On 2/19/2013 9:12 PM, Daniel Colascione wrote:
> I don't know why we suddenly started seeing this behavior.

Revision 111820, "Prefer `message1' over `message'.", caused us to start calling
message1 instead of message in sheap.c. message1 ends up calling message_dolog,
which tries to allocate a *Messages* buffer. Maybe Fdump_emacs should bind
Vmessage_max_log?

Anyway, I've installed the following patch to trunk, which fixes the problem. If
we want to go back to using message1, let's somehow make sure it doesn't log.

revno: 111830
committer: Daniel Colascione <dancol@dancol.org>
branch nick: cyg
timestamp: Tue 2013-02-19 21:43:53 -0800
message:
  2013-02-19  Daniel Colascione <dancol@dancol.org>

          * sheap.c (report_sheap_usage): Use message, not message1, so
         that we don't try to create a buffer while we're in the middle
         of dumping Emacs.  Explain why.
diff:
=== modified file 'src/ChangeLog'
--- src/ChangeLog       2013-02-19 03:29:28 +0000
+++ src/ChangeLog       2013-02-20 05:43:53 +0000
@@ -1,3 +1,9 @@
+2013-02-19  Daniel Colascione <dancol@dancol.org>
+
+       * sheap.c (report_sheap_usage): Use message, not message1, so
+       that we don't try to create a buffer while we're in the middle
+       of dumping Emacs.  Explain why.
+
 2013-02-19  Eli Zaretskii  <eliz@gnu.org>

        * w32proc.c (new_child): Avoid leaking handles if the subprocess

=== modified file 'src/sheap.c'
--- src/sheap.c 2013-02-17 16:49:27 +0000
+++ src/sheap.c 2013-02-20 05:43:53 +0000
@@ -91,5 +91,8 @@
   char buf[200];
   sprintf (buf, "Static heap usage: %d of %d bytes",
           bss_sbrk_ptr - bss_sbrk_buffer, STATIC_HEAP_SIZE);
-  message1 (buf);
+  /* Don't change this call to message1! message1 can log
+     messages, and at this point, we're not allowed to create
+     buffers.  */
+  message ("%s", buf);
 }