Hi.

I've fixed up the D demangler in gdb with one or two hicks left to address.  As the below auto-complete feature presents, the full function signatures are now shown.

(gdb) break 'gdbstress.testsig(
gdbstress.testsig(double)
gdbstress.testsig(float)
gdbstress.testsig(real)


Currently, I have been dwelling on the best way to express the following symbols:

__ctorMFZ
__dtorMFZ
__postblitMFZ
__initZ
__ClassZ
__vtblZ
__InterfaceZ
__ModuleInfoZ


This is the current mapping I've got in my local branch:

__ctor -> this
__dtor -> ~this
__postblit -> this(this)
__init -> init
__Class -> classinfo
__vtbl -> vtable
__interface -> interfaceinfo
__ModuleInfo -> moduleinfo


However I'm not sure if the names are sufficient, and whether or not they should really be prefixed with a double underscore to ensure that they don't conflict with user-declared members (especially the compiler-generated symbols).


What do you think?

(gdb) print 'gdbstress.testmembers.
gdbstress.testmembers.this(this, int)
gdbstress.testmembers.~this()
gdbstress.testmembers.this(this)
gdbstress.testmembers.init


Regards
Iain.

On 2013-12-30 13:24, Iain Buclaw wrote:

> However I'm not sure if the names are sufficient, and whether or not
> they should really be prefixed with a double underscore to ensure that
> they don't conflict with user-declared members (especially the
> compiler-generated symbols).

I think they should have some kind of prefix to avoid conflict with user declared symbols and somehow indicate they're special symbols.

-- 
/Jacob Carlborg

On Monday, 30 December 2013 at 12:57:57 UTC, Jacob Carlborg wrote:
> On 2013-12-30 13:24, Iain Buclaw wrote:
>
>> However I'm not sure if the names are sufficient, and whether or not
>> they should really be prefixed with a double underscore to ensure that
>> they don't conflict with user-declared members (especially the
>> compiler-generated symbols).
>
> I think they should have some kind of prefix to avoid conflict with user declared symbols and somehow indicate they're special symbols.

I had already gathered that much when I spoke to David on IRC. I was looking for some more concrete suggestions really.

Also Brad, I haven't received any ML messages since 29th? :-)

Regards,
Iain.

On Tuesday, 31 December 2013 at 12:22:54 UTC, Iain Buclaw wrote:
>
> Also Brad, I haven't received any ML messages since 29th? :-)
>

Oh never mind,, I think I know the problem. :(

On 30 December 2013 12:57, Jacob Carlborg <doob@me.com> wrote:
> On 2013-12-30 13:24, Iain Buclaw wrote:
>
>> However I'm not sure if the names are sufficient, and whether or not they should really be prefixed with a double underscore to ensure that they don't conflict with user-declared members (especially the compiler-generated symbols).
>
>
> I think they should have some kind of prefix to avoid conflict with user declared symbols and somehow indicate they're special symbols.
>


Special symbols can be suffixed with a $ perhaps.  I can't think of any other symbol that would work.

__init -> init$
__Class -> classinfo$
__vtbl -> vtable$
__interface -> interfaceinfo$
__ModuleInfo -> moduleinfo$


The alternative is to just go for prefixing with double underscore, given that the names are OK?

__init -> __init
__Class -> __classinfo
__vtbl -> __vtable
__interface -> __interfaceinfo
__ModuleInfo -> __moduleinfo


Regards
Iain

Iain Buclaw <ibuclaw@gdcproject.org> writes:

> On 30 December 2013 12:57, Jacob Carlborg <doob@me.com> wrote:
>> On 2013-12-30 13:24, Iain Buclaw wrote:
>>
>>> However I'm not sure if the names are sufficient, and whether or not they should really be prefixed with a double underscore to ensure that they don't conflict with user-declared members (especially the compiler-generated symbols).
>>
>>
> Special symbols can be suffixed with a $ perhaps.  I can't think of any other symbol that would work.
>
> __init -> init$
> __Class -> classinfo$
> __vtbl -> vtable$
> __interface -> interfaceinfo$
> __ModuleInfo -> moduleinfo$

I like this approach.  Do we need the "info" in classinfo$, interfaceinfo$, and moduleinfo$?

Jerry

On 02.01.2014 12:35, Iain Buclaw wrote:
>
> Special symbols can be suffixed with a $ perhaps.  I can't think of
> any other symbol that would work.
>
> __init -> init$
> __Class -> classinfo$
> __vtbl -> vtable$
> __interface -> interfaceinfo$
> __ModuleInfo -> moduleinfo$
>
>
> The alternative is to just go for prefixing with double underscore,
> given that the names are OK?
>
> __init -> __init
> __Class -> __classinfo
> __vtbl -> __vtable
> __interface -> __interfaceinfo
> __ModuleInfo -> __moduleinfo
>

I think double underscores should be hint enough to suggest "compiler generated stuff". Prepending $ might confuse expression evaluators (e.g. mago).

I always wondered why these symbols don't follow the standard mangling to start with...