Merge pull request #109 from udem-dlteam/laurent/dont-use-temporary-w…

…hen-assignment-used-in-condition

Dont use temporary when `x = f()` used in condition

examples/compiled/base64.sh

@@ -105,21 +105,21 @@ _encode() {
 }
 
 defarr _lut 256
-: $((__t1 = c = 0))
+: $((c = 0))
 _get() {
-  let c; let __t1
-  while _getchar __t1; [ $((c = __t1)) -ge 0 ]; do
+  let c
+  while _getchar c; [ $c -ge 0 ]; do
     if [ $((c = _$((_lut + c)))) -ge 0 ] ; then
       break
     fi
   done
   : $(($1 = c))
-  endlet $1 __t1 c
+  endlet $1 c
 }
 
-: $((__t1 = c4 = c3 = c2 = c1 = i = 0))
+: $((c4 = c3 = c2 = c1 = i = 0))
 _decode() {
-  let i; let c1; let c2; let c3; let c4; let __t1
+  let i; let c1; let c2; let c3; let c4
   i=0
   while [ $i -lt 256 ]; do
     : $((_$((_lut + i)) = -1))
@@ -130,21 +130,21 @@ _decode() {
     : $((_$((_lut + 255 & _$((_codes + i)))) = i))
     : $((i += 1))
   done
-  while _get __t1; [ $((c1 = __t1)) -ge 0 ]; do
-    if _get __t1; [ $((c2 = __t1)) -lt 0 ] ; then
+  while _get c1; [ $c1 -ge 0 ]; do
+    if _get c2; [ $c2 -lt 0 ] ; then
       exit 1
     fi
     printf \\$((((c1 << 2) | (c2 >> 4))/64))$((((c1 << 2) | (c2 >> 4))/8%8))$((((c1 << 2) | (c2 >> 4))%8))
-    if _get __t1; [ $((c3 = __t1)) -lt 0 ] ; then
+    if _get c3; [ $c3 -lt 0 ] ; then
       break
     fi
     printf \\$(((255 & ((c2 << 4) | (c3 >> 2)))/64))$(((255 & ((c2 << 4) | (c3 >> 2)))/8%8))$(((255 & ((c2 << 4) | (c3 >> 2)))%8))
-    if _get __t1; [ $((c4 = __t1)) -lt 0 ] ; then
+    if _get c4; [ $c4 -lt 0 ] ; then
       break
     fi
     printf \\$(((255 & ((c3 << 6) | c4))/64))$(((255 & ((c3 << 6) | c4))/8%8))$(((255 & ((c3 << 6) | c4))%8))
   done
-  endlet $1 __t1 c4 c3 c2 c1 i
+  endlet $1 c4 c3 c2 c1 i
 }
 
 : $((myargv = argc = 0))

