NetNews Usenet Archive 1992 #16/NN_1992_16.iso/spool/comp/lang/cplus/11759

Internet Message Format | 1992-07-29 | 1.8 KB

Xref: sparky comp.lang.c++:11759 gnu.g++.help:1059
Path: sparky!uunet!idacrd!purtill
From: [e-mail] (Mark Purtill)
Newsgroups: comp.lang.c++,gnu.g++.help
Subject: Re: porting to g++ (from other "dialects")
Message-ID: <[e-mail]>
Date: 29 Jul 92 22:21:06 GMT
References: <[e-mail]> <[e-mail]>
Organization: idacrd, princeton, nj
Lines: 27

[e-mail] (Joe Buck) writes:
>2.  The template implementation is still buggy, but it's been improving
>rapidly (2.2.2 is much better than 2.1).  g++ doesn't implement the cfront
>hack that lets you declare template functions in Foo.h and define them in
>Foo.c, and then do a hidden #include of Foo.c so everything will work.  If
>you do this in g++ you'll need to #include "Foo.c".

    That isn't actually quite how cfront works.  What it does is
generate a small file along the lines of:

    #include "Foo.h"
    #include "Foo.c"
    static Foo<int> _dummy_ ;

and then compiles THAT, so as to generate one copy of all of the
relevant functions.  This means, for instance, that "Foo.h" had better
not depend on anything else having been included before it, and that
you have to pass all the cpp/cfront/cc flags to the load phase (since
that's where the small file is generated and compiled).  In practice,
this all seems to be not worth the effort (especially when class Foo
is declared in "Bar.h" and defined int "Bar.C" and "Bar2.C", when you
have to start messing around with configuration files to tell the
template linker thingee where everything is).  It's especially fun
when you want the same code to compile under gcc as well....

^.-.^ Mark Purtill, [e-mail]; uunet!idacrd!purtill. [phone].
((")) \@_: IDA/CCR-P, Thanet Road, Princeton NJ 08540;           [phone].