Re: [PATCH, PR 57195] Allow mode iterators inside angle brackets

[Michael Collison]

Richard,

Here is a modified patch that takes your comments into account. Breaking 
on depth == 0 with '>' does not work due to the code looking for whitespace.

2015-08-25  Michael Collison  <michael.colli...@linaro.org>

    PR other/57195
    * read-md.c (read_name): Allow mode iterators inside angle
    brackets in rtl expressions.


--- a/gcc/read-md.c
+++ b/gcc/read-md.c
@@ -399,20 +399,31 @@ read_name (struct md_name *name)
 {
   int c;
   size_t i;
+  int angle_bracket_depth;

   c = read_skip_spaces ();

   i = 0;
+  angle_bracket_depth = 0;
   while (1)
     {
+      if (c == '<')
+    angle_bracket_depth++;
+
+      if ((c == '>') && (angle_bracket_depth > 0))
+      angle_bracket_depth--;
+
       if (c == ' ' || c == '\n' || c == '\t' || c == '\f' || c == '\r'
       || c == EOF)
     break;
-      if (c == ':' || c == ')' || c == ']' || c == '"' || c == '/'
-      || c == '(' || c == '[')
+      if (angle_bracket_depth == 0)
     {
-      unread_char (c);
-      break;
+      if (c == ':' || c == ')' || c == ']'
+          || c == '"' || c == '/' || c == '(' || c == '[')
+        {
+          unread_char (c);
+          break;
+        }
     }

       if (i == sizeof (name->buffer) - 1)

On 09/07/2015 02:46 AM, Richard Sandiford wrote:
Michael Collison <michael.colli...@linaro.org> writes:
This patch allow mode iterators inside angle brackets in machine
description files. I discovered the issue when attempting to use
iterators on match_operand's as follows:

match_operand:<VW:V_widen> 0 "s_register_operand" "=w")

The function 'read_name' is nor properly handling ':' inside angle brackets.

Bootstrapped on arm-linux.
Sorry for the slow review.

diff --git a/gcc/read-md.c b/gcc/read-md.c
index 9f158ec..0171fb0 100644
--- a/gcc/read-md.c
+++ b/gcc/read-md.c
@@ -399,17 +399,25 @@ read_name (struct md_name *name)
   {
     int c;
     size_t i;
+  bool in_angle_bracket;

     c = read_skip_spaces ();

     i = 0;
+  in_angle_bracket = false;
     while (1)
       {
+      if (c == '<')
+    in_angle_bracket = true;
+
+      if (c == '>')
+    in_angle_bracket = false;
+
         if (c == ' ' || c == '\n' || c == '\t' || c == '\f' || c == '\r'
         || c == EOF)
       break;
-      if (c == ':' || c == ')' || c == ']' || c == '"' || c == '/'
-      || c == '(' || c == '[')
+      if (((c == ':') and (!in_angle_bracket)) || c == ')' || c == ']'
+      || c == '"' || c == '/' || c == '(' || c == '[')
       {
         unread_char (c);
         break;
I think we should have a nesting depth rather than a boolean.
It also seems more natural to skip the final "if" statement above when
inside an angle bracket, rather than treating ':' as a special case.
(We'd still break at the end of the line in the case of a missing '>',
so the error reporting shouldn't be too bad.)

I suppose logically '>' with a nesting depth of 0 should also break
the loop.

Thanks for fixing this.

Richard


--
Michael Collison
Linaro Toolchain Working Group
michael.colli...@linaro.org