power operation on functions with syntax o^; power operation on relations with syntax ^^
(* Title: HOL/IMP/Compiler.thy
ID: $Id$
Author: Tobias Nipkow, TUM
Copyright 1996 TUM
This is an early version of the compiler, where the abstract machine
has an explicit pc. This turned out to be awkward, and a second
development was started. See Machines.thy and Compiler.thy.
*)
header "A Simple Compiler"
theory Compiler0 imports Natural begin
subsection "An abstract, simplistic machine"
text {* There are only three instructions: *}
datatype instr = ASIN loc aexp | JMPF bexp nat | JMPB nat
text {* We describe execution of programs in the machine by
an operational (small step) semantics:
*}
inductive_set
stepa1 :: "instr list \<Rightarrow> ((state\<times>nat) \<times> (state\<times>nat))set"
and stepa1' :: "[instr list,state,nat,state,nat] \<Rightarrow> bool"
("_ \<turnstile> (3\<langle>_,_\<rangle>/ -1\<rightarrow> \<langle>_,_\<rangle>)" [50,0,0,0,0] 50)
for P :: "instr list"
where
"P \<turnstile> \<langle>s,m\<rangle> -1\<rightarrow> \<langle>t,n\<rangle> == ((s,m),t,n) : stepa1 P"
| ASIN[simp]:
"\<lbrakk> n<size P; P!n = ASIN x a \<rbrakk> \<Longrightarrow> P \<turnstile> \<langle>s,n\<rangle> -1\<rightarrow> \<langle>s[x\<mapsto> a s],Suc n\<rangle>"
| JMPFT[simp,intro]:
"\<lbrakk> n<size P; P!n = JMPF b i; b s \<rbrakk> \<Longrightarrow> P \<turnstile> \<langle>s,n\<rangle> -1\<rightarrow> \<langle>s,Suc n\<rangle>"
| JMPFF[simp,intro]:
"\<lbrakk> n<size P; P!n = JMPF b i; ~b s; m=n+i \<rbrakk> \<Longrightarrow> P \<turnstile> \<langle>s,n\<rangle> -1\<rightarrow> \<langle>s,m\<rangle>"
| JMPB[simp]:
"\<lbrakk> n<size P; P!n = JMPB i; i <= n; j = n-i \<rbrakk> \<Longrightarrow> P \<turnstile> \<langle>s,n\<rangle> -1\<rightarrow> \<langle>s,j\<rangle>"
abbreviation
stepa :: "[instr list,state,nat,state,nat] \<Rightarrow> bool"
("_ \<turnstile>/ (3\<langle>_,_\<rangle>/ -*\<rightarrow> \<langle>_,_\<rangle>)" [50,0,0,0,0] 50) where
"P \<turnstile> \<langle>s,m\<rangle> -*\<rightarrow> \<langle>t,n\<rangle> == ((s,m),t,n) : ((stepa1 P)^*)"
abbreviation
stepan :: "[instr list,state,nat,nat,state,nat] \<Rightarrow> bool"
("_ \<turnstile>/ (3\<langle>_,_\<rangle>/ -(_)\<rightarrow> \<langle>_,_\<rangle>)" [50,0,0,0,0,0] 50) where
"P \<turnstile> \<langle>s,m\<rangle> -(i)\<rightarrow> \<langle>t,n\<rangle> == ((s,m),t,n) : (stepa1 P ^^ i)"
subsection "The compiler"
consts compile :: "com \<Rightarrow> instr list"
primrec
"compile \<SKIP> = []"
"compile (x:==a) = [ASIN x a]"
"compile (c1;c2) = compile c1 @ compile c2"
"compile (\<IF> b \<THEN> c1 \<ELSE> c2) =
[JMPF b (length(compile c1) + 2)] @ compile c1 @
[JMPF (%x. False) (length(compile c2)+1)] @ compile c2"
"compile (\<WHILE> b \<DO> c) = [JMPF b (length(compile c) + 2)] @ compile c @
[JMPB (length(compile c)+1)]"
declare nth_append[simp]
subsection "Context lifting lemmas"
text {*
Some lemmas for lifting an execution into a prefix and suffix
of instructions; only needed for the first proof.
