[PATCH] fix typo in tree-ssa-strlen.c (PR 86696)

[PATCH] fix typo in tree-ssa-strlen.c (PR 86696)

[Martin Sebor]

[-- Attachment #1: Type: text/plain, Size: 559 bytes --]

My yesterday's change to tree-ssa-strlen.c introduced a test
for INTEGER_TYPE where INTEGRAL_TYPE_P should have been used,
causing a subsequent ICE when the latter was supplied.

The attached patch corrects the test and also makes the subsequent
code more robust and be prepared for the test to fail.

Since the change also caused a couple pf regressions in the Ada
test suite, if there are no objections, I will commit the patch
later today to clear up those failures.

Retested on x86_64-linux with no regressions (including in the Ada
test suite).

Martin

[-- Attachment #2: gcc-86696.diff --]
[-- Type: text/x-patch, Size: 2316 bytes --]

PR tree-optimization/86696 - ICE in handle_char_store at gcc/tree-ssa-strlen.c:3332

gcc/ChangeLog:

	PR tree-optimization/86696
	* tree-ssa-strlen.c (get_min_string_length): Handle all integer
	types, including enums.
	(handle_char_store): Be prepared for the above function to fail.

gcc/testsuite/ChangeLog:

	PR tree-optimization/86696
	* gcc.dg/pr86696.C: New test.

Index: gcc/testsuite/g++.dg/pr86696.C
===================================================================
--- gcc/testsuite/g++.dg/pr86696.C	(nonexistent)
+++ gcc/testsuite/g++.dg/pr86696.C	(working copy)
@@ -0,0 +1,30 @@
+/* PR tree-optimization/86696 - ICE in handle_char_store at
+   gcc/tree-ssa-strlen.c
+   { dg-do compile }
+   { dg-options "-O2 -Wall -std=c++11" } */
+
+typedef char a;
+template <typename b> struct c {
+  int d;
+  b e;
+};
+struct f;
+class g {
+public:
+  void h(c<f>);
+};
+enum i {};
+enum j : a { k, l };
+struct f {
+  i m;
+  a n;
+  a o;
+  a p;
+  j family;
+};
+void fn1() {
+  i format{};
+  f info{format, a(), 0, 4, l};
+  g dest;
+  dest.h({format, info});
+}
Index: gcc/tree-ssa-strlen.c
===================================================================
--- gcc/tree-ssa-strlen.c	(revision 263030)
+++ gcc/tree-ssa-strlen.c	(working copy)
@@ -3150,7 +3150,7 @@ handle_pointer_plus (gimple_stmt_iterator *gsi)
 static HOST_WIDE_INT
 get_min_string_length (tree rhs, bool *full_string_p)
 {
-  if (TREE_CODE (TREE_TYPE (rhs)) == INTEGER_TYPE)
+  if (INTEGRAL_TYPE_P (TREE_TYPE (rhs)))
     {
       if (tree_expr_nonzero_p (rhs))
 	{
@@ -3315,9 +3315,16 @@ handle_char_store (gimple_stmt_iterator *gsi)
 	     Otherwise, we're storing an unknown value at offset OFFSET,
 	     so need to clip the nonzero_chars to OFFSET.  */
 	  bool full_string_p = storing_all_zeros_p;
-	  HOST_WIDE_INT len = (storing_nonzero_p
-			       ? get_min_string_length (rhs, &full_string_p)
-			       : 1);
+	  HOST_WIDE_INT len = 1;
+	  if (storing_nonzero_p)
+	    {
+	      /* Try to get the minimum length of the string (or
+		 individual character) being stored.  If it fails,
+		 STORING_NONZERO_P guarantees it's at least 1.  */
+	      len = get_min_string_length (rhs, &full_string_p);
+	      if (len < 0)
+		len = 1;
+	    }
 
 	  location_t loc = gimple_location (stmt);
 	  tree oldlen = si->nonzero_chars;

committed [PATCH] fix typo in tree-ssa-strlen.c (PR 86696)

[Martin Sebor]

I committed this in r263032.

On 07/27/2018 09:53 AM, Martin Sebor wrote:
> My yesterday's change to tree-ssa-strlen.c introduced a test
> for INTEGER_TYPE where INTEGRAL_TYPE_P should have been used,
> causing a subsequent ICE when the latter was supplied.
>
> The attached patch corrects the test and also makes the subsequent
> code more robust and be prepared for the test to fail.
>
> Since the change also caused a couple pf regressions in the Ada
> test suite, if there are no objections, I will commit the patch
> later today to clear up those failures.
>
> Retested on x86_64-linux with no regressions (including in the Ada
> test suite).
>
> Martin