This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH] Don't override operator new if GDB is built with -fsanitize=address
On 10/25/2016 11:38 AM, Pedro Alves wrote:
> How about this follow up?
On 10/25/2016 11:38 AM, Pedro Alves wrote:
> +++ b/gdb/common/new-op.c
> @@ -33,6 +33,12 @@
> new-handler function instead (std::set_new_handler) because we want
> to catch allocation errors from within global constructors too.
> + Skip overriding if building with -fsanitize=address though.
> + Address sanitizer wants to override operator new/delete too in
Bah, looks like "edit as new" with Thunderbird messed
up the patch... Here it is again...
>From e58eb16a5f27b1ee39c45642a80da7364763a07b Mon Sep 17 00:00:00 2001
From: Pedro Alves <palves@redhat.com>
Date: Tue, 25 Oct 2016 11:20:03 +0100
Subject: [PATCH] new-op.c: Add comment about -fsanitize=address
gdb/ChangeLog:
2016-10-25 Pedro Alves <palves@redhat.com>
* common/new-op.c: Add comment about -fsanitize=address.
---
gdb/common/new-op.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/gdb/common/new-op.c b/gdb/common/new-op.c
index f04c5cb..1eb4f94 100644
--- a/gdb/common/new-op.c
+++ b/gdb/common/new-op.c
@@ -33,6 +33,12 @@
new-handler function instead (std::set_new_handler) because we want
to catch allocation errors from within global constructors too.
+ Skip overriding if building with -fsanitize=address though.
+ Address sanitizer wants to override operator new/delete too in
+ order to detect malloc+delete and new+free mismatches. Our
+ versions would mask out ASan's, with the result of losing that
+ useful mismatch detection.
+
Note that C++ implementations could either have their throw
versions call the nothrow versions (libstdc++), or the other way
around (clang/libc++). For that reason, we replace both throw and
--
2.5.5