*}
lemma app_right_1:
assumes "is1 \<turnstile> \<langle>s1,i1\<rangle> -1\<rightarrow> \<langle>s2,i2\<rangle>"
shows "is1 @ is2 \<turnstile> \<langle>s1,i1\<rangle> -1\<rightarrow> \<langle>s2,i2\<rangle>"
using assms
by induct auto
lemma app_left_1:
assumes "is2 \<turnstile> \<langle>s1,i1\<rangle> -1\<rightarrow> \<langle>s2,i2\<rangle>"
shows "is1 @ is2 \<turnstile> \<langle>s1,size is1+i1\<rangle> -1\<rightarrow> \<langle>s2,size is1+i2\<rangle>"
using assms
by induct auto
declare rtrancl_induct2 [induct set: rtrancl]
lemma app_right:
assumes "is1 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>"
shows "is1 @ is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>"
using assms
proof induct
show "is1 @ is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s1,i1\<rangle>" by simp
next
fix s1' i1' s2 i2
assume "is1 @ is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s1',i1'\<rangle>"
and "is1 \<turnstile> \<langle>s1',i1'\<rangle> -1\<rightarrow> \<langle>s2,i2\<rangle>"
thus "is1 @ is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>"
by (blast intro: app_right_1 rtrancl_trans)
qed
lemma app_left:
assumes "is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>"
shows "is1 @ is2 \<turnstile> \<langle>s1,size is1+i1\<rangle> -*\<rightarrow> \<langle>s2,size is1+i2\<rangle>"
using assms
proof induct
show "is1 @ is2 \<turnstile> \<langle>s1,length is1 + i1\<rangle> -*\<rightarrow> \<langle>s1,length is1 + i1\<rangle>" by simp
next
fix s1' i1' s2 i2
assume "is1 @ is2 \<turnstile> \<langle>s1,length is1 + i1\<rangle> -*\<rightarrow> \<langle>s1',length is1 + i1'\<rangle>"
and "is2 \<turnstile> \<langle>s1',i1'\<rangle> -1\<rightarrow> \<langle>s2,i2\<rangle>"
thus "is1 @ is2 \<turnstile> \<langle>s1,length is1 + i1\<rangle> -*\<rightarrow> \<langle>s2,length is1 + i2\<rangle>"
by (blast intro: app_left_1 rtrancl_trans)
qed
lemma app_left2:
"\<lbrakk> is2 \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>; j1 = size is1+i1; j2 = size is1+i2 \<rbrakk> \<Longrightarrow>
is1 @ is2 \<turnstile> \<langle>s1,j1\<rangle> -*\<rightarrow> \<langle>s2,j2\<rangle>"
by (simp add: app_left)
lemma app1_left:
assumes "is \<turnstile> \<langle>s1,i1\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle>"
shows "instr # is \<turnstile> \<langle>s1,Suc i1\<rangle> -*\<rightarrow> \<langle>s2,Suc i2\<rangle>"
proof -
from app_left [OF assms, of "[instr]"]
show ?thesis by simp
qed
subsection "Compiler correctness"
declare rtrancl_into_rtrancl[trans]
converse_rtrancl_into_rtrancl[trans]
rtrancl_trans[trans]
text {*
The first proof; The statement is very intuitive,
but application of induction hypothesis requires the above lifting lemmas
*}
theorem
assumes "\<langle>c,s\<rangle> \<longrightarrow>\<^sub>c t"
shows "compile c \<turnstile> \<langle>s,0\<rangle> -*\<rightarrow> \<langle>t,length(compile c)\<rangle>" (is "?P c s t")
using assms
proof induct
show "\<And>s. ?P \<SKIP> s s" by simp
next
show "\<And>a s x. ?P (x :== a) s (s[x\<mapsto> a s])" by force
next
fix c0 c1 s0 s1 s2
assume "?P c0 s0 s1"
hence "compile c0 @ compile c1 \<turnstile> \<langle>s0,0\<rangle> -*\<rightarrow> \<langle>s1,length(compile c0)\<rangle>"
by (rule app_right)
moreover assume "?P c1 s1 s2"
hence "compile c0 @ compile c1 \<turnstile> \<langle>s1,length(compile c0)\<rangle> -*\<rightarrow>
\<langle>s2,length(compile c0)+length(compile c1)\<rangle>"
proof -
show "\<And>is1 is2 s1 s2 i2.
is2 \<turnstile> \<langle>s1,0\<rangle> -*\<rightarrow> \<langle>s2,i2\<rangle> \<Longrightarrow>
is1 @ is2 \<turnstile> \<langle>s1,size is1\<rangle> -*\<rightarrow> \<langle>s2,size is1+i2\<rangle>"
using app_left[of _ 0] by simp
qed
ultimately have "compile c0 @ compile c1 \<turnstile> \<langle>s0,0\<rangle> -*\<rightarrow>
\<langle>s2,length(compile c0)+length(compile c1)\<rangle>"
by (rule rtrancl_trans)
thus "?P (c0; c1) s0 s2" by simp
next
fix b c0 c1 s0 s1
let ?comp = "compile(\<IF> b \<THEN> c0 \<ELSE> c1)"
assume "b s0" and IH: "?P c0 s0 s1"
hence "?comp \<turnstile> \<langle>s0,0\<rangle> -1\<rightarrow> \<langle>s0,1\<rangle>" by auto
also from IH
have "?comp \<turnstile> \<langle>s0,1\<rangle> -*\<rightarrow> \<langle>s1,length(compile c0)+1\<rangle>"
by(auto intro:app1_left app_right)
also have "?comp \<turnstile> \<langle>s1,length(compile c0)+1\<rangle> -1\<rightarrow> \<langle>s1,length ?comp\<rangle>"
by(auto)
finally show "?P (\<IF> b \<THEN> c0 \<ELSE> c1) s0 s1" .
