httpd-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <wr...@lnd.com>
Subject RE: "free" borland C++ compiler
Date Sun, 25 Jun 2000 17:56:07 GMT
> From: TOKILEY@aol.com [mailto:TOKILEY@aol.com]
> Sent: Sunday, June 25, 2000 12:33 PM
> 
> 
> In a message dated 00-06-25 12:35:22 EDT, Ozgur Urgenc writes...
> 
> > C++ Builder Help :
> > [snip]
> 
> Hi Ozgur...
> 
> What VERSION of C++ Builder is that help text from? 
> 5.5? 5.4? 5.x?
> 
> I don't see some of the options on earlier versions... at
> least not in the exact syntax shown... but not 
> sure this is an issue since I believe any proposed 
> Apache/Borland support is going to be 
> 'minimum requirement: BC 5.5'.

Perhaps not... but that's the only goal of the ASF at the
moment, to provide -everyone- Apache compilable for free
on Win32, without the particularities of cygwin (I'm not
dismissing cygwin, just suggesting it is its own beast :)

If Borland users benefit, great!  If MS releases the compiler
with the operating system, great again (and they would have
a lot less arguments from the Justice Department, given that
they -can- cite precidence.)

So I'm not trying to -exclude- pre-5.5, simply 5.5 is free.
If you two and others can untangle it so bcc 4.0 can build
it, that would be fine.  But, for example, we are only 
'supporting' MSVC 5.0 and 6.0 right now, though others could
-probably- build it with a little hacking.

Let's not dig too far further back than bcc 4.0 for now.

> It's certainly possible to create some 'callable' binaries that
> are totally generic and can be 'called' or 'imported' seamlessly
> from an MSVC or BORLAND or ( whoever ) compiled version
> of the .LIB's and .DLL's but I don't recall ever having reached
> that goal without using different .DEF files... have you?

I would expect this is the only acceptable solution.  Module authors
distributing binaries expect to link no matter what.

If this means we have to maul the .def file, then perhaps we need
to start looking at that too.  You'll notice the .pl script matches
for .mak right now, but could have another section to pick up .def
files (though we better be sure they are the Win32 defs :)
 
> Borland products have always 'loved' the INT FAR PASCAL
> thing and they default to it a lot but I'm not sure it's an option.
> The whole 'left-right right-left who cleans the stack' thing is 
> what becomes the real issue for people trying to interface
> with the resulting binaries, especially if they are writing in
> assembler. Actually, C++ makes things even worse with
> regards to being tickety-boo about calling conventions in
> linked/called modules, but that's another story.

It's an argument already on the MS side, and I've always come down
against c convention -unless- you have a variable argument list
(which has nothing to do with passing an array of varargs.)
 
> I believe we are talking Apache 2.0 and BC 5.5 only here
> but I haven't heard 'officially' from ASF about that yet.

If we can support it, great, but we've run out of time to really
thresh it out for 1.3.13, with the bug fixes we really need to
get rolling on it.  It -does- -not- mean that we can't provide
a .pl to convert to a Borland build after 1.3.13 is out. 

(b.t.w. - very cool, I hadn't tried it against the 2.0a4 build
 myself yet, just beating on 1.3.13.  Thanks for trying that :)
 
> This one is for ASF community...
> 
> Is there any such thing yet as a set of 'standard' 2.0 APR
> interfacing sample programs for both Standard 'C' and
> C++ that can be used as 'test' programs to make sure
> that whatever solution is used DOES, in fact, create
> output binaries that are 'seamlessly usable' by people
> trying to code for APR using whatever compiler/linker
> they choose? I thought I saw a 'C++' program float by
> that was complaining about APR 'extern 'C' stuff but I'm 
> not sure.

Everything 'exposed' in 2.0 should be very clearly tagged in
extern "C" blocks.  That means nothing should be attacked by
name mangling.  If bcc has a problem, that's a problem, but
not for the ASF.  Those specs have been pretty clear for some
time now.

Mime
View raw message