Re: ANSIfy cp/parser.c

[Mark Mitchell <mark at codesourcery dot com>]

--On Friday, January 10, 2003 01:33:08 PM +0100 Gabriel Dos Reis 
<gdr@integrable-solutions.net> wrote:

> Well, I do believe that the "code duplication" issue is a realistic
> and killer argument.  We have datapoints showing how that beats us in
> the past (and present).  However, wouldn't it be possible to set up
> machinery to share codes between those front-ends?
The problem with the c-common strategy is that we have to keep trying
to refactor code that doesn't need to be refactored -- just reused.

As a small example, consider the C++ code that handles variables going
out of scope.  It checks for a destructor, and if there is one, runs
a cleanup.  Well, that will work fine in C -- there are no destructors.
If profiling shows those checks to be expensive, conditionalize the
function.  But, there's no reason to use function-pointers and callbacks
and so forth to move most of the code into c-common.

Declarators are perhaps a better example.  If we really do create a
sensible declarator data-structure there's no point putting it in
c-common, and then having lots of shared code for processing them, but
then extra bits for C++ constructs like qualified names and
pointers-to-members.  The right thing to do is simply to use the same
code, but not check for qualified names or pointers-to-members in C.

--
Mark Mitchell                mark@codesourcery.com
CodeSourcery, LLC            http://www.codesourcery.com