[PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Luis Machado]

[-- Attachment #1: Type: text/plain, Size: 951 bytes --]

Hi,

I noticed a few tests failing for ppc64. Those are related to printing 
virtual tables for objects.

Some of these failures are related to how the output from that command 
comes out in a target that uses function descriptors.

This is the regular non-function-descriptor output:

info vtbl a
vtable for 'A' @ 0x401858 (subobject @ 0x603220):
[0]: 0x400d9a <A::f()>

vtable for 'V' @ 0x401880 (subobject @ 0x603230):
[0]: 0x400ea2 <VB::vvb()>
[1]: 0x400e16 <V::vv()>

And the output for a target that does function descriptors:

info vtbl a
vtable for 'A' @ 0x10013518 (subobject @ 0x10013e20):
[0]: @0x10013b88: 0x10001c6c <A::f()>

vtable for 'V' @ 0x10013540 (subobject @ 0x10013e30):
[0]: @0x10013cd8: 0x10001f74 <VB::vvb()>
[1]: @0x10013c48: 0x10001df4 <V::vv()>

As you can see, there are additional fields for each virtual function 
pointer, and that is the address of the function descriptor.

The attached patch takes care of this.

OK?

[-- Attachment #2: vtbl_fd.diff --]
[-- Type: text/x-patch, Size: 694 bytes --]

2013-06-03  Luis Machado  <lgustavo@codesourcery.com>

	* gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra output
	from targets that use function descriptors in the virtual tables.

diff --git a/gdb/testsuite/gdb.cp/virtfunc.exp b/gdb/testsuite/gdb.cp/virtfunc.exp
index 2509cc7..4b73482 100644
--- a/gdb/testsuite/gdb.cp/virtfunc.exp
+++ b/gdb/testsuite/gdb.cp/virtfunc.exp
@@ -234,7 +234,7 @@ proc make_one_vtable_result {name args} {
     set result "vtable for '${name}' @ $hex .subobject @ $hex.:$nls"
     set count 0
     foreach func $args {
-	append result ".${count}.: $hex <$func..>${nls}"
+	append result ".${count}.:( @$hex:)? $hex <$func..>${nls}"
 	incr count
     }
 

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Tom Tromey]

>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:

Luis> 2013-06-03  Luis Machado  <lgustavo@codesourcery.com>
Luis> 	* gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra output
Luis> 	from targets that use function descriptors in the virtual tables.

Ok.  Thanks for looking at this.

Tom

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Luis Machado]

[-- Attachment #1: Type: text/plain, Size: 995 bytes --]

On 06/03/2013 06:17 PM, Tom Tromey wrote:
>>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:
>
> Luis> 2013-06-03  Luis Machado  <lgustavo@codesourcery.com>
> Luis> 	* gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra output
> Luis> 	from targets that use function descriptors in the virtual tables.
>
> Ok.  Thanks for looking at this.

Before checking in the patch, i figured out the rest of the problems.

Newer GDB's seem to have fixed a problem with displaying thunks in the 
virtual tables. Older ones did not demangle those names properly.

With that said, ppc64 uses dot symbols for those thunks, so we need to 
account for those in the testcase as well.

Here's the updated patch. I escaped dot once (\.) instead of twice. So 
hopefully this is the correct way. I often get confused with escaping in 
dejagnu.

With this fix, i see only a single failure for virtfunc.exp on ppc64. 
The other failure is more involved and i'm still chasing the root cause.

