[PATCH] pei386, ld auto-import

[Charles Wilson <cwilson at ece dot gatech dot edu>]

In this message:

Missing variable 'pe_data_import_dll'
http://sources.redhat.com/ml/binutils/2001-08/msg00186.html

Nick Clifton reported that the recently accepted auto-import changes to 
ld on pei386 caused a build problem on other pe platforms that don't 
support DLLs (specifically arm-epoc-pe).

This was because the auto-import changes added a reference in pe-dll.c 
to a variable in pe.em -- but that variable wasn't defined unless 
DLL_SUPPORT was defined.

Now, Nick posted a workaround to the problem here:
http://sources.redhat.com/ml/binutils/2001-08/msg00187.html
but AFAICT it was never applied to CVS.  (Plus, it was just a 
workaround).  In that message, Nick also suggested the following:

   I still
   feel however, that there ought to be a better way to solve this
   problem.  My suggestion is that the definition of DLL_SUPPORT ought
   to be set in ld/configure.tgt rather than ld/emultemp/pe.em and then
   tested in ld/pe-dll.c before it uses variables that are only defined
   in pe.em.

That's not what I did, because ld/configure.tgt seems to only set shell 
variables, which are later used to control Makefile options.  However, 
the list of acceptable shell variables is fixed: targ_emul, 
targ_extra_emuls, targ_extra_libpath, and targ_extra_ofiles.  These are 
then (eventually) used by configure to fixup the Makefile.in's when 
generating the Makefile's.  So, sure, it's *possible* to do what Nick 
suggests -- by changing configure.in/configure so that it will replace 
@DLL_SUPPORT@, adding a few more rules to Makefile.in, adding 
DLL_SUPPORT where appropriate in configure.tgt, etc., but ...

It really doesn't seem right to me that "DLL_SUPPORT" be added to the 
list of stuff controlled by configure.tgt, which is currently limited to 
things like targ_*.  Comments?

So, the attached patch is identical to the one I posted a few days ago, 
but the Subject line of that earlier message was probably misleading. 
Anyway, the attached patch fixes this problem by:
  a) make the offending variable (pe_data_import_dll) static within pe.em
  b) add a new function in pe.em (pe_get_data_import_dll_name) which 
either returns this variable (#ifdef DLL_SUPPORT), or returns a constant 
  string (#else).  However, at least the function itself is always defined.

Could somebody on arm-epoc-pe or other non-DLL-supporting pe platform 
please verify that this patch allows successful compilation/operation of 
binutils?  the patched version does work on cygwin (but the unpatched 
version worked on cygwin, too)

Thanks,
Chuck
2001-09-21  Charles Wilson  <cwilson@ece.gatech.edu>

	* emultempl/pe.em(pe_data_import_dll): Make static.
	(pe_get_data_import_dll_name): New accessor function.
	* pe-dll.c(pe_create_import_fixup): call 
	pe_get_data_import_dll_name() from pe.em, instead of
	directly accessing pe_data_import_dll variable from pe.em.
Index: ld/pe-dll.c
===================================================================
RCS file: /cvs/src/src/ld/pe-dll.c,v
retrieving revision 1.30
diff -u -r1.30  old/ld/pe-dll.c new/ld/pe-dll.c
--- old/ld/pe-dll.c	2001/09/19 05:33:33	1.30
+++ new/ld/pe-dll.c	2001/09/21 06:09:49
@@ -123,6 +123,9 @@
 static void
 add_bfd_to_link PARAMS ((bfd *, const char *, struct bfd_link_info *));
 
+extern char *
+pe_get_data_import_dll_name PARAMS(( ));  /* Defined in emultempl/pe.em.  */
+
 /* For emultempl/pe.em.  */
 
 def_file * pe_def_file = 0;
@@ -2065,10 +2068,9 @@
     }
 
   {
-    extern char * pe_data_import_dll;  /* Defined in emultempl/pe.em.  */
-    
-    bfd *b = make_import_fixup_entry (name, fixup_name, pe_data_import_dll,
-				      output_bfd);
+    bfd *b = make_import_fixup_entry (name, fixup_name,
+                                      pe_get_data_import_dll_name(), 
+                                      output_bfd);
     add_bfd_to_link (b, b->filename, &link_info);
   }
 }
Index: ld/emultempl/pe.em
===================================================================
RCS file: /cvs/src/src/ld/emultempl/pe.em,v
retrieving revision 1.52
diff -u -r1.52  old/ld/emultempl/pe.em new/ld/emultempl/pe.em
--- old/ld/emultempl/pe.em	2001/09/18 10:10:21	1.52
+++ new/ld/emultempl/pe.em	2001/09/21 06:09:51
@@ -153,6 +153,7 @@
 static char *pe_implib_filename = NULL;
 static int pe_enable_auto_image_base = 0;
 static char *pe_dll_search_prefix = NULL;
+static char *pe_data_import_dll = NULL;
 #endif
 
 extern const char *output_filename;
@@ -879,7 +880,32 @@
   return 1;
 }
 
-char *pe_data_import_dll;
+/* Previously, pe-dll.c directly accessed pe_data_import_dll,
+ * which was only defined if DLL_SUPPORT.  This cause a build
+ * failure on certain targets. At least this function will
+ * exist regardless of whether DLL_SUPPORT is defined or not.
+ *
+ * However, it's still a kludge.  pe-dll.c shouldn't directly
+ * call any functions other than the gld_${EMULATION_NAME}_* 
+ * Nick suggests the following method:
+ *   I still feel however, that there ought to be a better 
+ *   way to solve this problem.  My suggestion is that the 
+ *   definition of DLL_SUPPORT ought to be set in ld/configure.tgt 
+ *   rather than ld/emultemp/pe.em and then tested in ld/pe-dll.c 
+ *   before it uses variables that are only defined in pe.em.
+ * However, I don't understand this.  ld/configure.tgt sets SHELL
+ * variables, not #define variables.  You'd need #define variables
+ * to #ifdef out the offending sections of code from pe-dll.c
+ */
+char *
+pe_get_data_import_dll_name ()
+{
+#ifdef DLL_SUPPORT
+  return pe_data_import_dll;
+#else
+  return "unknown";
+#endif
+}
 
 static void
 pe_find_data_imports ()