Re: [PATCH 1/2] gdb: fix crash when pretty printing anonymous struct with base class

On Mon, 17 Aug 2026 10:59:01 -0400
Simon Marchi <[email protected]> wrote:

> As I was reviewing another patch, I was wondering if we allowed
> type->name() to be nullptr.  The answer is yes, but there are some spots
> that don't check for nullptr, when they should.  Claude came up with a
> reproducer that inspired the included test case.  It is arguably a
> corner case, but it makes GDB crash.
> 
> The crash happens when trying to pretty-print (with "set print pretty
> on") a value of an anonymous struct type that has a base.  Given this:
> 
>     struct base
>     {
>       int a;
>     };
> 
>     struct : base
>     {
>       int b;
>     } v4 = { { 1 }, 2 };
> 
> We get:
> 
>     $ ./gdb -nx -q --data-directory=data-directory testsuite/outputs/gdb.cp/anon-struct/anon-struct -ex "with print pretty -- p v4"
>     Reading symbols from testsuite/outputs/gdb.cp/anon-struct/anon-struct...
>     $1 = {
>       <base> = {
>         a = 1
>       },
>     /home/smarchi/src/binutils-gdb/gdb/ui-file.h:77:30: runtime error: null pointer passed as argument 1, which is declared to never be null
> 
> The crash happens here:
> 
>     #4  0x00007ffff600dec8 in __ubsan_handle_nonnull_arg_abort () from /usr/lib/libubsan.so.1
>     #5  0x00005555637daa35 in ui_file::puts (this=0x7c1ff1c022d0, str=0x0) at /home/smarchi/src/binutils-gdb/gdb/ui-file.h:77
>     #6  0x0000555566f2b0ab in gdb_puts (linebuffer=0x0, stream=0x7c1ff1c022d0) at /home/smarchi/src/binutils-gdb/gdb/utils.c:1912
>     #7  0x00005555644e5cd4 in cp_print_value_fields (val=0x7d0ff1c2aa00, stream=0x7c1ff1c022d0, recurse=0, options=0x7bfff094acb0, dont_print_vb=0x0, dont_print_statmem=0) at /home/smarchi/src/binutils-gdb/gdb/cp-valprint.c:213
>     #8  0x00005555640de708 in c_value_print_struct (val=0x7d0ff1c2aa00, stream=0x7c1ff1c022d0, recurse=0, options=0x7bfff094acb0) at /home/smarchi/src/binutils-gdb/gdb/c-valprint.c:385
>     #9  0x00005555640df215 in c_value_print_inner (val=0x7d0ff1c2aa00, stream=0x7c1ff1c022d0, recurse=0, options=0x7bfff094acb0) at /home/smarchi/src/binutils-gdb/gdb/c-valprint.c:441
>     #10 0x000055556563828c in language_defn::value_print_inner (this=0x5555734a0700 <cplus_language_defn>, val=0x7d0ff1c2aa00, stream=0x7c1ff1c022d0, recurse=0, options=0x7bfff094acb0) at /home/smarchi/src/binutils-gdb/gdb/language.c:658
>     #11 0x0000555566fb7749 in common_val_print (value=0x7d0ff1c2aa00, stream=0x7c1ff1c022d0, recurse=0, options=0x7bfff094aa70, language=0x5555734a0700 <cplus_language_defn>) at /home/smarchi/src/binutils-gdb/gdb/valprint.c:1120
> 
> when we try to print the nullptr type name:
> 
>     if (options->prettyformat)
>       {
> 	gdb_printf (stream, "\n");
> 	print_spaces (2 + 2 * recurse, stream);
> 	gdb_puts ("members of ", stream);
> 	gdb_puts (type->name (), stream);
> 	gdb_puts (":", stream);
>       }
> 
> Fix it by using type->safe_name() instead of type->name().  This results
> in:
> 
>     $1 = {
>       <base> = {
>         a = 1
>       },
>       members of <unnamed type>:
>       b = 2
>     }
> 
> Claude pointed out that p-valprint.c has more or less the same code, so
> I changed it there too, but I did not write a test for that one.
> 

Both parts LGTM.

Approved-By: Kevin Buettner <[email protected]>