next
fix b c0 c1 s0 s1
let ?comp = "compile(\<IF> b \<THEN> c0 \<ELSE> c1)"
assume "\<not>b s0" and IH: "?P c1 s0 s1"
hence "?comp \<turnstile> \<langle>s0,0\<rangle> -1\<rightarrow> \<langle>s0,length(compile c0) + 2\<rangle>" by auto
also from IH
have "?comp \<turnstile> \<langle>s0,length(compile c0)+2\<rangle> -*\<rightarrow> \<langle>s1,length ?comp\<rangle>"
by (force intro!: app_left2 app1_left)
finally show "?P (\<IF> b \<THEN> c0 \<ELSE> c1) s0 s1" .
next
fix b c and s::state
assume "\<not>b s"
thus "?P (\<WHILE> b \<DO> c) s s" by force
next
fix b c and s0::state and s1 s2
let ?comp = "compile(\<WHILE> b \<DO> c)"
assume "b s0" and
IHc: "?P c s0 s1" and IHw: "?P (\<WHILE> b \<DO> c) s1 s2"
hence "?comp \<turnstile> \<langle>s0,0\<rangle> -1\<rightarrow> \<langle>s0,1\<rangle>" by auto
also from IHc
have "?comp \<turnstile> \<langle>s0,1\<rangle> -*\<rightarrow> \<langle>s1,length(compile c)+1\<rangle>"
by (auto intro: app1_left app_right)
also have "?comp \<turnstile> \<langle>s1,length(compile c)+1\<rangle> -1\<rightarrow> \<langle>s1,0\<rangle>" by simp
also note IHw
finally show "?P (\<WHILE> b \<DO> c) s0 s2".
qed
text {*
Second proof; statement is generalized to cater for prefixes and suffixes;
needs none of the lifting lemmas, but instantiations of pre/suffix.
*}
(*
theorem assumes A: "\<langle>c,s\<rangle> \<longrightarrow>\<^sub>c t"
shows "\<And>a z. a@compile c@z \<turnstile> \<langle>s,size a\<rangle> -*\<rightarrow> \<langle>t,size a + size(compile c)\<rangle>"
(is "\<And>a z. ?P c s t a z")
proof -
from A show "\<And>a z. ?thesis a z"
proof induct
case Skip thus ?case by simp
next
case Assign thus ?case by (force intro!: ASIN)
next
fix c1 c2 s s' s'' a z
assume IH1: "\<And>a z. ?P c1 s s' a z" and IH2: "\<And>a z. ?P c2 s' s'' a z"
from IH1 IH2[of "a@compile c1"]
show "?P (c1;c2) s s'' a z"
by(simp add:add_assoc[THEN sym])(blast intro:rtrancl_trans)
next
(* at this point I gave up converting to structured proofs *)
(* \<IF> b \<THEN> c0 \<ELSE> c1; case b is true *)
apply(intro strip)
(* instantiate assumption sufficiently for later: *)
apply(erule_tac x = "a@[?I]" in allE)
apply(simp)
(* execute JMPF: *)
apply(rule converse_rtrancl_into_rtrancl)
apply(force intro!: JMPFT)
(* execute compile c0: *)
apply(rule rtrancl_trans)
apply(erule allE)
apply assumption
(* execute JMPF: *)
apply(rule r_into_rtrancl)
apply(force intro!: JMPFF)
(* end of case b is true *)
apply(intro strip)
apply(erule_tac x = "a@[?I]@compile c0@[?J]" in allE)
apply(simp add:add_assoc)
apply(rule converse_rtrancl_into_rtrancl)
apply(force intro!: JMPFF)
apply(blast)
apply(force intro: JMPFF)
apply(intro strip)
apply(erule_tac x = "a@[?I]" in allE)
apply(erule_tac x = a in allE)
apply(simp)
apply(rule converse_rtrancl_into_rtrancl)
apply(force intro!: JMPFT)
apply(rule rtrancl_trans)
apply(erule allE)
apply assumption
apply(rule converse_rtrancl_into_rtrancl)
apply(force intro!: JMPB)
apply(simp)
done
*)
text {* Missing: the other direction! I did much of it, and although
the main lemma is very similar to the one in the new development, the
lemmas surrounding it seemed much more complicated. In the end I gave
up. *}
end