examples/compiled/c4.sh

@@ -828,7 +828,7 @@ _main() { # argc: $2, argv: $3
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
     return
   fi
-  if _open __t1 $((_$argv_)) 0; [ $((fd = __t1)) -lt 0 ] ; then
+  if _open fd $((_$argv_)) 0; [ $fd -lt 0 ] ; then
     printf "could not open("
     _put_pstr __ $((_$argv_))
     printf ")\n"
@@ -837,25 +837,25 @@ _main() { # argc: $2, argv: $3
     return
   fi
   poolsz=$((256 * 1024))
-  if _malloc __t1 $poolsz; [ $((!(_sym = __t1))) != 0 ] ; then
+  if _malloc _sym $poolsz; [ $((!_sym)) != 0 ] ; then
     printf "could not malloc(%d) symbol area\n" $poolsz
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
     return
   fi
-  if _malloc __t1 $poolsz; [ $((!(_le = _e = __t1))) != 0 ] ; then
+  if _malloc _e $poolsz; [ $((!(_le = _e))) != 0 ] ; then
     printf "could not malloc(%d) text area\n" $poolsz
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
     return
   fi
-  if _malloc __t1 $poolsz; [ $((!(_data = __t1))) != 0 ] ; then
+  if _malloc _data $poolsz; [ $((!_data)) != 0 ] ; then
     printf "could not malloc(%d) data area\n" $poolsz
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
     return
   fi
-  if _malloc __t1 $poolsz; [ $((!(sp = __t1))) != 0 ] ; then
+  if _malloc sp $poolsz; [ $((!sp)) != 0 ] ; then
     printf "could not malloc(%d) stack area\n" $poolsz
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
@@ -879,13 +879,13 @@ _main() { # argc: $2, argv: $3
   : $((_$((_id + _Tk)) = _Char))
   _next __
   idmain=$_id
-  if _malloc __t1 $poolsz; [ $((!(_lp = _p = __t1))) != 0 ] ; then
+  if _malloc _p $poolsz; [ $((!(_lp = _p))) != 0 ] ; then
     printf "could not malloc(%d) source area\n" $poolsz
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))
     return
   fi
-  if _read __t1 $fd $_p $((poolsz - 1)); [ $((i = __t1)) -le 0 ] ; then
+  if _read i $fd $_p $((poolsz - 1)); [ $i -le 0 ] ; then
     printf "read() returned %d\n" $i
     : $(($1 = -1))
     : $((__tmp = $1)) $((argc = $4)) $((argv_ = $5)) $((fd = $6)) $((bt = $7)) $((ty = $8)) $((poolsz = $9)) $((idmain = ${10})) $((pc = ${11})) $((sp = ${12})) $((bp = ${13})) $((a = ${14})) $((cycle = ${15})) $((i = ${16})) $((t = ${17})) $((__t1 = ${18})) $(($1 = __tmp))

examples/compiled/cp.sh

@@ -22,13 +22,13 @@ _malloc() { # $2 = object size
 defarr() { _malloc $1 $2; }
 
 defarr _buffer 1024
-: $((__t1 = len = c = dst = src = args = argc = 0))
+: $((len = c = dst = src = args = argc = 0))
 _main() { let argc $2; let args $3
-  let src; let dst; let c; let len; let __t1
+  let src; let dst; let c; let len
   if [ $argc != 3 ] ; then
     printf "Usage: cp <source> <destination>\n"
     : $(($1 = 1))
-    endlet $1 __t1 len c dst src args argc
+    endlet $1 len c dst src args argc
     return
   fi
   _open src $((_$((args + 1)))) 0
@@ -39,10 +39,10 @@ _main() { let argc $2; let args $3
   if [ $dst = 0 ] ; then
     _file_error __ $((_$((args + 2))))
   fi
-  while _read __t1 $src $_buffer 1024; [ $((len = __t1)) != 0 ]; do
+  while _read len $src $_buffer 1024; [ $len != 0 ]; do
     _write __ $dst $_buffer $len
   done
-  endlet $1 __t1 len c dst src args argc
+  endlet $1 len c dst src args argc
 }
 
 # Runtime library

examples/compiled/wc-stdin.sh

@@ -8,15 +8,15 @@ _is_word_separator() { let c $2
   endlet $1 c
 }
 
-: $((__t1 = last_sep = sep = c = chars = words = lines = 0))
+: $((last_sep = sep = c = chars = words = lines = 0))
 _main() {
-  let lines; let words; let chars; let c; let sep; let last_sep; let __t1
+  let lines; let words; let chars; let c; let sep; let last_sep
   lines=0
   words=0
   chars=0
   sep=0
   last_sep=0
-  while _getchar __t1; [ $((c = __t1)) != -1 ]; do
+  while _getchar c; [ $c != -1 ]; do
     : $((chars += 1))
     if [ $c = $__NEWLINE__ ] ; then
       : $((lines += 1))
@@ -28,7 +28,7 @@ _main() {
     last_sep=$sep
   done
   printf "%d %d %d\n" $lines $words $chars
-  endlet $1 __t1 last_sep sep c chars words lines
+  endlet $1 last_sep sep c chars words lines
 }
 
