This is the mail archive of the
gdb-patches@sourceware.org
mailing list for the GDB project.
Re: [PATCH 2/2] Support setting breakpoints on C++ method pointers
- From: Patrick Palka <patrick at parcs dot ath dot cx>
- To: gdb-patches at sourceware dot org
- Cc: Patrick Palka <patrick at parcs dot ath dot cx>
- Date: Sat, 15 Nov 2014 22:57:19 -0500
- Subject: Re: [PATCH 2/2] Support setting breakpoints on C++ method pointers
- Authentication-results: sourceware.org; auth=none
- References: <1411355257-10861-1-git-send-email-patrick at parcs dot ath dot cx> <1411429189-17235-1-git-send-email-patrick at parcs dot ath dot cx> <CA+C-WL_V=9S2Ok7kkB=ev5==Qwo6qvMahtvRsN_HisiF7hrC4g at mail dot gmail dot com>
On Tue, Sep 30, 2014 at 5:49 PM, Patrick Palka <patrick@parcs.ath.cx> wrote:
> On Mon, Sep 22, 2014 at 7:39 PM, Patrick Palka <patrick@parcs.ath.cx> wrote:
>> (Oops, I accidentally sent the older version of this patch a 2nd time.
>> Sorry for the noise.)
>>
>> { v2: I noticed that non-virtual method pointers are also affected by
>> this commit. That is, breaking on non-virtual method pointers now also
>> works whereas previously it didn't. So I rewrote the commit message and
>> augmented the testcase to also test non-virtual method pointers. I also
>> tweaked the documentation for the function cplus_method_ptr_to_value(). }
>>
>> -- >8 --
>>
>> This patch adds support for setting breakpoints on C++ method pointers,
>> both virtual and non-virtual. For example:
>>
>> struct x
>> {
>> virtual void f ();
>> void g ();
>> }
>>
>> void x::f () { }
>> void x::g () { }
>>
>> struct y : x
>> {
>> }
>>
>> (gdb) break *&y::f
>> Value can't be converted to integer.
>> (gdb) break *&y::g
>> Value can't be converted to integer.
>>
>> Breaking on these expressions doesn't currently work because GDB doesn't
>> know how to convert a METHODPTR to an address in the function
>> value_as_address(). We have to teach value_as_address() how to extract
>> a symbolic address out from a METHODPTR in order for the above example
>> to work.
>>
>> This patch tweaks value_as_address() to call cplus_method_ptr_to_value()
>> in order to extract a pointer value out of a METHODPTR. The latter
>> function does most of the work, but it needs a few tweaks. Firstly,
>> this patch makes the first argument of gnuv3_method_ptr_to_value(), i.e.
>> the argument corresponding to a "this" object pointer, optional.
>> Secondly, when the "this" pointer is omitted and a virtual method
>> pointer is passed in then we attempt to extract a pointer value by doing
>> a lookup of the virtual method pointer's symbolic name in the symbol
>> table.
>>
>> Tested on x86_64-unknown-linux-gnu.
>>
>> gdb/ChangeLog
>> * cp-abi.h (cplus_method_ptr_to_value): Document behavior for a
>> THIS_P that's NULL.
>> * gnu-v3-abi.c (gnuv3_method_ptr_to_value): Support method
>> resolution without a given "this" object.
>> * value.c (value_as_address): Handle TYPE_CODE_METHODPTR values.
>>
>> gdb/testsuite/ChangeLog
>> * gdb.cp/method-ptr.exp: Add tests for breaking on non-virtual
>> method pointers.
>> (check_virtual_method_ptr_resolution): New parameter VIRTUAL.
>> Also test that setting a breakpoint on the given method pointer
>> works correctly.
>> * gdb.cp/method-ptr.cc: Introduce a few non-virtual methods
>> to test.
