Hi Eric,

* Eric Blake wrote on Tue, Nov 11, 2008 at 08:16:55PM CET:
> I'm debating about committing this patch.  It shaves off a lot of blank lines 
> in configure output.  But I also found at least one use case in the wild that 
> changes semantics if I commit it:
> http://lists.gnu.org/archive/html/bug-coreutils/2008-11/msg00049.html

Ouch.  Are you certain that removing this newline is really worth the
price of a silent incompatibility?

> By making AC_DEFINE no longer force a leading newline, users that define a 
> shell variable then call AC_DEFINE on the same line have become localized 
> assignments, rather than persistent.  How much should we worry about breaking 
> existing scripts, given that I did not see much common use of
>  var=val AC_DEFINE
> in my searching?

We should worry very much about any kind of valid construct in user's
scripts, IMVHO.

If you persist in it, how about letting autoconf warn about instances of
  ^[\t ]*[^\t ]+[\t ]*\<AC_DEFINE

in the input (inconsistent regex notation)?

I fear that the number of hard-to-detect incompatibilities is really
growing a bit more than we would like, in the current development tree.

Cheers,
Ralf


Reply via email to