Re: [Patch/i386]: -mrtd vs system library declarations

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

Michael Matz <matz@suse.de> writes:

| Hi,
| 
| On 22 Jul 2003, Gabriel Dos Reis wrote:
| 
| > | >     extern string-literal { declaration-seq }
| > | >     extern string-literal declaration
| > |
| > | In principle I agree with you; however, adding support for this
| > | notation to the C front end would be tremendously more work than
| > | adding some machine-specific #pragmas.
| >
| > I hear you, but I'm skeptical about the overall cost.
| 
| Well, usually you are vehemently against introducing language extensions
| (for good reasons).

First of all, I'm not in principle against language extensions.
I happen to be against extensions that do not take into account
interactions with other language features.

|   extern "stdcall" bla();
| would be one for C.  So you are inconsistent at least in your requests,
| especially because the #pragma solution _does_ provide something inside
| the language.

The issue to consider is that the C programming language is the not
the only programming language affected in this issue.  We also have C++.
Standard C++ already provides mechanism to specify a calling convention, with 
semantics checks within the language (they interact with the rest of
the type-system).  That is an important aspect to consider.

Now the question is what for the C programming language?
*Either* way, it is an extension to the C language.  
Choosing the pragma approach does more damage than accepting the
linkage-specification syntax, specially given its interaction with C++.
Therefore I strongly advise against taking the pragma road.

The linkage-specification approach has the advantage that it is
well-understood and widely known. And it happens to accomodate
other languages involved in this issue.

-- Gaby