FYI: autoreconf (Was: YYSTYPE_IS_TRIVIAL)

[Akim Demaille]

| I currently get the following in my attempt to bootstrap:
| 
| 433-sic $ ./bootstrap
| Bootstrapping CVS Bison...
| autoreconf: working in `.'
| autoreconf: running: aclocal -I m4 --output=aclocal.m4t
| autoreconf: `aclocal.m4' is created
| autoreconf: running: gettextize --copy --force

Err???  How come --intl was not passed to gettextize???  It is for me!

Anyway, I have found a flaw in autoreconf, and I changed bootstrap
too.  I tested the combination several times, and it appears to work
reliably.

First, the Autoconf part:

Index: ChangeLog
from  Akim Demaille  <address@hidden>

        * bin/autoreconf.in (autoreconf): Run automake after autoconf and
        autoheader, so that automake does not complain about a missing
        config.h.in that was to be created.

Index: bin/autoreconf.in
===================================================================
RCS file: /cvsroot/autoconf/autoconf/bin/autoreconf.in,v
retrieving revision 1.94
diff -u -u -r1.94 autoreconf.in
--- bin/autoreconf.in 20 Apr 2002 06:09:01 -0000 1.94
+++ bin/autoreconf.in 24 Apr 2002 09:33:12 -0000
@@ -449,29 +449,16 @@
 
 
   # ------------------ #
-  # Running automake.  #
-  # ------------------ #
-
-  # Assumes that there is a Makefile.am in the topmost directory.
-  if (!-f 'Makefile.am')
-    {
-      verbose "$configure_ac: not using Automake";
-    }
-  else
-    {
-      # We should always run automake, and let it decide whether it shall
-      # update the file or not.  In fact, the effect of `$force' is already
-      # included in `$automake' via `--no-force'.
-      xsystem ($automake);
-    }
-
-
-  # ------------------ #
   # Running autoconf.  #
   # ------------------ #
 
-  # Don't try to be smarter than `autoconf', which does its own
-  # up to date checks.
+  # Don't try to be smarter than `autoconf', which does its own up to
+  # date checks.
+  #
+  # We prefer running autoconf before autoheader, because (i) the
+  # latter runs the former, and (ii) autoconf is stricter than
+  # autoheader.  So all in all, autoconf should give better error
+  # messages.
   xsystem ($autoconf);
 
 
@@ -481,7 +468,12 @@
 
   # We now consider that if AC_CONFIG_HEADERS is used, then autoheader
   # is used too.
-  # Just as for autoconf, up to date ness is performed by the tool itself.
+  #
+  # Just as for autoconf, up to date ness is performed by the tool
+  # itself.
+  #
+  # Run it before automake, since the latter checks the presence of
+  # config.h.in when it sees an AC_CONFIG_HEADERS.
   if (!$uses_autoheader)
     {
       verbose "$configure_ac: not using Autoheader";
@@ -489,6 +481,24 @@
   else
     {
       xsystem ($autoheader);
+    }
+
+
+  # ------------------ #
+  # Running automake.  #
+  # ------------------ #
+
+  # Assumes that there is a Makefile.am in the topmost directory.
+  if (!-f 'Makefile.am')
+    {
+      verbose "$configure_ac: not using Automake";
+    }
+  else
+    {
+      # We should always run automake, and let it decide whether it shall
+      # update the file or not.  In fact, the effect of `$force' is already
+      # included in `$automake' via `--no-force'.
+      xsystem ($automake);
     }
 
   chdir $cwd


-- 
Ashamed.