Re: plans?

[Paolo Bonzini]

On 08/26/2010 08:03 PM, Ralf Wildenhues wrote:
> do you intend to release 2.68 soon, so that GNOME packages (with URLs in
> AC_INIT's BUG-ADDRESS) can build again?
>
> If not, I'd like to ping the atomic config.cache and the long lines in
> testsuite patches.  Actually, I might want to ping them regardless,
> since more intrusive changes are going in ATM anyway.

I think this is a problem with the current release mode of Autoconf. 
Since 2.64 there has been no release without a serious regression:

** The macro AC_LANG_COMPILER no longer fails on embedded systems that
   lack fopen in the C library, such as AVR or RTEMS (regression
   introduced in 2.64).

** The following undocumented autoconf macros, removed in Autoconf 2.64,
   have been reinstated:
   AH_CHECK_HEADERS

** The macros AC_TYPE_INT8_T, AC_TYPE_INT16_T, AC_TYPE_INT32_T, and
   AC_TYPE_INT64_T work again.  Regression introduced in 2.65.

** AC_CONFIG_SUBDIRS with more than one subdirectory at a time works
   again.  Regression introduced in 2.66.

** AC_CHECK_SIZEOF of a pointer type works again.  Regression
   introduced in 2.66.

** AC_INIT again allows URLs with '?' for its BUG-REPORT argument.
   Regression introduced in 2.66 [and present in 2.67].

Guys, Autoconf cannot afford this.  Miss the boat for a widespread Linux 
distro, and you risk buggy scripts in the wild for years literally.

We _really_, _really_ need a stable branch of Autoconf where every 
regression is treated the same way as security problems and released in 
a matter of days.  And where only regressions are fixed using the _least 
invasive_ approach.

Paolo