 # Character constants

examples/compiled/welcome.sh

@@ -2,20 +2,20 @@
 set -e -u -f
 LC_ALL=C
 
-: $((__t1 = i = name = 0))
+: $((i = name = 0))
 _main() {
-  let name; let i; let __t1
+  let name; let i
   _malloc name 100
   i=0
   printf "What is your name?\n"
-  while { _getchar __t1; [ $((_$((name + i)) = __t1)) != -1 ]; } && [ $((_$((name + i)))) != $__NEWLINE__ ]; do
+  while { _getchar _$((name + i)); [ $((_$((name + i)))) != -1 ]; } && [ $((_$((name + i)))) != $__NEWLINE__ ]; do
     : $((i += 1))
   done
   : $((_$((name + i)) = __NUL__))
   printf "Hello, "
   _put_pstr __ $name
   printf "\n"
-  endlet $1 __t1 i name
+  endlet $1 i name
 }
 
 # Character constants

sh.c

@@ -67,6 +67,7 @@ bool comp_body(ast node, STMT_CTX stmt_ctx);
 bool comp_statement(ast node, STMT_CTX stmt_ctx);
 void mark_mutable_variables_body(ast node);
 void handle_enum_struct_union_type_decl(ast node);
+ast handle_side_effects_go(ast node, int executes_conditionally);
 
 // Because concatenating strings is very expensive and a common operation, we
 // use a tree structure to represent the concatenated strings. That way, the
@@ -948,23 +949,69 @@ ast replaced_fun_calls_tail = 0;
 ast conditional_fun_calls = 0;
 ast conditional_fun_calls_tail = 0;
 ast literals_inits = 0;
-int executes_conditionally = 0;
-int contains_side_effects = 0;
+bool contains_side_effects = 0;
+
+ast handle_fun_call_side_effect(ast node, ast assign_to, bool executes_conditionally) {
+  int start_gensym_ix = gensym_ix;
+  ast new_tail;
+  ast sub2;
+
+  if (assign_to == 0) {
+    assign_to = fresh_ident(); // Unique identifier for the function call
+    start_gensym_ix = gensym_ix;
+
+    // At this point, the temporary identifier of the variable is not live and
+    // can be used to evaluate the function arguments. This reduces the number
+    // of temporary variables.
+    gensym_ix -= 1;
+  }
+
+  // Traverse the arguments and replace them with the result of handle_side_effects_go
+  sub2 = get_child(node, 1);
+  if (sub2 != 0) { // Check if not an empty list
+    if (get_op(sub2) == ',') {
+      while (get_op(sub2) == ',') {
+        set_child(sub2, 0, handle_side_effects_go(get_child(sub2, 0), executes_conditionally));
+        sub2 = get_child(sub2, 1);
+      }
+    } else { // sub2 is the first argument, not wrapped in a cons cell
+      sub2 = handle_side_effects_go(sub2, executes_conditionally);
+      set_child(node, 1, sub2);
+    }
+  }
+
+  // All the temporary variables used for the function parameters can be
+  // reused after the function call, so resetting the gensym counter.
+  gensym_ix = start_gensym_ix;
+
+  new_tail = new_ast2(',', assign_to, node);
+  new_tail = new_ast2(',', new_tail, 0);
+  if (executes_conditionally) {
+    if (conditional_fun_calls == 0) { conditional_fun_calls = new_tail; }
+    else { set_child(conditional_fun_calls_tail, 1, new_tail); }
+    conditional_fun_calls_tail = new_tail;
+  }
+  else {
+    if (replaced_fun_calls == 0) { replaced_fun_calls = new_tail; }
+    else { set_child(replaced_fun_calls_tail, 1, new_tail); }
+    replaced_fun_calls_tail = new_tail;
+  }
+
+  return assign_to;
+}
 
