Expected output of gdb.cp/no-dmgl-verbose.exp

Expected output of gdb.cp/no-dmgl-verbose.exp

[Luis Machado]

I'm trying to determine why this particular test is failing (for both 
aarch64-linux and x86_64-linux on Ubuntu 18.04/20.04) and what the 
expected outcome is.

In my case, the only symbol I see for function "f" is the following:

"f(std::__cxx11::basic_string<char, std::char_traits<char>, 
std::allocator<char> >)"

There is no "f(std::string)" nor "f(std::basic_string<char, 
std::char_traits<char>, std::allocator<char> >)" symbol.

Given we've done quite a few changes to how demangling is applied and 
how we match the names, I wonder if this test is no longer meaningful.

Does anyone know?

Re: Expected output of gdb.cp/no-dmgl-verbose.exp

[Keith Seitz]

On 4/12/21 11:16 AM, Luis Machado via Gdb wrote:
> I'm trying to determine why this particular test is failing (for both
> aarch64-linux and x86_64-linux on Ubuntu 18.04/20.04) and what the
> expected outcome is.
> 
> In my case, the only symbol I see for function "f" is the following:
> 
> "f(std::__cxx11::basic_string<char, std::char_traits<char>,
> std::allocator<char> >)"
> 
> There is no "f(std::string)" nor "f(std::basic_string<char,
> std::char_traits<char>, std::allocator<char> >)" symbol.
> 

Wow, such a long time ago! IIRC...

This test was written to ensure that "std::string" was used
instead of "std::basic_string<...>" (which is now called something
else again). Something about "that's what nm and other tools tell users,
so we should not mention the symbol's real (linkage) name, either."

The "no-dmgl-verbose" refers to the DMGL_VERBOSE option
of the demangler, which does this.

There is even special logic in our typedef-replacing machinery to
enforce this. [NOTE: `ignore_typedefs' in cp-support.c is also "out-
of-date" wrt to this naming change. So none of that has likely been
"working" for many years, but then at very quick glance, "nm -C" doesn't
do this anymore, either.]

I was against it then, so I certainly do not mind getting rid of this test
or limiting it to C++ installations that typedef the "older" std::basic_string<...>
to std::string.

Keith

Re: Expected output of gdb.cp/no-dmgl-verbose.exp

[Luis Machado]

On 4/12/21 3:45 PM, Keith Seitz wrote:
> On 4/12/21 11:16 AM, Luis Machado via Gdb wrote:
>> I'm trying to determine why this particular test is failing (for both
>> aarch64-linux and x86_64-linux on Ubuntu 18.04/20.04) and what the
>> expected outcome is.
>>
>> In my case, the only symbol I see for function "f" is the following:
>>
>> "f(std::__cxx11::basic_string<char, std::char_traits<char>,
>> std::allocator<char> >)"
>>
>> There is no "f(std::string)" nor "f(std::basic_string<char,
>> std::char_traits<char>, std::allocator<char> >)" symbol.
>>
> 
> Wow, such a long time ago! IIRC...

Indeed. I vaguely remember that, back in ~2011 IIRC. But obviously not 
enough to figure out what was going on.

> 
> This test was written to ensure that "std::string" was used
> instead of "std::basic_string<...>" (which is now called something
> else again). Something about "that's what nm and other tools tell users,
> so we should not mention the symbol's real (linkage) name, either."
> 
> The "no-dmgl-verbose" refers to the DMGL_VERBOSE option
> of the demangler, which does this.
> 
> There is even special logic in our typedef-replacing machinery to
> enforce this. [NOTE: `ignore_typedefs' in cp-support.c is also "out-
> of-date" wrt to this naming change. So none of that has likely been
> "working" for many years, but then at very quick glance, "nm -C" doesn't
> do this anymore, either.]

Ah, that makes it much more clear. Thanks for the explanation.

> 
> I was against it then, so I certainly do not mind getting rid of this test
> or limiting it to C++ installations that typedef the "older" std::basic_string<...>
> to std::string.

I think keeping it may be more confusing than useful. I'll propose its 
removal on gdb-patches@ and we'll see where we go from there.

Thanks Keith.

> 
> Keith
> 

Re: Expected output of gdb.cp/no-dmgl-verbose.exp

[David Blaikie]

FWIW - at least some Clang folks are working on adding a
"preferred_name" attribute ( https://reviews.llvm.org/D91311 ) which
would be used to help tools understand that std::basic_string<char,
... is best known as std::string. Perhaps we'll end up with a
DW_AT_preferred_name attribute in DWARF to make this more explicit/not
needing consumers to guess or otherwise encode heuristically which
names should be used.

