Re: [Gcl-devel] -Wall checked in

[Camm Maguire]

Greetings!

Gregory Wright <address@hidden> writes:

> Hi Camm,
> 
> On Sat, 2002-07-20 at 03:25, Camm Maguire wrote:
> > Greetings!  Am pleased to announce that I just checked in a rather
> > large mod to get a clean build under -Wall.  Several issues uncovered
> > and addressed at least preliminarily.  maxima 5.6 and cvs build and
> > run fine, and gcl compiles itself as well as ever.  Seems slightly
> > faster too, though this is not a scientific observation.  we'll see
> > shortly how the Debian autobuilders like this one.
> > 
> > Please test thoroughly.
> > 
> > Just a few notes for the future:
> > 1) It would be nice to find a way to go completely with ANSI C
> > varargs, in place of the traditional method.  The problem here is that
> > one cannot represent *all* args as variable in this way, and such a
> > possibility is built in to the code in funlink.c (c_apply_n).  
> 
> ANSI C would be good. It does bring up the interesting policy question:
> should gcl work with any compiler other than gcc? Does anyone use it
> with something else? I ask because it affects how much cruft can be
> removed from the configure.{in,ac} (see below).
> 

Personally, I have no interest in non-gcc compilers.  But some might,
and I don't think we get much mileage out of assuming that gcc will be
the compiler.  We don't currently use GNUC extensions, for example,
and I think that we should protect them with configure and defines
should we ever do so.  

It would be great if configure had an 'include' mechanism, whereby we
could split the huge file up into a tree which would be invoked
according to the results of earlier tests.  I do agree that
configure.in looks rather crufty now.

> > 
> > 2) Need to minimalize the new header h/protoize.h
> > 
> > 3) Need to declare all functions with proper arguments -- right now
> >    gcl requires certain functions to be declared with an empty
> >    argument list to pass the function pointers around correctly.
> > 
> > 4) Need to revisit the 'lintian' changes I've made to the list
> >    compiler.  Several unused variables should not be declared, rather
> >    than the empty use of said variables I've inserted into the C
> >    output.  Likewise with unused labels.
> > 
> > 5) Need to beautify and clean the code at some point.  Changes made
> >    thus far were minimal for -Wall, and not intended to change
> >    functionality (significantly).
> 
> Yes, there are some stretches of code that are hard to read. Should a
> standard 'indent' command be used on the C and the lisp run through an
> agreed upon pretty printer?
> 

I think yes.

> > 
> > 6) Need to finalize the build method, modern makefiles, have configure
> >    spit out most if not all what we put in by hand not in *.h 
> > 
> 
> I'll suggest automake + autoconf for developer builds; this won't affect
> users who will still be able to build with ./configure && make && make
> install.
> 

Agreed!  Do we need a bootstrap shell script?

> The possibly disruptive change is cleaning up the directory structure.
> We should probably move the build support (e.g., add-defs) into a
> scripts directory. This aligns us with the standard gnu-ish tree: top
> level is essential documentation (README, INSTALL), build scripts
> (configure, Makefile.*) and subdirectories for all of the code. This
> structure helps keep automake's makefiles clean and easy to manage.
> 

If you could give me a sample 'ls -lR' or 'find -type f' on a tree
that looks standard to you, I'd be happy to handle the cvs end.

> I'd be happy to work on this since I've already started.
> 

Great!


Take care,

> 
> > 7) Need to turn to ansi compliance, import clocc test tree.
> > 
> > 8) Need to have a sfasldlsym.c option where bfd doesn't work, or get
> >    bfd to work everywhere :-).
> > 
> > 9) release 2.5.0, not necessarily in this order :-)
> > 
> > Take care,
> > -- 
> > Camm Maguire                                                address@hidden
> > ==========================================================================
> > "The earth is but one country, and mankind its citizens."  --  Baha'u'llah
> > 
> > _______________________________________________
> > Gcl-devel mailing list
> > address@hidden
> > http://mail.gnu.org/mailman/listinfo/gcl-devel
> -- 
> 
> Gregory Wright
> Chief Technical Officer
> PacketStorm Communications, Inc.
> 20 Meridian Road
> Eatontown, New Jersey 07724
> 
> 1 732 544-2434 ext. 206
> 1 732 544-2437 [fax]
> address@hidden
> 
> 
> 
> 

-- 
Camm Maguire                                            address@hidden
==========================================================================
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah