Re: PATCH and comments Re: PATCH: autoloaded parameters

[Sven Wischnowsky <wischnow@xxxxxxxxxxxxxxxxxxxxxxx>]

Bart Schaefer wrote:

> First comment:  This new code does not link with --disable-dynamic.

Grrr, Sven, Wake Up.

> Patch below.

Thanks.

> On Apr 28, 11:51am, Sven Wischnowsky wrote:
> } Subject: PATCH: autoloaded parameters
> }
> } Autoloaded parameters are stored in it like normal parameters, but
> } with a new flag (PM_AUTOLOAD). Then I changed the getnode-functions of 
> } paramtab to use a new function getparamnode() which calls
> } gethashnode2() and then checks if the Param returned has the
> } PM_AUTOLOAD flag set. If it has, the module is loaded.
> [...]
> }   - To avoid having a `local' with the name of a parameter that will
> }     be autoloaded trigger the loading of the module, I had to change
> }     calls like `paramtab->getnode()' to `gethashnode2(paramtab,...)'
> 
> Hrm.  Is this really the only place to hook this in?  Seems like doing
> it there means that you load the module as soon as anybody references
> the parameter, even if all that's done is to test whether it's set.
> 
> Heck, it means the module gets loaded if "unset ..." is used on the
> parameter!

No, it doesn't. Because I've also hooked it into createparam().

But, yes, parameters are autoloaded when tested for settedness (err,
well, you know what I mean). Somehow I liked that... To change this,
we would, as you know, have to fiddle some more in paramsubst().

> And what happens in createparamtable() if there's an environment variable
> with the same name as one of the autoloaded parameters?  Or is it not
> possible to have an autoloaded parameter at that point?

Well, setupvals()->createparamtable() is called before
init_bltinmods() registers autoloaded modules. So, currently there
shouldn't be a problem.

> }   - For now, I made `typeset' without arguments not print the
> }     autoloaded parameters because we have virtually no information
> }     about them as long as the module is not loaded. But since `which'
> }     reports builtins even when their module isn't loaded yet, we might
> }     want to change this. (What exactly should it print then?)
> 
> It should probably show the type as "undefined" the same as is done for
> autoloaded functions.

Hm, yes, sounds reasonable.

Bye
 Sven

--- os/params.c	Thu Apr 29 11:57:27 1999
+++ Src/params.c	Fri Apr 30 11:07:42 1999
@@ -2717,11 +2717,13 @@
     Param p = (Param) hn;
     char *t, **u;
 
-    if (p->flags & (PM_UNSET | PM_AUTOLOAD))
+    if (p->flags & PM_UNSET)
 	return;
 
     /* Print the attributes of the parameter */
     if (printflags & PRINT_TYPE) {
+	if (p->flags & PM_AUTOLOAD)
+	    printf("undefined ");
 	if (p->flags & PM_INTEGER)
 	    printf("integer ");
 	else if (p->flags & PM_ARRAY)
@@ -2755,6 +2757,11 @@
     }
 
     quotedzputs(p->nam, stdout);
+
+    if (p->flags & PM_AUTOLOAD) {
+	putchar('\n');
+	return;
+    }
     if (printflags & PRINT_KV_PAIR)
 	putchar(' ');
     else

--
Sven Wischnowsky                         wischnow@xxxxxxxxxxxxxxxxxxxxxxx