Patchwork Don't track useless expressions in -Wsequence-point (PR c++/46401)

login
register
mail settings
Submitter Jakub Jelinek
Date Nov. 15, 2010, 8:25 p.m.
Message ID <20101115202533.GF29412@tyan-ft48-01.lab.bos.redhat.com>
Download mbox | patch
Permalink /patch/71282/
State New
Headers show

Comments

Jakub Jelinek - Nov. 15, 2010, 8:25 p.m.
Hi!

-Wsequence-point algorithm is quadratic (both in memory and compile time),
so the fewer things that can't be usefully warned about we track the better.

CALL_EXPRs that aren't ECF_CONST will never pass operand_equal_p, so
it doesn't make sense to track them - they will not be warned about
(lvalue_p returns true if the CALL_EXPR returns REFERENCE_TYPE).

Similarly, STRING_CSTs, while they are lvalues, can't be written into
and thus will not be ever reported by -Wsequence-point.

With these two changes the testcase compiles instantly (as opposed
after several minutes).

Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

2010-11-15  Jakub Jelinek  <jakub@redhat.com>

	PR c++/46401
	* c-common.c (warning_candidate_p): Don't track non-const calls
	or STRING_CSTs.

	* g++.dg/warn/Wsequence-point-3.C: New test.


	Jakub
Joseph S. Myers - Nov. 15, 2010, 8:42 p.m.
On Mon, 15 Nov 2010, Jakub Jelinek wrote:

> Hi!
> 
> -Wsequence-point algorithm is quadratic (both in memory and compile time),
> so the fewer things that can't be usefully warned about we track the better.
> 
> CALL_EXPRs that aren't ECF_CONST will never pass operand_equal_p, so
> it doesn't make sense to track them - they will not be warned about
> (lvalue_p returns true if the CALL_EXPR returns REFERENCE_TYPE).
> 
> Similarly, STRING_CSTs, while they are lvalues, can't be written into
> and thus will not be ever reported by -Wsequence-point.
> 
> With these two changes the testcase compiles instantly (as opposed
> after several minutes).
> 
> Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk?

OK in the absence of any C++ maintainer objections within 24 hours.

Patch

--- gcc/c-family/c-common.c.jj	2010-11-15 09:27:54.000000000 +0100
+++ gcc/c-family/c-common.c	2010-11-15 16:16:07.000000000 +0100
@@ -2324,10 +2324,26 @@  warn_for_collisions (struct tlist *list)
 static int
 warning_candidate_p (tree x)
 {
-  /* !VOID_TYPE_P (TREE_TYPE (x)) is workaround for cp/tree.c
+  if (DECL_P (x) && DECL_ARTIFICIAL (x))
+    return 0;
+
+  /* VOID_TYPE_P (TREE_TYPE (x)) is workaround for cp/tree.c
      (lvalue_p) crash on TRY/CATCH. */
-  return !(DECL_P (x) && DECL_ARTIFICIAL (x))
-    && TREE_TYPE (x) && !VOID_TYPE_P (TREE_TYPE (x)) && lvalue_p (x);
+  if (TREE_TYPE (x) == NULL_TREE || VOID_TYPE_P (TREE_TYPE (x)))
+    return 0;
+
+  if (!lvalue_p (x))
+    return 0;
+
+  /* No point to track non-const calls, they will never satisfy
+     operand_equal_p.  */
+  if (TREE_CODE (x) == CALL_EXPR && (call_expr_flags (x) & ECF_CONST) == 0)
+    return 0;
+
+  if (TREE_CODE (x) == STRING_CST)
+    return 0;
+
+  return 1;
 }
 
 /* Return nonzero if X and Y appear to be the same candidate (or NULL) */
--- gcc/testsuite/g++.dg/warn/Wsequence-point-3.C.jj	2010-11-15 17:15:42.000000000 +0100
+++ gcc/testsuite/g++.dg/warn/Wsequence-point-3.C	2010-11-15 16:35:35.000000000 +0100
@@ -0,0 +1,20 @@ 
+// PR c++/46401
+// { dg-do compile }
+// { dg-options "-Wsequence-point" }
+
+struct S
+{
+  S ();
+  S &operator<< (const char *);
+  S (const S &);
+};
+
+#define N1(n) << #n
+#define N2(n) N1(n)
+#define N3(n) N2(n##0) N2(n##1) N2(n##2) N2(n##3) N2(n##4) \
+	      N2(n##5) N2(n##6) N2(n##7) N2(n##8) N2(n##9)
+#define N4(n) N3(n##0) N3(n##1) N3(n##2) N3(n##3) N3(n##4) \
+	      N3(n##5) N3(n##6) N3(n##7) N3(n##8) N3(n##9)
+#define N5(n) N4(n##0) N4(n##1) N4(n##2) N4(n##3) N4(n##4) \
+	      N4(n##5) N4(n##6) N4(n##7) N4(n##8) N4(n##9)
+S s = S () N5(a) N5(b);