Re: Apologies (Was Re: AC_INIT translates PACKAGE to lower case)

[Akim Demaille]

[The technical part of your message is CC'd to Autoconf: I don't want
to have to duplicate this explanation in one year to someone asking
the same questions.  That's what the archive is here for.]

>>>>> "Ralf" == Ralf Corsepius <address@hidden> writes:

Ralf> 1. Processing the 1st argument of AC_INIT: "Lowercasing", +
Ralf> "stripping the gnu". IMO, this unnecessarily complicates things
Ralf> and doesn't actually help anybody. Replacing non
Ralf> printable-characters with underscores, for a PACKAGE_TARNAME
Ralf> however is reasonable.

Your POV.  OK, point taken.  It was already clear.

Ralf> 2. The default behavior of AC_INIT: IMO, most of the tasks it
Ralf> tries to perform are beyond the scope of autoconf's duties and
Ralf> are not actually useful as defaults, because they are not
Ralf> applicable or at least of limited used for autoconf.

Your POV again.

Ralf> Example: Why should AC_INIT provide an email address or
Ralf> PACKAGE_BUGREPORT? If a user really wants one, let him put a
Ralf> AC_PACKAGE_ADDRESS(address@hidden) or similar into his
Ralf> configure.ac.

No, I've always been in favor of using m4_define, because there is no
point in introducing many macros for a single job (defining).  But
I've told that that's ``ugly hacks''.

Ralf> I know, some of the PACKAGE_*'s are used to provide information
Ralf> to configure --help, but I fail to see why autoconf should
Ralf> AC_DEFINE or propagate them _uncondionally_ as
Ralf> CPPFLAGS/-DPACKAGE_*.

Many because that's just _your_ POV.  Did it occur to you that I might
have received demands opposed to your POV?  *Several* of them?  Here's
the oldest.

| From: "Nelson H. F. Beebe" <address@hidden>
| Subject: autoconf-2.52 and PACKAGE_xxx vars
| To: Akim Demaille <address@hidden>
| Cc: address@hidden
| X-Sent: 23 weeks, 10 hours, 33 minutes, 12 seconds ago
| 
| [This is temporarily off the autoconf list].
| 
| The addition of the PACKAGE_xxx variables generated from the extended
| AC_INIT() macro is a welcome addition to autoconf-2.5x.
| 
| I've often had similar variables defined in my configure.in file, so
| now there is a good chance of standardizing them.  However, it appears
| that their wider utility has not been foreseen, so that I had to do
| this:
| 
| AC_INIT(bibclean,2.13,address@hidden)
| PACKAGE_DATE="27-Aug-2001"
| 
| AH_TEMPLATE(PACKAGE_BUGREPORT,        [Package bug-reporting address])
| AH_TEMPLATE(PACKAGE_DATE,     [Package date])
| AH_TEMPLATE(PACKAGE_NAME,     [Package name])
| AH_TEMPLATE(PACKAGE_STRING,   [Package string])
| AH_TEMPLATE(PACKAGE_TARNAME,  [Package tarname])
| AH_TEMPLATE(PACKAGE_VERSION,  [Package version])
| 
| AC_DEFINE_UNQUOTED(PACKAGE_BUGREPORT, "$PACKAGE_BUGREPORT")
| AC_DEFINE_UNQUOTED(PACKAGE_DATE,      "$PACKAGE_DATE")
| AC_DEFINE_UNQUOTED(PACKAGE_NAME,      "$PACKAGE_NAME")
| AC_DEFINE_UNQUOTED(PACKAGE_STRING,    "$PACKAGE_STRING")
| AC_DEFINE_UNQUOTED(PACKAGE_TARNAME,   "$PACKAGE_TARNAME")
| AC_DEFINE_UNQUOTED(PACKAGE_VERSION,   "$PACKAGE_VERSION")
| 
| in order to get my config.h file to contain:
| 
| /* Package bug-reporting address */
| #define PACKAGE_BUGREPORT "address@hidden"
| 
| /* Package date */
| #define PACKAGE_DATE "27-Aug-2001"
| 
| /* Package name */
| #define PACKAGE_NAME "bibclean"
| 
| /* Package string */
| #define PACKAGE_STRING "bibclean 2.13"
| 
| /* Package tarname */
| #define PACKAGE_TARNAME "bibclean"
| 
| /* Package version */
| #define PACKAGE_VERSION "2.13"
| 
| Is there a cleaner way?  Also, these new variables should likely
| be documented in the manual.
| 
| 
-------------------------------------------------------------------------------
| - Nelson H. F. Beebe                    Tel: +1 801 581 5254                  
-
| - Center for Scientific Computing       FAX: +1 801 585 1640, +1 801 581 4148 
-
| - University of Utah                    Internet e-mail: address@hidden  -
| - Department of Mathematics, 322 INSCC      address@hidden  address@hidden -
| - 155 S 1400 E RM 233                       address@hidden                    
-
| - Salt Lake City, UT 84112-0090, USA    URL: http://www.math.utah.edu/~beebe  
-
| 
-------------------------------------------------------------------------------
| 



Ralf> 3. The monolytical AC_INIT-API. Using several AC_*-macros
Ralf> instead of one AC_INIT to perform everything would probably be
Ralf> better and help not to pull-in unnecessary symbols and hereby
Ralf> pollute various namespaces.

Sure, I would use m4_define.  Certainly not additional macros, there
is already enough to be maintained in Autoconf.

But again, some users think that's ``ugly hacks''.

Ralf> 4. automake should not apply PACKAGE_TARNAME as value for
Ralf> PACKAGE, but should apply something that is backward compatible
Ralf> to the former PACKAGE (Non-lowercased, user-provided) as default
Ralf> The 4th argument to AC_INIT provides a means to implement this,
Ralf> IMO, however is "just a hack" to work around autoconf's
Ralf> lowercasing.

See what I mean?  You still don't get it.  And you're still talking
about ``backward compatibility''.  Everybody wants a better synergy
between Automake and Autoconf, and here's what happens when one tries.


Ralf> Finally, let me emphasize that I do not want nor intend to argue
Ralf> nor fight for this. Our opinions apparently diverge and I don't
Ralf> seem to be able to convince you, that's all ... I can't exclude
Ralf> the possibility that we may still be talking past each other,
Ralf> either.

:)  Indeed, I won't change anything unless someone proves me there is
something which is *functionally* wrong.  All the rest is
yet-another-means-for-me-and-others-to-waste-time.  Sorry about this,
I perfectly understand the frustration it can cause, but someone has
to cut somewhere, and it turns out currently I'm that guy.