Quoting and the expr parser

Quoting and the expr parser

[Keith Seitz]

Hi,

I have a simple question:

Why must one type "my_class::'operator+'" instead of simply
"my_class::operator+"?

*Should* gdb accept the latter; after all, it's not ambiguous (assuming 
it's not overloaded).

Keith

Re: Quoting and the expr parser

[Mark Kettenis]

> Date: Tue, 05 Aug 2008 12:24:04 -0700
> From: Keith Seitz <keiths@redhat.com>
> 
> Hi,
> 
> I have a simple question:
> 
> Why must one type "my_class::'operator+'" instead of simply
> "my_class::operator+"?
> 
> *Should* gdb accept the latter; after all, it's not ambiguous (assuming 
> it's not overloaded).

GDB probably should accept that if the selected language is C++.  The
problem though is that it is actually pretty difficult to make GDB
accept that wiouout the additional quotes.  Normally a '+' sign isn't
allowed in a function name, so you'd have to teach the lexer/parser
that it is allowed but only if there the "operator" keyword is used
right in front of it.  This is a fundamental problem with C++.  It is
so badly designed that it is almost impossible to write a proper
parser for it.

Re: Quoting and the expr parser

[Daniel Jacobowitz]

On Tue, Aug 05, 2008 at 12:24:04PM -0700, Keith Seitz wrote:
> Hi,
>
> I have a simple question:
>
> Why must one type "my_class::'operator+'" instead of simply
> "my_class::operator+"?
>
> *Should* gdb accept the latter; after all, it's not ambiguous (assuming  
> it's not overloaded).

Yes, it should.  This is an omission in c-exp.y; cp-name-parser.y
(not used for expressions) handles it correctly.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Quoting and the expr parser

[André Pönitz]

On Tuesday 05 August 2008 21:45:00 Mark Kettenis wrote:
> > Date: Tue, 05 Aug 2008 12:24:04 -0700
> > From: Keith Seitz <keiths@redhat.com>
> > 
> > Hi,
> > 
> > I have a simple question:
> > 
> > Why must one type "my_class::'operator+'" instead of simply
> > "my_class::operator+"?
> > 
> > *Should* gdb accept the latter; after all, it's not ambiguous (assuming 
> > it's not overloaded).
> 
> GDB probably should accept that if the selected language is C++.  The
> problem though is that it is actually pretty difficult to make GDB
> accept that wiouout the additional quotes.  Normally a '+' sign isn't
> allowed in a function name, so you'd have to teach the lexer/parser
> that it is allowed but only if there the "operator" keyword is used
> right in front of it.

[This sounds like a one-line rule for a parser generator. But that's
not the point, see below]

> This is a fundamental problem with C++.  It is 
> so badly designed that it is almost impossible to write a proper
> parser for it.

While this seems like a good opportunity for a bit of C++ bashing I'd
like to point out that the problem is _not_ a "C++ parser". The problem
is "gdb cannot read back what gdb writes out", and that's a very 
generic one, completely independent of the language.

There are several places where gdb produces type information that
cannot be easily handed back to gdb because it needs to be quoted
in order to survive the gdb CLI parsing.

Now quoting would be not a problem if there were rules for it.
Unfortunately, for gdb the quoting rules are far from obvious, let
alone from being documented. It would be _so easy_ if there
were a 'replace x by \x' or 'put quotes around it' or even 'encode
it in base64' or 'only use mangled names'.


André 

PS: As usual: These are my opinions, not the company's.