Re: regression in autoconf-2.62 vs. 2.61

[Ralf Wildenhues]

Hello Karsten,

Thanks for the bug report.

* Karsten Hopp wrote on Wed, Jun 04, 2008 at 05:00:49PM CEST:
> (second attempt, first one got stuck in the maintainer queue)

Usually there is some latency due to the human involved in the
first-time moderation, and/or the system involved in spam filtering.

> https://bugzilla.redhat.com/show_bug.cgi?id=449245 lists two issues
> after updating autoconf in Fedora to 2.62.
>
> - same line comments are now really broken where they worked before
> even though this was discouraged.  The documentation should probably
> reflect this new behaviour if this breakage was intentionally.

Either that, or the behaviour should be fixed.  I prefer at least a
simple fix.

> - autoconf shouldn't temper with symbols it hasn't been told to
> substitute.

Well, that, however, is done intentionally; see this comment in
lib/autoconf/status.m4:
    # Replace #undef with comments.  This is necessary, for example,
    # in the case of _POSIX_SOURCE, which is predefined and required
    # on some systems where configure will not decide to define it.

(maybe we should document that?)

Let's have a test case to see what happens:

cat >configure.in <<\EOF
AC_INIT
AC_CONFIG_HEADER(config.h)
AC_DEFINE([DEFINED_SYMBOL], [1])
if false; then
  AC_DEFINE([UNDEFINED_SYMBOL], [1])
fi
AC_OUTPUT
EOF
cat >config.h.in <<\EOF
#undef DEFINED_SYMBOL /* some comment */
#undef UNDEFINED_SYMBOL /* some other comment */
#undef DEFINED_SYMBOL
#undef UNDEFINED_SYMBOL
EOF

Then, with Autoconf 2.13, 2.59, 2.61, 2.62, let's run
  autoconf
  ./configure

to get the following config.h files, respectively:

--- 2.13 ---
/* config.h.  Generated automatically by configure.  */
#define DEFINED_SYMBOL 1 /* some comment */
/* #undef UNDEFINED_SYMBOL */ /* some other comment */
#define DEFINED_SYMBOL 1
/* #undef UNDEFINED_SYMBOL */

--- 2.59 ---
/* config.h.  Generated by configure.  */
/* #undef DEFINED_SYMBOL */ /* some comment */
/* #undef UNDEFINED_SYMBOL */ /* some other comment */
#define DEFINED_SYMBOL 1
/* #undef UNDEFINED_SYMBOL */

--- 2.61 ---
/* config.h.  Generated from config.h.in by configure.  */
#undef DEFINED_SYMBOL /* some comment */
#undef UNDEFINED_SYMBOL /* some other comment */
#define DEFINED_SYMBOL 1
/* #undef UNDEFINED_SYMBOL */

--- 2.62 ---
/* config.h.  Generated from config.h.in by configure.  */
#define DEFINED_SYMBOL 1
/* #undef UNDEFINED_SYMBOL /* some other comment */ */
#define DEFINED_SYMBOL 1
/* #undef UNDEFINED_SYMBOL */


The commenting out of line 2 with 2.59 looks like a bug to me.
Likewise, the first two #undef line with 2.61 look buggy to me.

The fact that 2.62 removes the comment after the defined symbol
looks like a QoI issue (i.e., would be nice to fix, but no error
if not).  The nested comment, however, is clearly undesirable.

So here's a patch for a simple fix, to generates this instead:
--- 
/* config.h.  Generated from config.h.in by configure.  */
#define DEFINED_SYMBOL 1
/* undef UNDEFINED_SYMBOL */
#define DEFINED_SYMBOL 1
/* undef UNDEFINED_SYMBOL */
---

Question is if that's good enough.  Judging from autoconf history as
seen above, comments have been problematic ever since 2.59.  I haven't
tried incomplete (multi-line) comments, or C++ style comments.

I'm willing to add a test case after we've decided what to be desirable.

Cheers,
Ralf

2008-06-04  Ralf Wildenhues  <address@hidden>

        * lib/autoconf/status.m4 (_AC_OUTPUT_HEADERS_PREPARE): For
        undefined preprocessor macros that are followed by a comment
        in the header template, do not create nested comments in the
        output.
        * NEWS: Update.
        Report by Karsten Hopp <address@hidden>.

diff --git a/NEWS b/NEWS
index ccf4418..467c7b8 100644
--- a/NEWS
+++ b/NEWS
@@ -12,6 +12,9 @@ GNU Autoconf NEWS - User visible changes.
 ** Autotest testsuites accept an option --jobs[=N] for parallel testing.
    Alternatively, testsuites can act as GNU make jobserver client.
 
+** Config header templates `#undef UNDEFINED /* comment */' do not lead to
+   nested comments any more; regression introduced in 2.62.
+
 
 * Major changes in Autoconf 2.62 (2008-04-05) [stable]
   Released by Eric Blake, based on git versions 2.61a.*.
diff --git a/lib/autoconf/status.m4 b/lib/autoconf/status.m4
index 50be77b..d5ed323 100644
--- a/lib/autoconf/status.m4
+++ b/lib/autoconf/status.m4
@@ -832,9 +832,9 @@ cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
   }
   split(mac1, mac2, "(") #)
   macro = mac2[1]
+  prefix = substr(line, 1, index(line, defundef) - 1)
   if (D_is_set[macro]) {
     # Preserve the white space surrounding the "#".
-    prefix = substr(line, 1, index(line, defundef) - 1)
     print prefix "define", macro P[macro] D[macro]
     next
   } else {
@@ -842,7 +842,7 @@ cat >>$CONFIG_STATUS <<_ACEOF || ac_write_fail=1
     # in the case of _POSIX_SOURCE, which is predefined and required
     # on some systems where configure will not decide to define it.
     if (defundef == "undef") {
-      print "/*", line, "*/"
+      print "/*", prefix defundef, macro, "*/"
       next
     }
   }