Re: cpplib problem

[Zack Weinberg <zack at wolery dot cumb dot org>]

On Mon, Mar 13, 2000 at 08:44:35PM -0800, Ulrich Drepper wrote:
> The current egcs cannot generate dependencies using -dM correctly.  I
> don't know when this was introduced since the last time I successfully
> compiled glibc is several weeks back.

Dependencies?  You mean the report on macros defined, right?  That is
broken...

$ echo "#define foo bar" | ./cpp -dM
Segmentation fault
$

> Anyhow, looking at the crash I think I came up with a usable patch.
> The only thing I could imagine people might want is to not use NULL as
> the indicator for an builtin and instead use, say, (void*)-1.  This
> would help catching other initialization errors.

This patch can't possibly be right.  The value field of struct
hashnode is a union.

The right patch should be to have dump_hash_helper verify that the
type field is T_MACRO.  Here's a patch that does that.  (Looks like we
need to emit newlines there, too.)

With this applied:

$ echo "#define foo bar" | ./cpp -dM
#define foo bar
$

I take it you got all the way to the glibc testsuite?  Last time I
tried to compile libc with a gcc snapshot, it would crash inside the
dynamic linker while trying to run rpcgen.

zw

	* cpphash.c (dump_hash_helper): Only dump nodes of type
	T_MACRO.  Emit a newline after each definition.

===================================================================
Index: cpphash.c
--- cpphash.c	2000/03/13 22:01:07	1.55
+++ cpphash.c	2000/03/14 05:57:04
@@ -1727,7 +1727,11 @@ dump_hash_helper (h, p)
   HASHNODE *hp = (HASHNODE *)h;
   cpp_reader *pfile = (cpp_reader *)p;
 
-  _cpp_dump_definition (pfile, hp->name, hp->length, hp->value.defn);
+  if (hp->type == T_MACRO)
+    {
+      _cpp_dump_definition (pfile, hp->name, hp->length, hp->value.defn);
+      CPP_PUTC (pfile, '\n');
+    }
   return 1;
 }