note that (nat * nat : Type) is NOT equivalent to    
 (nat * nat)%type, whose inferred type is legacy type "Set".        

Syntax for referring to canonical structures:                   
      [the struct_type of proj_val by proj_fun]                  
 This form denotes the Canonical instance s of the Structure     
 type struct_type whose proj_fun projection is proj_val, i.e.,   
 such that proj_fun s = proj_val. Typically proj_fun will be one 
 of the record field accessors of struct_type, but this need not 
 be the case; it can be, for instance, a field of a record type  
 to which struct_type coerces; proj_val will likewise be coerced 
 to the return type of proj_fun. In all but the simplest cases,  
 proj_fun should be eta-expanded to allow for the insertion of   
 implicit arguments.                                             
   In the common case where proj_fun itself is a coercion, the   
 "by" part can be omitted entirely; in this case it is inferred  
 by casting s : struct_type to the inferred type of proj_val.    
 Obviously the latter can be fixed by using an implicit cast on  
 proj_val, and it is highly recommended to do so when the return 
 type intended for proj_fun is "Type", as the type inferred for  
 proj_val may vary because of sort polymorphism (it could be Set 
 or Prop).                                                       
   Special note for telescopes (structures with coercion-based   
 inheritance): when using the "the" form to generate a           
 substructure from a canonical hierarchy, on should always       
 project or coerce the value to the BASE structure, because Coq  
 will only find a Canonical derived structure for the Canonical  
 base structure -- not for a base structure that is specific to  
 proj_value.                                                     

The following are "format only" versions of the above notations.   
 Since Coq doesn't provide this option, we fake it by splitting the 
 "the" keyword. We need to do this because the formatter will be    
 thrown off by application collapsing, coercion insertion and beta  
 reduction in the right hand sides of the above notations.          

Helper notation for canonical structure inheritance support.           
 This is a workaround for the poor interaction between delta reduction  
 and canonical projections in Coq's unification algorithm, by which     
 transparent definitions hide canonical structures, i.e., in            
   Canonical Structure a_type_struct := @Struct a_type ...              
   Definition my_type := a_type.                                        
 my_type doesn't effectively inherit the struct structure from a_type.  
 Our solution is to redeclare the structure, as follows                 
   Canonical Structure my_type_struct :=                                
     Eval hnf in [struct of my_type].                                   
 The special notation [str of _] must be defined for each Strucure      
 "str" with constructor "Str", typically as follows                     
   Definition repack_str s :=                                           
      let: Str _ x y ... z := s return {type of Str for s} -> str in    
      fun k => k _ x y ... z.                                           
    Notation "[ 'str' 'of' T 'for' s ]" := (@repack_str s (@Str T))     
      (at level 0, format "[ 'str'  'of'  T  'for'  s ]") : form_scope. 
    Notation "[ 'str' 'of' T ]" := (repack_str (fun x => @Str T x))     
      (at level 0, format "[ 'str'  'of'  T ]") : form_scope.           
 The notation for the match return predicate is defined below; the eta  
 expansion in the second form serves both to distinguish it from the    
 first and to avoid the delta reduction problem.                        
   There are several variations on the notation and the definition of   
 the "repack" function, for telescopes, mixin classes, and join         
 (multiple inheritance) classes; see fintype.v and ssralg.v for         
 examples; they involve inferring the structure from instances of       
 reflexivity or from phantoms (see below), rather than directly from    
 the constructor as above.                                              

A generic "phantom" type (actually, the unit type with a phantom      
 parameter). This can be used for type definitions that require some   
 Structure on one of their parameters, to allow Coq to infer said      
 structure rather that having to supply it explicitly or to resort to  
 the "[is _ <: _]" notation, which interacts poorly with Notation.     
   The definition of a (co)inductive type with a parameter p : p_type, 
 that uses the operations of a structure                               
  Structure p_str : Type := p_Str {                                    
    p_repr :> p_type; p_op : p_repr -> ...}                            
 should be given as                                                    
  Inductive indt_type (p : p_str) : Type := Indt ... .                 
  Definition indt_of (p : p_str) & phantom p_type p := indt_type p.    
  Notation "{ 'indt' p }" := (indt_of (Phantom p)).                    
  Definition indt p x y ... z : {indt p} := @Indt p x y ... z.         
  Notation "[ 'indt' x y ... z ]" := (indt x y ... z).                 
 That is, the concrete type and its constructor should be shadowed by  
 definitions that use a phantom argument to infer and display the true 
 value of p (in practice, the "indt" constructor often performs        
 additional functions, like "locking" the representation (see below).  
   We also define a simpler version ("phant" / "Phant") of phantom for 
 the common case where p_type is Type.                                 

Internal tagging used by the implementation of the ssreflect elim. 

Term tagging (user-level).                                              
 We provide two strengths of term tagging :                               
  - (nosimpl t) simplifies to t EXCEPT in a definition; more precisely,   
    given Definition foo := (nosimpl bar), foo (or (foo t')) will NOT be  
    expanded by the simpl tactic unless it is in a forcing context (e.g., 
    in match foo t' with ... end, foo t' will be reduced if this allows   
    match to be reduced. Note that nosimpl bar is simply notation for     
    a term that beta-iota reduces to bar; hence unfold foo will replace   
    foo by bar, and fold foo will replace bar by foo. A final warning:    
    nosimpl only works if no reduction is apparent in t; in particular    
    Definition foo x := nosimpl t. will usually not work.                 
    CAVEAT: nosimpl should not be used inside a Section, because the end  
    of section "cooking" removes the iota redex.                          
  - (locked t) is provably equal to t, but is not convertible to t; it    
    provides support for abstract data types, and selective rewriting.    
    The equation t == (locked t) is provided as a lemma, but it should    
    only be used for selective rewriting (see below). For ADTs, the       
    unlock tactic should be used to remove locking.                       
 locked is also used as a placeholder for the implementation of flexible  
 matching.                                                                
 Addendum: it appears that the use of a generic key confuses the term     
 comparison heuristic of the Coq kernel, which thinks all locked terms    
 have the same "head constant", and therefore immediately jumps to        
 comparing their LAST argument. Furthermore, Coq still needs to delta     
 expand a locked constant when comparing unequal terms, and, given the    
 total absence of caching of comparisons, this causes an exponential      
 blowup in comparisons that return false on terms that are built from     
 many combinators, which is quite common in a modular development.        
   As a stopgap, we use the module system to create opaque constants      
 with an expansion lemma; this is pretty clumsy because design of the     
 module language does not support such small-scale usage very well. See   
 the definiiton of card and subset in fintype.v for examples of this.     
   Of course the unlock tactic will not support the expansion of this new 
 kind of opaque constants; to compensate for that we use "unlockable"     
 canonical structures to store the expansion lemmas, which can then be    
 retrieved by a generic "unlock" rewrite rule. Note that because of a     
 technical limitation of the implementation of canonical projection       
 in ssreflect 1.1, unlock must weaken the intensional equality between    
 the constant and its definition to an extensional one.                   

To unlock opaque constants. 

This should be Definition locked := locked_with master_key, 
 but for compatibility with the ml4 code.                    

Needed for locked predicates, in particular for eqType's. 

The basic closing tactic "done".                                       

The internal lemmas for the have tactics.                                

Internal  N-ary congruence lemma for the congr tactic 

View lemmas that don't use reflection.