 // We can't have function calls and other side effects in $(( ... )), so we need to handle them separately.
 // For unconditional function calls, they are replaced with unique identifiers and returned as a list with their new identifiers.
 // For pre/post-increments/decrements, we map them to a pre-side-effects and replace with the corresponding operation.
 // Note that pre/post-increments/decrements of function calls are not supported.
-ast handle_side_effects_go(ast node, int executes_conditionally) {
+ast handle_side_effects_go(ast node, bool executes_conditionally) {
   int op = get_op(node);
   int nb_children = get_nb_children(node);
-  int start_gensym_ix;
   ast sub1;
   ast sub2;
   ast previous_conditional_fun_calls;
   ast left_conditional_fun_calls;
   ast right_conditional_fun_calls;
-  ast new_tail;
 
   if (nb_children == 0) {
     if (op == IDENTIFIER || op == IDENTIFIER_INTERNAL || op == IDENTIFIER_STRING || op == IDENTIFIER_DOLLAR || op == INTEGER || op == CHARACTER) {
@@ -995,51 +1042,19 @@ ast handle_side_effects_go(ast node, int executes_conditionally) {
     }
   } else if (nb_children == 2) {
     if (op == '(') { // Function call
-      sub1 = fresh_ident(); // Unique identifier for the function call
-
-      start_gensym_ix = gensym_ix;
-
-      // At this point, the temporary identifier of the variable is not live and
-      // can be used to evaluate the function arguments. This reduces the number
-      // of temporary variables.
-      gensym_ix -= 1;
-
-      // Traverse the arguments and replace them with the result of handle_side_effects_go
-      sub2 = get_child(node, 1);
-      if (sub2 != 0) { // Check if not an empty list
-        if (get_op(sub2) == ',') {
-          while (get_op(sub2) == ',') {
-            set_child(sub2, 0, handle_side_effects_go(get_child(sub2, 0), executes_conditionally));
-            sub2 = get_child(sub2, 1);
-          }
-        } else { // sub2 is the first argument, not wrapped in a cons cell
-          sub2 = handle_side_effects_go(sub2, executes_conditionally);
-          set_child(node, 1, sub2);
-        }
-      }
-
-      // All the temporary variables used for the function parameters can be
-      // reused after the function call, so resetting the gensym counter.
-      gensym_ix = start_gensym_ix;
-
-      new_tail = new_ast2(',', sub1, node);
-      new_tail = new_ast2(',', new_tail, 0);
-      if (executes_conditionally) {
-        if (conditional_fun_calls == 0) { conditional_fun_calls = new_tail; }
-        else { set_child(conditional_fun_calls_tail, 1, new_tail); }
-        conditional_fun_calls_tail = new_tail;
-      }
-      else {
-        if (replaced_fun_calls == 0) { replaced_fun_calls = new_tail; }
-        else { set_child(replaced_fun_calls_tail, 1, new_tail); }
-        replaced_fun_calls_tail = new_tail;
+      return handle_fun_call_side_effect(node, 0, executes_conditionally);
+    } else if (op == '=') {
+      if (get_op(get_child(node, 1)) == '(') { // Function call
+        // In that case, we reuse the left hand side of the assignment as the result location
+        return handle_fun_call_side_effect(get_child(node, 1), get_child(node, 0), executes_conditionally);
+      } else {
+        sub1 = handle_side_effects_go(get_child(node, 0), executes_conditionally);
+        sub2 = handle_side_effects_go(get_child(node, 1), executes_conditionally); // We could inline that one since the assignment to the global variable is done after the last handle_side_effects_go call
+        return new_ast2(op, sub1, sub2);
       }
-
-      return sub1;
     } else if (op == '&' || op == '|' || op == '<' || op == '>' || op == '+' || op == '-' || op == '*' || op == '/'
-      || op == '%' || op == '^' || op == ',' || op == EQ_EQ || op == EXCL_EQ || op == LT_EQ || op == GT_EQ || op == LSHIFT || op == RSHIFT || op == '=' || op == '['
+      || op == '%' || op == '^' || op == ',' || op == EQ_EQ || op == EXCL_EQ || op == LT_EQ || op == GT_EQ || op == LSHIFT || op == RSHIFT || op == '['
       || op == '.' || op == ARROW ) {
-      // We can't place handle_side_effects_go directly in new_ast2 call because six-cc creates a global variable that gets overwritten in the other handle_side_effects_go calls
       sub1 = handle_side_effects_go(get_child(node, 0), executes_conditionally);
       sub2 = handle_side_effects_go(get_child(node, 1), executes_conditionally); // We could inline that one since the assignment to the global variable is done after the last handle_side_effects_go call
       return new_ast2(op, sub1, sub2);