On Mon, Apr 12, 2021 at 1:02 PM Luis Machado via Gdb <gdb@sourceware.org> wrote:
>
> On 4/12/21 3:45 PM, Keith Seitz wrote:
> > On 4/12/21 11:16 AM, Luis Machado via Gdb wrote:
> >> I'm trying to determine why this particular test is failing (for both
> >> aarch64-linux and x86_64-linux on Ubuntu 18.04/20.04) and what the
> >> expected outcome is.
> >>
> >> In my case, the only symbol I see for function "f" is the following:
> >>
> >> "f(std::__cxx11::basic_string<char, std::char_traits<char>,
> >> std::allocator<char> >)"
> >>
> >> There is no "f(std::string)" nor "f(std::basic_string<char,
> >> std::char_traits<char>, std::allocator<char> >)" symbol.
> >>
> >
> > Wow, such a long time ago! IIRC...
>
> Indeed. I vaguely remember that, back in ~2011 IIRC. But obviously not
> enough to figure out what was going on.
>
> >
> > This test was written to ensure that "std::string" was used
> > instead of "std::basic_string<...>" (which is now called something
> > else again). Something about "that's what nm and other tools tell users,
> > so we should not mention the symbol's real (linkage) name, either."
> >
> > The "no-dmgl-verbose" refers to the DMGL_VERBOSE option
> > of the demangler, which does this.
> >
> > There is even special logic in our typedef-replacing machinery to
> > enforce this. [NOTE: `ignore_typedefs' in cp-support.c is also "out-
> > of-date" wrt to this naming change. So none of that has likely been
> > "working" for many years, but then at very quick glance, "nm -C" doesn't
> > do this anymore, either.]
>
> Ah, that makes it much more clear. Thanks for the explanation.
>
> >
> > I was against it then, so I certainly do not mind getting rid of this test
> > or limiting it to C++ installations that typedef the "older" std::basic_string<...>
> > to std::string.
>
> I think keeping it may be more confusing than useful. I'll propose its
> removal on gdb-patches@ and we'll see where we go from there.
>
> Thanks Keith.
>
> >
> > Keith
> >

Re: Expected output of gdb.cp/no-dmgl-verbose.exp

[Tom de Vries]

On 4/12/21 8:55 PM, Luis Machado via Gdb wrote:
> On 4/12/21 3:45 PM, Keith Seitz wrote:
>> On 4/12/21 11:16 AM, Luis Machado via Gdb wrote:
>>> I'm trying to determine why this particular test is failing (for both
>>> aarch64-linux and x86_64-linux on Ubuntu 18.04/20.04) and what the
>>> expected outcome is.
>>>
>>> In my case, the only symbol I see for function "f" is the following:
>>>
>>> "f(std::__cxx11::basic_string<char, std::char_traits<char>,
>>> std::allocator<char> >)"
>>>
>>> There is no "f(std::string)" nor "f(std::basic_string<char,
>>> std::char_traits<char>, std::allocator<char> >)" symbol.
>>>
>>
>> Wow, such a long time ago! IIRC...
> 
> Indeed. I vaguely remember that, back in ~2011 IIRC. But obviously not
> enough to figure out what was going on.
> 
>>
>> This test was written to ensure that "std::string" was used
>> instead of "std::basic_string<...>" (which is now called something
>> else again). Something about "that's what nm and other tools tell users,
>> so we should not mention the symbol's real (linkage) name, either."
>>
>> The "no-dmgl-verbose" refers to the DMGL_VERBOSE option
>> of the demangler, which does this.
>>
>> There is even special logic in our typedef-replacing machinery to
>> enforce this. [NOTE: `ignore_typedefs' in cp-support.c is also "out-
>> of-date" wrt to this naming change. So none of that has likely been
>> "working" for many years, but then at very quick glance, "nm -C" doesn't
>> do this anymore, either.]
> 
> Ah, that makes it much more clear. Thanks for the explanation.
> 
>>
>> I was against it then, so I certainly do not mind getting rid of this
>> test
>> or limiting it to C++ installations that typedef the "older"
>> std::basic_string<...>
>> to std::string.
> 
> I think keeping it may be more confusing than useful. I'll propose its
> removal on gdb-patches@ and we'll see where we go from there.
> 

The FAIL was mentioned here:
https://sourceware.org/bugzilla/show_bug.cgi?id=19436#c7

Thanks,
- Tom