Luis


[-- Attachment #2: virtfunc_fix.diff --]
[-- Type: text/x-patch, Size: 732 bytes --]

2013-06-04  Luis Machado  <lgustavo@codesourcery.com>

	* gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra output
	from targets that use function descriptors in the virtual tables.
	Handle presence of dot symbols.

diff --git a/gdb/testsuite/gdb.cp/virtfunc.exp b/gdb/testsuite/gdb.cp/virtfunc.exp
index 2509cc7..1c2268a 100644
--- a/gdb/testsuite/gdb.cp/virtfunc.exp
+++ b/gdb/testsuite/gdb.cp/virtfunc.exp
@@ -234,7 +234,7 @@ proc make_one_vtable_result {name args} {
     set result "vtable for '${name}' @ $hex .subobject @ $hex.:$nls"
     set count 0
     foreach func $args {
-	append result ".${count}.: $hex <$func..>${nls}"
+	append result ".${count}.:( @$hex:)? $hex <(\.)?$func..>${nls}"
 	incr count
     }
 

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Luis Machado]

On 06/04/2013 10:53 AM, Luis Machado wrote:
> On 06/03/2013 06:17 PM, Tom Tromey wrote:
>>>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:
>>
>> Luis> 2013-06-03  Luis Machado  <lgustavo@codesourcery.com>
>> Luis>     * gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra
>> output
>> Luis>     from targets that use function descriptors in the virtual
>> tables.
>>
>> Ok.  Thanks for looking at this.
>
> Before checking in the patch, i figured out the rest of the problems.
>
> Newer GDB's seem to have fixed a problem with displaying thunks in the
> virtual tables. Older ones did not demangle those names properly.
>
> With that said, ppc64 uses dot symbols for those thunks, so we need to
> account for those in the testcase as well.
>
> Here's the updated patch. I escaped dot once (\.) instead of twice. So
> hopefully this is the correct way. I often get confused with escaping in
> dejagnu.
>
> With this fix, i see only a single failure for virtfunc.exp on ppc64.
> The other failure is more involved and i'm still chasing the root cause.

For convenience, here is an example of how the output looks for ppc64. 
Thunk symbols have a dot prefix.

vtable for 'D' @ 0x10013248 (subobject @ 0x10013ed0):
[0]: @0x10013c00: 0x10001d64 <.non-virtual thunk to E::vg()>
[1]: @0x10013ca8: 0x10001eec <D::vd()>

vtable for 'V' @ 0x10013290 (subobject @ 0x10013ef0):
[0]: @0x10013cd8: 0x10001f74 <VB::vvb()>
[1]: @0x10013c60: 0x10001e38 <.virtual thunk to E::vv()>

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Will Newton]

On 4 June 2013 10:03, Luis Machado <lgustavo@codesourcery.com> wrote:
> On 06/04/2013 10:53 AM, Luis Machado wrote:
>>
>> On 06/03/2013 06:17 PM, Tom Tromey wrote:
>>>>>>>>
>>>>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:
>>>
>>>
>>> Luis> 2013-06-03  Luis Machado  <lgustavo@codesourcery.com>
>>> Luis>     * gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra
>>> output
>>> Luis>     from targets that use function descriptors in the virtual
>>> tables.
>>>
>>> Ok.  Thanks for looking at this.
>>
>>
>> Before checking in the patch, i figured out the rest of the problems.
>>
>> Newer GDB's seem to have fixed a problem with displaying thunks in the
>> virtual tables. Older ones did not demangle those names properly.
>>
>> With that said, ppc64 uses dot symbols for those thunks, so we need to
>> account for those in the testcase as well.
>>
>> Here's the updated patch. I escaped dot once (\.) instead of twice. So
>> hopefully this is the correct way. I often get confused with escaping in
>> dejagnu.
>>
>> With this fix, i see only a single failure for virtfunc.exp on ppc64.
>> The other failure is more involved and i'm still chasing the root cause.
>
>
> For convenience, here is an example of how the output looks for ppc64. Thunk
> symbols have a dot prefix.
>
> vtable for 'D' @ 0x10013248 (subobject @ 0x10013ed0):
> [0]: @0x10013c00: 0x10001d64 <.non-virtual thunk to E::vg()>
> [1]: @0x10013ca8: 0x10001eec <D::vd()>
>
> vtable for 'V' @ 0x10013290 (subobject @ 0x10013ef0):
> [0]: @0x10013cd8: 0x10001f74 <VB::vvb()>
> [1]: @0x10013c60: 0x10001e38 <.virtual thunk to E::vv()>

ARM also has a similar problem, as Thumb addresses end up looking like:

[0]: 0x8e7d <non-virtual thunk to E::vg()+1>

I'm not sure if it's a problem with the test or if there is a missing
call to addr_bits_remove somewhere.

