Type-lookup by name for type in namespace fails with -fdebug-types-section after 93b8bea

Type-lookup by name for type in namespace fails with -fdebug-types-section after 93b8bea

[Sterling Augustine]

Hi Tom,

We have found a strange error with -fdebug-types-section, and as best I can
tell, your commit 93b8bea4143cafae79076076c64aaa4c46a9b73c is the culprit,
although I don't understand how or why.

In any event, before this change, the test-case below works fine, and after
it, lookup of a type in a namespace by name fails. In particular, "ptype
hide::Bar" should find the type. But it fails both inside python with
gdb.lookup_type and with normal commands.

Any idea what is happening? Reproducer follows.

$ g++ -v
...
gcc version 9.2.1 20191102 (Debian 9.2.1-17)
$ cat b.cpp
namespace hide { struct Bar {  int y; }; }
struct Foo {  hide::Bar* b; };

inline void x() { hide::Bar b; }

static Foo aFoo;

void * f() {
  return aFoo.b;
}

int main() {
  f();
  return 0;
}
$ g++ -fdebug-types-section b.cpp -g
$ ./build/gdb/gdb -ex "b f" -ex run -ex "ptype hide::Bar" ./a.out
...
GNU gdb (GDB) 8.2.50.20180720-git
...
Reading symbols from ./a.out...done.
Breakpoint 1 at 0x1129: file b.cpp, line 11.
Starting program: /usr/local/google/home/saugustine/gdb/a.out

Breakpoint 1, Python Exception <type 'exceptions.NameError'> Installation
error: gdb.execute_unwinders function is missing:
f () at b.cpp:11
11  return aFoo.b;
No symbol "Bar" in namespace "hide".
(gdb)

Re: Type-lookup by name for type in namespace fails with -fdebug-types-section after 93b8bea

[Sterling Augustine]

[Updating Tom Tromey's address.]

On Mon, Mar 16, 2020 at 10:16 AM Sterling Augustine <saugustine@google.com>
wrote:

> Hi Tom,
>
> We have found a strange error with -fdebug-types-section, and as best I
> can tell, your commit 93b8bea4143cafae79076076c64aaa4c46a9b73c is the
> culprit, although I don't understand how or why.
>
> In any event, before this change, the test-case below works fine, and
> after it, lookup of a type in a namespace by name fails. In particular, "ptype
> hide::Bar" should find the type. But it fails both inside python with
> gdb.lookup_type and with normal commands.
>
> Any idea what is happening? Reproducer follows.
>
> $ g++ -v
> ...
> gcc version 9.2.1 20191102 (Debian 9.2.1-17)
> $ cat b.cpp
> namespace hide { struct Bar {  int y; }; }
> struct Foo {  hide::Bar* b; };
>
> inline void x() { hide::Bar b; }
>
> static Foo aFoo;
>
> void * f() {
>   return aFoo.b;
> }
>
> int main() {
>   f();
>   return 0;
> }
> $ g++ -fdebug-types-section b.cpp -g
> $ ./build/gdb/gdb -ex "b f" -ex run -ex "ptype hide::Bar" ./a.out
> ...
> GNU gdb (GDB) 8.2.50.20180720-git
> ...
> Reading symbols from ./a.out...done.
> Breakpoint 1 at 0x1129: file b.cpp, line 11.
> Starting program: /usr/local/google/home/saugustine/gdb/a.out
>
> Breakpoint 1, Python Exception <type 'exceptions.NameError'> Installation
> error: gdb.execute_unwinders function is missing:
> f () at b.cpp:11
> 11  return aFoo.b;
> No symbol "Bar" in namespace "hide".
> (gdb)
>

Re: Type-lookup by name for type in namespace fails with -fdebug-types-section after 93b8bea

[Sterling Augustine]

On Mon, Mar 16, 2020 at 10:16 AM Sterling Augustine
<saugustine@google.com> wrote:
>
> Hi Tom,
>
> We have found a strange error with -fdebug-types-section, and as best I can tell, your commit 93b8bea4143cafae79076076c64aaa4c46a9b73c is the culprit, although I don't understand how or why.
>
> In any event, before this change, the test-case below works fine, and after it, lookup of a type in a namespace by name fails. In particular, "ptype hide::Bar" should find the type. But it fails both inside python with gdb.lookup_type and with normal commands.
>
> Any idea what is happening? Reproducer follows.
>
> $ g++ -v
> ...
> gcc version 9.2.1 20191102 (Debian 9.2.1-17)
> $ cat b.cpp
> namespace hide { struct Bar {  int y; }; }
> struct Foo {  hide::Bar* b; };
>
> inline void x() { hide::Bar b; }
>
> static Foo aFoo;
>
> void * f() {
>   return aFoo.b;
> }
>
> int main() {
>   f();
>   return 0;
> }
> $ g++ -fdebug-types-section b.cpp -g
> $ ./build/gdb/gdb -ex "b f" -ex run -ex "ptype hide::Bar" ./a.out
> ...
> GNU gdb (GDB) 8.2.50.20180720-git
> ...
> Reading symbols from ./a.out...done.
> Breakpoint 1 at 0x1129: file b.cpp, line 11.
> Starting program: /usr/local/google/home/saugustine/gdb/a.out
>
> Breakpoint 1, Python Exception <type 'exceptions.NameError'> Installation error: gdb.execute_unwinders function is missing:
> f () at b.cpp:11
> 11  return aFoo.b;
> No symbol "Bar" in namespace "hide".
> (gdb)

I have found the source of this problem. The patch I identified above
relies on start_symtab to set list_in_scope. For type units,
setup_type_unit_groups ordinarily calls start_symtab. However, when
setup_type_unit_groups is called the second time (ie, first_time ends
up false), setup_type_unit_groups doesn't call start_symtab, but does
a small amount of housekeeping to recover the previously started
symtab.

This patch sets list_in_scope in addition to the other housekeeping it
does, and this fixes the problem. OK for trunk?

diff --git a/gdb/ChangeLog b/gdb/ChangeLog
index 186660bf9f..23dee0f503 100644
--- a/gdb/ChangeLog
+++ b/gdb/ChangeLog
@@ -1,3 +1,7 @@
+2020-03-24  Sterling Augustine <saugustine@google.com>
+
+       * dwarf2/read.c (setup_type_unit_groups): Set list_in_scope.
+
 2020-03-20  Tom Tromey  <tromey@adacore.com>

        * dwarf2/loc.h (dwarf2_evaluate_property): Make "addr_stack"
diff --git a/gdb/dwarf2/read.c b/gdb/dwarf2/read.c
index 0e879e08a0..65afb43b04 100644
--- a/gdb/dwarf2/read.c
+++ b/gdb/dwarf2/read.c
@@ -10954,6 +10954,7 @@ dwarf2_cu::setup_type_unit_groups (struct die_info *die)
                        compunit_language (cust),
                        0, cust));

+      list_in_scope = get_builder ()->get_file_symbols ();
       auto &file_names = line_header->file_names ();
       for (i = 0; i < file_names.size (); ++i)
        {

Re: Type-lookup by name for type in namespace fails with -fdebug-types-section after 93b8bea

[Tom Tromey]

>>>>> "Sterling" == Sterling Augustine <saugustine@google.com> writes:

Sterling> +2020-03-24  Sterling Augustine <saugustine@google.com>
Sterling> +
Sterling> +       * dwarf2/read.c (setup_type_unit_groups): Set list_in_scope.
Sterling> +

Thanks.  This is ok.

Tom