Re: gdb linux testing regressions (compared to 7/24)

[Daniel Berlin <dberlin@redhat.com>]

Jimmy Guo <guo@cup.hp.com> writes:

> FYI:
> 
> Latest gdb source has a test regression of 200+ as compared to the 7/24
> tree, on i686-pc-linux-gnu.  For example, gdb.base/break.exp:
> ...
> print marker2(99)
> 
> read_register_bytes:  Couldn't update register 16.
> 
> (gdb) FAIL: gdb.base/break.exp: hit breakpoint on called function
> ...
> 
> Regression info is provided below ...
> 

...

> gdb.c++/classes.exp
>  | F |: base class param->a
>  | F |: base class param->x
>  | F |: inherited class param->a
>  | F |: inherited class param->x
>  | F |: base class param.a
>  | F |: base class param.x
>  | F |: inherited class param.a
>  | F |: inherited class param.x
> 

Known.

But it's actually

FAIL: base class (&param)->a
FAIL: base class (&param)->x
FAIL: inherited class (&param)->a
FAIL: inherited class (&param)->x
FAIL: continue to enums2


now.

> gdb.c++/demangle.exp
>  | F |: gnu:
>__thunk_64__0RL__list__Q29CosNaming20_proxy_NamingContextUlRPt25_CORBA_Unbounded_Sequence1ZQ29CosNaming7BindingRPQ29CosNaming15BindingIterator


I claim this can't be unambiguously demangled, as it has more than one
legal demangling.

This will have to be marked XFAIL.

I'll add more tests that break the demangler before i fixed it, to
show this is worth it (I know it's worth it. Before, customers
complained about things not being able to be demangled. Now, they
don't).



> 
> gdb.c++/namespace.exp
>  | F |: print 'AAA::xyzq'('x')
>  | F |: print 'BBB::CCC::xyzq'('x')
> 

Unreproducable.


I get "FAIL: info func xyzq" instead, but it's clearly not failing.
info func xyzq^M
All functions matching regular expression "xyzq":^M
^M  
File ./gdb.c++/namespace.cc:^M
int AAA::A_xyzq(int);^M
char AAA::xyzq(char);^M
int BBB::B_xyzq(int);^M
char BBB::CCC::xyzq(char);^M
char BBB::Class::xyzq(char);^M
char BBB::xyzq(char);^M

All of these properly match xyzq.


> gdb.c++/templates.exp
>  | F |: print t5i.value()

I can only reproduce under stabs
It looks like a name gets demangled properly the firsttime, when we
print it the second, it's garbage.

This tells me it's a method stub problem, and i'm on it.

Under dwarf2 i get no failures at all.

--Dan