Re: [RFC] [PATCH] Return intptr when dereferencing integer

[Denis Revunov <rnovds@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 3016 bytes --]

>>>>> "Tom" == Tom Tromey via Gdb-patches <gdb-patches@sourceware.org>
writes:

Tom> Does this impact any test?

I cannot be 100% sure because i get slightly different test results on
consecutive runs even without any changes, but it seems the only failures
appearing in gdb.sum diff (with/without the patch) are timeouts and some
server stuff, and there's like 2-5 of them. So we're probably fine.

Tom> If not, can a test be written for it?

Sure, see a diff in the end.

Tom> I guess from context the scenario is something like: "print *23",
Tom> and here gdb acts like you typed "print *(int*)23"?

That's right, and my point is to make it act like "print *(intptr_t*)23".
Because int is intptr_t everywhere except 64-bit.

Here's a diff for the test:

---
 gdb/testsuite/gdb.base/pointers.c   | 5 ++++-
 gdb/testsuite/gdb.base/pointers.exp | 4 ++++
 2 files changed, 8 insertions(+), 1 deletion(-)

diff --git a/gdb/testsuite/gdb.base/pointers.c
b/gdb/testsuite/gdb.base/pointers.c
index 96f2b52f5e8..d3a9689b46a 100644
--- a/gdb/testsuite/gdb.base/pointers.c
+++ b/gdb/testsuite/gdb.base/pointers.c
@@ -1,4 +1,4 @@
-
+#include <stdint.h>
 #if !defined (__STDC__) && !defined (_AIX)
 #define signed  /**/
 #endif
@@ -71,6 +71,8 @@ float ** ptr_to_ptr_to_float;

 int y;

+const intptr_t intptr_max = INTPTR_MAX;
+intptr_t intptr_p;

 typedef long k[5];

@@ -170,6 +172,7 @@ void dummy()

   v_float_array[0] = v_float;
   v_double_array[0] = v_double;
+  intptr_p = (intptr_t)&intptr_max;

 }

diff --git a/gdb/testsuite/gdb.base/pointers.exp
b/gdb/testsuite/gdb.base/pointers.exp
index d72fa5ad613..5787b873cca 100644
--- a/gdb/testsuite/gdb.base/pointers.exp
+++ b/gdb/testsuite/gdb.base/pointers.exp
@@ -202,6 +202,10 @@ gdb_test "print *( *(matrix+1) +2)" " = 5" \
 gdb_test "print **ptr_to_ptr_to_float" " = 100" \
     "print through ptr to ptr"

+# when we attempt to dereference an integer,
+# it should be treated as a pointer to intptr_t.
+gdb_test "print *intptr_p == intptr_max" " = 1" "deref size is intptr"
+
 # tests for pointers
 # with elementary type variables and pointers.
 #
-- 
2.41.0



On Thu, 27 Jul 2023 at 19:15, Tom Tromey <tom@tromey.com> wrote:

> >>>>> "Denis" == Denis Revunov via Gdb-patches <gdb-patches@sourceware.org>
> writes:
>
> Denis> The code which makes default dereference type an int was intruduced
> in
> Denis> gdb-3.4(1c997a4ae869) in 1989, 10 years before AMD64 was released.
> With
> Denis> current adoption of 64-bit it makes sense to change default
> dereference
> Denis> size to match the native address size, which will enable us to
> easily
> Denis> inspect .got entries and follow chains of pointers on 64-bit systems
> Denis> without excessive casting. This change won't affect 32-and-less bit
> Denis> systems since intptr is the same as int there.
>
> Does this impact any test?
> If not, can a test be written for it?
>
> I guess from context the scenario is something like: "print *23",
> and here gdb acts like you typed "print *(int*)23"?
>
> Tom
>