All of lore.kernel.org
 help / color / mirror / Atom feed
From: Nicolai Stange <nicstange@gmail.com>
To: linux-sparse@vger.kernel.org
Cc: Nicolai Stange <nicstange@gmail.com>,
	Christopher Li <sparse@chrisli.org>,
	Josh Triplett <josh@joshtriplett.org>,
	Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
Subject: [PATCH v2 05/13] expression: examine constness of conditionals at evaluation only
Date: Mon, 25 Jan 2016 15:55:09 +0100	[thread overview]
Message-ID: <878u3dg14i.fsf@gmail.com> (raw)
In-Reply-To: <87twm1g1go.fsf@gmail.com> (Nicolai Stange's message of "Mon, 25 Jan 2016 15:47:51 +0100")

Currently, the propagation of expressions' constness flags through
conditional expressions is done in two steps:
- Several flags are speculatively set at expression parsing time
- and possibly cleared again at evaluation time.

Set aside this unfortunate split of code, the early propagation of
constness flags is not able to recognize constant expressions such as
  0 ? __builtin_choose_expr(0, 0, 0) : 0
  0 ? 0 : __builtin_choose_expr(0, 0, 0)
since the final expression to be thrown into the conditional
expression is known only after evaluation.

Move the whole calculation of conditional expressions' constness flags
to the evaluation phase.

Introduce support for tracking arithmetic constness propagation through
conditional expressions.

Signed-off-by: Nicolai Stange <nicstange@gmail.com>
---
 evaluate.c                         | 18 ++++++++----------
 expression.c                       |  7 -------
 validation/constexpr-conditional.c | 34 ++++++++++++++++++++++++++++++++++
 3 files changed, 42 insertions(+), 17 deletions(-)
 create mode 100644 validation/constexpr-conditional.c

diff --git a/evaluate.c b/evaluate.c
index 5138d40..97da51d 100644
--- a/evaluate.c
+++ b/evaluate.c
@@ -1114,16 +1114,14 @@ static struct symbol *evaluate_conditional_expression(struct expression *expr)
 		true = &expr->cond_true;
 	}
 
-	if (expr->flags) {
-		unsigned char flags_mask = 0;
-		unsigned char flags;
-
-		expr_set_flag(&flags_mask, EXPR_FLAG_INT_CONST_EXPR);
-		flags = (expr->conditional->flags & flags_mask);
-		flags &= (*true)->flags & expr->cond_false->flags;
-		if (!flags)
-			expr->flags = EXPR_FLAG_NONE;
-	}
+	expr->flags = (expr->conditional->flags & (*true)->flags &
+		expr->cond_false->flags);
+	expr_flags_decay_consts(&expr->flags);
+	/*
+	 * A conditional operator never yields an address constant
+	 * [6.6(9)].
+	 */
+	expr_clear_flag(&expr->flags, EXPR_FLAG_ADDR_CONST_EXPR);
 
 	lclass = classify_type(ltype, &ltype);
 	rclass = classify_type(rtype, &rtype);
diff --git a/expression.c b/expression.c
index 377834e..792c2a5 100644
--- a/expression.c
+++ b/expression.c
@@ -858,13 +858,6 @@ struct token *conditional_expression(struct token *token, struct expression **tr
 		token = parse_expression(token->next, &expr->cond_true);
 		token = expect(token, ':', "in conditional expression");
 		token = conditional_expression(token, &expr->cond_false);
-		if (expr->left && expr->cond_false) {
-			expr->flags = expr->left->flags &
-				expr->cond_false->flags;
-			if (expr->cond_true)
-				expr->flags &= expr->cond_true->flags;
-			expr_flags_decay_consts(&expr->flags);
-		}
 	}
 	return token;
 }
diff --git a/validation/constexpr-conditional.c b/validation/constexpr-conditional.c
new file mode 100644
index 0000000..a3331b3
--- /dev/null
+++ b/validation/constexpr-conditional.c
@@ -0,0 +1,34 @@
+static int a[] = {
+	[0 ? : 0] = 0,						// OK
+	[1 ? : 0] = 0,						// OK
+	[0 ? 0 : 0] = 0,					// OK
+	[1 ? 0 : 0] = 0,					// OK
+	[0 ? 0 : __builtin_choose_expr(0, 0, 0)] = 0,		// OK
+	[1 ? __builtin_choose_expr(0, 0, 0) : 0] = 0,		// OK
+	[0 ? __builtin_choose_expr(0, 0, 0) : 0] = 0,		// OK
+	[1 ? 1 : __builtin_choose_expr(0, 0, 0)] = 0,		// OK
+	[__builtin_choose_expr(0, 0, 0) ? : 0] = 0,		// OK
+	[__builtin_choose_expr(0, 0, 1) ? : 0] = 0,		// OK
+	[0. ? : 0] = 0,					// KO
+	[0 ? 0. : 0] = 0,					// KO
+	[1 ? : 0.] = 0,					// KO
+	[__builtin_choose_expr(0, 0., 0) ? : 0] = 0,		// OK
+	[__builtin_choose_expr(0, 0, 0.) ? : 0] = 0,		// KO
+	[0 ? __builtin_choose_expr(0, 0., 0) : 0] = 0,		// OK
+	[0 ? __builtin_choose_expr(0, 0, 0.) : 0] = 0,		// KO
+	[1 ? 0 : __builtin_choose_expr(0, 0., 0)] = 0,		// OK
+	[1 ? 0 : __builtin_choose_expr(0, 0, 0.)] = 0,		// KO
+};
+
+/*
+ * check-name: Expression constness propagation in conditional expressions
+ *
+ * check-error-start
+constexpr-conditional.c:12:13: error: bad constant expression
+constexpr-conditional.c:13:19: error: bad constant expression
+constexpr-conditional.c:14:12: error: bad constant expression
+constexpr-conditional.c:16:42: error: bad constant expression
+constexpr-conditional.c:18:48: error: bad constant expression
+constexpr-conditional.c:20:14: error: bad constant expression
+ * check-error-end
+ */
-- 
2.7.0