>> ---
>> gdb/cp-abi.h | 4 ++-
>> gdb/gnu-v3-abi.c | 65 +++++++++++++++++++++++++------------
>> gdb/testsuite/gdb.cp/method-ptr.cc | 6 ++++
>> gdb/testsuite/gdb.cp/method-ptr.exp | 23 +++++++++++--
>> gdb/value.c | 8 +++++
>> 5 files changed, 81 insertions(+), 25 deletions(-)
>>
>> diff --git a/gdb/cp-abi.h b/gdb/cp-abi.h
>> index 7d4b7f3..1f38605 100644
>> --- a/gdb/cp-abi.h
>> +++ b/gdb/cp-abi.h
>> @@ -162,7 +162,9 @@ void cplus_print_method_ptr (const gdb_byte *contents,
>> int cplus_method_ptr_size (struct type *to_type);
>>
>> /* Return the method which should be called by applying METHOD_PTR to
>> - *THIS_P, and adjust *THIS_P if necessary. */
>> + *THIS_P, and adjust *THIS_P if necessary. If THIS_P is NULL then
>> + return the method that would be called if METHOD_PTR was applied
>> + to an object of METHOD_PTR's domain type (e.g. the type X in &X::f). */
>> struct value *cplus_method_ptr_to_value (struct value **this_p,
>> struct value *method_ptr);
>>
>> diff --git a/gdb/gnu-v3-abi.c b/gdb/gnu-v3-abi.c
>> index ccb0be6..718694b 100644
>> --- a/gdb/gnu-v3-abi.c
>> +++ b/gdb/gnu-v3-abi.c
>> @@ -719,36 +719,59 @@ gnuv3_method_ptr_to_value (struct value **this_p, struct value *method_ptr)
>> gdbarch = get_type_arch (domain_type);
>> vbit = gnuv3_decode_method_ptr (gdbarch, contents, &ptr_value, &adjustment);
>>
>> - /* First convert THIS to match the containing type of the pointer to
>> - member. This cast may adjust the value of THIS. */
>> - *this_p = value_cast (final_type, *this_p);
>> + if (this_p != NULL)
>> + {
>> + /* First convert THIS to match the containing type of the pointer to
>> + member. This cast may adjust the value of THIS. */
>> + *this_p = value_cast (final_type, *this_p);
>>
>> - /* Then apply whatever adjustment is necessary. This creates a somewhat
>> - strange pointer: it claims to have type FINAL_TYPE, but in fact it
>> - might not be a valid FINAL_TYPE. For instance, it might be a
>> - base class of FINAL_TYPE. And if it's not the primary base class,
>> - then printing it out as a FINAL_TYPE object would produce some pretty
>> - garbage.
>> + /* Then apply whatever adjustment is necessary. This creates a somewhat
>> + strange pointer: it claims to have type FINAL_TYPE, but in fact it
>> + might not be a valid FINAL_TYPE. For instance, it might be a
>> + base class of FINAL_TYPE. And if it's not the primary base class,
>> + then printing it out as a FINAL_TYPE object would produce some pretty
>> + garbage.
>>
>> - But we don't really know the type of the first argument in
>> - METHOD_TYPE either, which is why this happens. We can't
>> - dereference this later as a FINAL_TYPE, but once we arrive in the
>> - called method we'll have debugging information for the type of
>> - "this" - and that'll match the value we produce here.
>> + But we don't really know the type of the first argument in
>> + METHOD_TYPE either, which is why this happens. We can't
>> + dereference this later as a FINAL_TYPE, but once we arrive in the
>> + called method we'll have debugging information for the type of
>> + "this" - and that'll match the value we produce here.
>>
>> - You can provoke this case by casting a Base::* to a Derived::*, for
>> - instance. */
>> - *this_p = value_cast (builtin_type (gdbarch)->builtin_data_ptr, *this_p);
>> - *this_p = value_ptradd (*this_p, adjustment);
>> - *this_p = value_cast (final_type, *this_p);
>> + You can provoke this case by casting a Base::* to a Derived::*, for
>> + instance. */
>> + *this_p = value_cast (builtin_type (gdbarch)->builtin_data_ptr, *this_p);
>> + *this_p = value_ptradd (*this_p, adjustment);
>> + *this_p = value_cast (final_type, *this_p);
>> + }
>>
>> if (vbit)
>> {
>> LONGEST voffset;
>>
>> voffset = ptr_value / TYPE_LENGTH (vtable_ptrdiff_type (gdbarch));
>> - return gnuv3_get_virtual_fn (gdbarch, value_ind (*this_p),
>> - method_type, voffset);
>> +
>> + /* If we don't have a "this" object to apply the method pointer to,
>> + then retrieve the value of the virtual method by looking up its
>> + symbolic name within the symbol table. */
>> + if (this_p == NULL)
>> + {
>> + const char *physname;
>> + struct symbol *sym;
>> +
>> + physname = gnuv3_find_method_in (domain_type, voffset, adjustment);
>> + if (physname == NULL)
>> + return NULL;
>> +
>> + sym = lookup_symbol (physname, NULL, VAR_DOMAIN, NULL);
>> + if (sym == NULL)
>> + return NULL;
>> +
>> + return value_of_variable (sym, NULL);
>> + }
>> + else
>> + return gnuv3_get_virtual_fn (gdbarch, value_ind (*this_p),
>> + method_type, voffset);
>> }
>> else
>> return value_from_pointer (lookup_pointer_type (method_type), ptr_value);
>> diff --git a/gdb/testsuite/gdb.cp/method-ptr.cc b/gdb/testsuite/gdb.cp/method-ptr.cc
>> index db47484..4e1524a 100644
>> --- a/gdb/testsuite/gdb.cp/method-ptr.cc
>> +++ b/gdb/testsuite/gdb.cp/method-ptr.cc
>> @@ -20,11 +20,14 @@ struct x
>> virtual void f ();
>> virtual void g ();
>> virtual void h ();
>> +
>> + void a ();
>> };
>>
>> void x::f () { }
>> void x::g () { }
>> void x::h () { }
>> +void x::a () { }
>>
>> struct y : x
>> {
>> @@ -32,11 +35,14 @@ struct y : x
>>
>> virtual void j ();
>> virtual void k ();
>> +
>> + void b ();
>> };
>>
>> void y::f () { }
>> void y::j () { }
>> void y::k () { }
>> +void y::b () { }
>>
>> struct z : y
>> {
>> diff --git a/gdb/testsuite/gdb.cp/method-ptr.exp b/gdb/testsuite/gdb.cp/method-ptr.exp
>> index 732b861..ca6be4b 100644
>> --- a/gdb/testsuite/gdb.cp/method-ptr.exp
>> +++ b/gdb/testsuite/gdb.cp/method-ptr.exp
>> @@ -33,23 +33,38 @@ if ![test_debug_format "DWARF 2"] {
>> return 0
>> }
>>
>> -# Check that the virtual method pointer NAME resolves to symbol SYMBOL.