--
Will Newton
Toolchain Working Group, Linaro

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Luis Machado]

On 06/04/2013 11:10 AM, Will Newton wrote:
> On 4 June 2013 10:03, Luis Machado <lgustavo@codesourcery.com> wrote:
>> On 06/04/2013 10:53 AM, Luis Machado wrote:
>>>
>>> On 06/03/2013 06:17 PM, Tom Tromey wrote:
>>>>>>>>>
>>>>>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:
>>>>
>>>>
>>>> Luis> 2013-06-03  Luis Machado  <lgustavo@codesourcery.com>
>>>> Luis>     * gdb.cp/virtfunc.exp (make_one_vtable_result): Handle extra
>>>> output
>>>> Luis>     from targets that use function descriptors in the virtual
>>>> tables.
>>>>
>>>> Ok.  Thanks for looking at this.
>>>
>>>
>>> Before checking in the patch, i figured out the rest of the problems.
>>>
>>> Newer GDB's seem to have fixed a problem with displaying thunks in the
>>> virtual tables. Older ones did not demangle those names properly.
>>>
>>> With that said, ppc64 uses dot symbols for those thunks, so we need to
>>> account for those in the testcase as well.
>>>
>>> Here's the updated patch. I escaped dot once (\.) instead of twice. So
>>> hopefully this is the correct way. I often get confused with escaping in
>>> dejagnu.
>>>
>>> With this fix, i see only a single failure for virtfunc.exp on ppc64.
>>> The other failure is more involved and i'm still chasing the root cause.
>>
>>
>> For convenience, here is an example of how the output looks for ppc64. Thunk
>> symbols have a dot prefix.
>>
>> vtable for 'D' @ 0x10013248 (subobject @ 0x10013ed0):
>> [0]: @0x10013c00: 0x10001d64 <.non-virtual thunk to E::vg()>
>> [1]: @0x10013ca8: 0x10001eec <D::vd()>
>>
>> vtable for 'V' @ 0x10013290 (subobject @ 0x10013ef0):
>> [0]: @0x10013cd8: 0x10001f74 <VB::vvb()>
>> [1]: @0x10013c60: 0x10001e38 <.virtual thunk to E::vv()>
>
> ARM also has a similar problem, as Thumb addresses end up looking like:
>
> [0]: 0x8e7d <non-virtual thunk to E::vg()+1>
>
> I'm not sure if it's a problem with the test or if there is a missing
> call to addr_bits_remove somewhere.

Hum. I'd double check printcmd.c:build_address_symbolic to make sure it 
is doing the right thing.

I think E::vg(), E::vg()+2, E::vg()+3 is what should be displayed, correct?

Luis

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Tom Tromey]

>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:

Luis> Here's the updated patch. I escaped dot once (\.) instead of twice. So
Luis> hopefully this is the correct way. I often get confused with escaping
Luis> in dejagnu.

This is ok.  Thanks.

Tom

Re: [PATCH, c++ testsuite] Fix a few failures in gdb.cp/virtfunc.exp

[Luis Machado]

On 06/05/2013 05:36 PM, Tom Tromey wrote:
>>>>>> "Luis" == Luis Machado <lgustavo@codesourcery.com> writes:
>
> Luis> Here's the updated patch. I escaped dot once (\.) instead of twice. So
> Luis> hopefully this is the correct way. I often get confused with escaping
> Luis> in dejagnu.
>
> This is ok.  Thanks.

Thanks. Committed now.

Luis