C++ ABI: Mangling of call_expr

[Matt Austern]

On Tuesday, June 24, 2003, at 08:16  AM, Mark Mitchell wrote:

> On Tue, 2003-06-24 at 03:32, Nathan Sidwell wrote:
>> Giovanni Bajo wrote:
>>> Hello,
>>>
>>> I can't understand how mangle.c is supposed to mangle a call 
>>> expression.
>>> Currently, it just segfaults on it. The expression production from 
>>> the ABI
>>> says:
>>>
>>> <expression> ::= <unary operator-name> <expression>
>>>             ::= <binary operator-name> <expression> <expression>
>>>             ::= <trinary operator-name> <expression> <expression>
>>> <expression>
>>> [...]
>>>
>>> But what is a CALL_EXPR? Its two-letter mangled name is "cl", but it
>>> certainly can't be either unary, binary, or trinary. In fact, in
>>> operators.def, it is defined with arity "-1", which means "any arity 
>>> is
>>> allowed". But then, I can't see which production is supposed to use.
>> looks like a bug in the ABI spec. To be demanglable, there must be a 
>> count
>> of the number of parameters encoded.
>>
>> Mark, have I missed something?
>
> Only that this is a known issue, with no known solution.
>
> Steve Adamczyk and I agreed at one point that the ISO C++ standard
> allows far too much complexity in expressions in this context:
>
>   S<sizeof (new (p) X(3.2, a->b (c)))>
>
> is pretty goofy.
>
> However, the standard does presently seem to allow that.
>
> I was pretty sure I had raised this issue on the ABI mailing list, but 
> I
> don't see that message, so maybe I never did.

The real question is whether you or Steve has raised this issue on
the C++ committee's mailing list.  I agree with both of you that
arbitrarily complicated expressions like this are ghastly.  (Not
that I have a good suggestion of exactly where to draw the line
between expressions that should be allowed and ones that
shouldn't.)

			--Matt