>> -proc check_virtual_method_ptr_resolution { name symbol } {
>> +# Check that the method pointer NAME resolves to symbol SYMBOL. Set VIRTUAL
>> +# to 1 if NAME is a virtual method pointer (default), 0 otherwise.
>> +proc check_virtual_method_ptr_resolution { name symbol {virtual 1} } {
>> global decimal
>>
>> # Printing the expression &NAME should show the resolved symbol SYMBOL.
>> - gdb_test "print &$name" "\\$$decimal = &virtual $symbol\\(\\)\\s"
>> + if {$virtual != 0} {
>> + gdb_test "print &$name" "\\s&virtual $symbol\\(\\)\\s"
>> + } else {
>> + gdb_test "print &$name" "\\s<$symbol\\(\\)>\\s"
>> + }
>> +
>> + # Breaking on the expression &NAME should create a breakpoint on the symbol
>> + # SYMBOL.
>> + set breakpoint_line [gdb_get_line_number $symbol]
>> + gdb_test "break *&$name" \
>> + "Breakpoint $decimal at .*, line $breakpoint_line\\.\\s"
>> + delete_breakpoints
>> }
>>
>> check_virtual_method_ptr_resolution "x::f" "x::f"
>> check_virtual_method_ptr_resolution "x::g" "x::g"
>> check_virtual_method_ptr_resolution "x::h" "x::h"
>> +check_virtual_method_ptr_resolution "x::a" "x::a" 0
>>
>> check_virtual_method_ptr_resolution "y::f" "y::f"
>> check_virtual_method_ptr_resolution "y::g" "x::g"
>> check_virtual_method_ptr_resolution "y::h" "x::h"
>> check_virtual_method_ptr_resolution "y::j" "y::j"
>> check_virtual_method_ptr_resolution "y::k" "y::k"
>> +check_virtual_method_ptr_resolution "y::a" "x::a" 0
>> +check_virtual_method_ptr_resolution "y::b" "y::b" 0
>>
>> check_virtual_method_ptr_resolution "z::f" "y::f"
>> check_virtual_method_ptr_resolution "z::g" "z::g"
>> @@ -58,3 +73,5 @@ check_virtual_method_ptr_resolution "z::j" "z::j"
>> check_virtual_method_ptr_resolution "z::k" "y::k"
>> check_virtual_method_ptr_resolution "z::l" "z::l"
>> check_virtual_method_ptr_resolution "z::m" "z::m"
>> +check_virtual_method_ptr_resolution "z::a" "x::a" 0
>> +check_virtual_method_ptr_resolution "z::b" "y::b" 0
>> diff --git a/gdb/value.c b/gdb/value.c
>> index fdc8858d..63ff363 100644
>> --- a/gdb/value.c
>> +++ b/gdb/value.c
>> @@ -2639,6 +2639,14 @@ value_as_address (struct value *val)
>> return gdbarch_addr_bits_remove (gdbarch, value_as_long (val));
>> #else
>>
>> + if (TYPE_CODE (value_type (val)) == TYPE_CODE_METHODPTR)
>> + {
>> + val = cplus_method_ptr_to_value (NULL, val);
>> + if (val == NULL)
>> + error (_("Method pointer can't be converted to an address."));
>> + }
>> +
>> +
>> /* There are several targets (IA-64, PowerPC, and others) which
>> don't represent pointers to functions as simply the address of
>> the function's entry point. For example, on the IA-64, a
>> --
>> 2.1.1.273.g97b8860
>>
>
> Ping.
Ping.