Comment.

src/alloc.c

@@ -1393,7 +1393,13 @@ allocate_string_data (s, nchars, nbytes)
 #ifdef DOUG_LEA_MALLOC
       /* Prevent mmap'ing the chunk.  Lisp data may not be mmap'ed
 	 because mapped region contents are not preserved in
-	 a dumped Emacs.  */
+	 a dumped Emacs.
+
+         In case you think of allowing it in a dumped Emacs at the
+         cost of not being able to re-dump, there's another reason:
+         mmap'ed data typically have an address towards the top of the
+         address space, which won't fit into an EMACS_INT (at least on
+         32-bit systems with the current tagging scheme).  --fx  */
       mallopt (M_MMAP_MAX, 0);
 #endif
 

src/buffer.c

@@ -420,10 +420,11 @@ The value is never nil.  */)
   XSETBUFFER (buf, b);
   Vbuffer_alist = nconc2 (Vbuffer_alist, Fcons (Fcons (name, buf), Qnil));
 
-  /* Fixme:  Protect against errors, which would trigger infinite
-     regress?  */
+  /* An error in calling the function here (should someone redfine it)
+     can lead to infinite regress until you run out of stack.  rms
+     says that's not worth protecting against.  */
   if (!NILP (Ffboundp (Qucs_set_table_for_input)))
-    /* buff is on buffer-alist, so no gcpro */
+    /* buf is on buffer-alist, so no gcpro.  */
     call1 (Qucs_set_table_for_input, buf);
 
   return buf;