  parent reply	other threads:[~2016-01-25 14:55 UTC|newest]

Thread overview: 43+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-01-25 14:47 [PATCH v2 00/13] improve constexpr handling Nicolai Stange
2016-01-25 14:49 ` [PATCH v2 01/13] expression: introduce additional expression constness tracking flags Nicolai Stange
2016-01-25 21:51   ` Luc Van Oostenryck
2016-01-26 15:26     ` Nicolai Stange
2016-01-26 15:37       ` Nicolai Stange
2016-01-25 14:51 ` [PATCH v2 02/13] expression: examine constness of casts at evaluation only Nicolai Stange
2016-01-25 22:02   ` Luc Van Oostenryck
2016-01-26 16:11     ` Nicolai Stange
2016-01-25 14:52 ` [PATCH v2 03/13] expression: examine constness of binops and alike " Nicolai Stange
2016-01-26  0:14   ` Luc Van Oostenryck
2016-01-26 15:50     ` Nicolai Stange
2016-01-26 17:24       ` Luc Van Oostenryck
2016-01-27 10:42         ` Nicolai Stange
2016-01-27 18:00           ` Luc Van Oostenryck
2016-01-26  0:59   ` Luc Van Oostenryck
2016-01-25 14:53 ` [PATCH v2 04/13] expression: examine constness of preops " Nicolai Stange
2016-01-26  1:10   ` Luc Van Oostenryck
2016-01-25 14:55 ` Nicolai Stange [this message]
2016-01-26  1:16   ` [PATCH v2 05/13] expression: examine constness of conditionals " Luc Van Oostenryck
2016-01-25 14:56 ` [PATCH v2 06/13] expression, evaluate: add support for recognizing address constants Nicolai Stange
2016-01-26  1:27   ` Luc Van Oostenryck
2016-01-26  3:10   ` Luc Van Oostenryck
2016-01-25 14:57 ` [PATCH v2 07/13] evaluate: check static storage duration objects' intializers' constness Nicolai Stange
2016-01-26  1:42   ` Luc Van Oostenryck
2016-01-26 16:08     ` Nicolai Stange
2016-01-26 17:56       ` Luc Van Oostenryck
2016-01-26 20:18         ` Luc Van Oostenryck
2016-02-01  3:00     ` Nicolai Stange
2016-01-25 14:59 ` [PATCH v2 08/13] expression: recognize references to labels as address constants Nicolai Stange
2016-01-26  1:45   ` Luc Van Oostenryck
2016-01-25 15:00 ` [PATCH v2 09/13] expression: examine constness of __builtin_offsetof at evaluation only Nicolai Stange
2016-01-26  1:57   ` Luc Van Oostenryck
2016-02-01  3:06     ` Nicolai Stange
2016-01-25 15:02 ` [PATCH v2 10/13] symbol: flag builtins constant_p, safe_p and warning as constexprs Nicolai Stange
2016-01-26  2:00   ` Luc Van Oostenryck
2016-01-25 15:03 ` [PATCH v2 11/13] evaluate: relax some constant expression rules for pointer expressions Nicolai Stange
2016-01-26  2:05   ` Luc Van Oostenryck
2016-01-25 15:04 ` [PATCH v2 12/13] expression, evaluate: support compound literals as address constants Nicolai Stange
2016-01-26  2:07   ` Luc Van Oostenryck
2016-01-25 15:05 ` [PATCH v2 13/13] symbol: do not inherit storage modifiers from base types at examination Nicolai Stange
2016-01-26  2:54   ` Luc Van Oostenryck
2016-01-25 21:01 ` [PATCH v2 00/13] improve constexpr handling Luc Van Oostenryck
2016-01-25 21:26   ` Nicolai Stange

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=878u3dg14i.fsf@gmail.com \
    --to=nicstange@gmail.com \
    --cc=josh@joshtriplett.org \
    --cc=linux-sparse@vger.kernel.org \
    --cc=luc.vanoostenryck@gmail.com \
    --cc=sparse@chrisli.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.