Re: [OpenBIOS] [PATCH] next-property

[Olivier Danet]

On 03/02/2014 15:21, Artyom Tarasenko wrote:
> On Sat, Feb 1, 2014 at 9:54 PM, Olivier Danet  wrote:
> > obp_nextprop(node,name) calls the Forth word "next-property" then pops the
> > result. "next-property" should therefore always return something :
> > A string when the property provided is not the last one of the node.
> > Zero when the property provided is the last one of the node.
> > Zero when the property provided does not exist in the current node.
> >
> > Previously, next-property returned nothing when the property didn't exist.
> >
> > "next-property" behavior with last and nonexistant properties differs
> > between implementations of OF.
>
> Right. As discussed before, the current behavior matches the OBP behavior on
> sun4u machines.
>
> It would be nice to keep the comatibility with OBP. Can the C function
> obp_nextprop(node,name) be modified instead?
>
> Artyom
> > This problem was found with NetBSD on Sparc32. It ennumerates peripherals
> > when starting X11 and asks for the "name" and "device_type" property of many
> > nodes, including some which have neither "name" nor "device_type"
> > properties.
> > Signed-off-by: Olivier Danet
> >
> > --- forth/device/property.fs    (révision 1257)
> > +++ forth/device/property.fs    (copie de travail)
> > @@ -70,7 +70,7 @@
> >       2drop r> >dn.properties @
> >     else
> >       r> find-property dup if @ then
> > ?dup if >prop.next @ then
> >
> > dup if >prop.next @ then
> > then
> >
> > ?dup if
> >
> > ===================================================================

OpenBIOS's current behavior does not match any other !

For "next-property" we have 3 cases :
    a) The property is not the last
    b) The property is the last
    c) The property does not exist

Current OpenBIOS : a) -1 "string"    b) 0        c) nothing
Standard         : a) true "string"  b) false    c) false
Sun Ultra10      : a) -1 "string"    b) nothing  c) nothing
PowerPC iBook    : a) -1 "string"    b) -1 0 0   c) 0
Previous patch   : a) -1 "string"    b) 0        c) 0

Here is a patch adhering to the Ultra10's behavior, maybe other
Sun4u computers do differently :

===================================================================
--- arch/sparc32/romvec.c    (révision 1257)
+++ arch/sparc32/romvec.c    (copie de travail)
@@ -173,8 +173,9 @@

const char *obp_nextprop(int node, const char *name)
  {
-    int found;
+    int stackcnt;

+    stackcnt = dstackcnt;
      if (!name || *name == '\0') {
          // NULL name means get first property
          push_str("");
@@ -184,14 +185,13 @@
      }
      PUSH(node);
      fword("next-property");
-    found = POP();
-    if (!found) {
+    if (dstackcnt == stackcnt) {
          DPRINTF("obp_nextprop(0x%x, %s) (not found)\n", node, name);

return "";
      } else {
          char *str;
-
+        POP(); /* true */
          POP(); /* len */
          str = (char *) POP();

===================================================================
--- forth/device/property.fs    (révision 1257)
+++ forth/device/property.fs    (copie de travail)
@@ -70,14 +70,12 @@
      2drop r> >dn.properties @
    else
      r> find-property dup if @ then
-    ?dup if >prop.next @ then
+    dup if >prop.next @ then
    then

?dup if
      >prop.name @ dup cstrlen true
      ( phandle name-str name-len true )
-  else
-    false
    then
  ;

===================================================================

The 'standard' way makes next-property more useable as
a Forth word, but, really, I don't care.

Olivier