lopper:assist: Generate xparameters and CMake variables from the misc…

[sathishkumar-amd]

…_props node

Update Lopper to consume the misc_props node generated by SDT in pcw.dtsi and automatically propagate its properties into bare-metal xparameters.h macros and CMake metadata. This enables new user-defined properties added under misc_props to flow through to the BSP without requiring additional Lopper assist changes.

[sathishkumar-amd]

Hi @onkarharsh,

Please review this?

[onkarharsh]

The changes look good to me.
The Integration test shows some failures which do not seem to be related to the changes done here.

[zeddii]

Thanks @sathishkumar-amd — forwarding misc_props into xparameters + CMake so new SDT properties reach the BSP without an assist change is a nice additive approach, and it's a clean no-op for boards without a misc_props node. A few things before it goes in.

1. Identifier sanitization is incomplete (the main one). Both generators build the macro/variable name with:

var_name = prop_name.replace('xlnx,', '').replace('-', '_').upper()

That only handles the xlnx,foo-bar case. Any other DT-legal name emits an invalid C/CMake identifier: a non-xlnx vendor prefix (arm,foo → #define ARM,FOO), a dot (foo.bar → FOO.BAR), or a leading digit (2nd-x → 2ND_X). Since the goal is "arbitrary user-defined properties flow through without assist changes," this defeats the premise for anything outside that one naming pattern. A general scrub — replace any non-[A-Za-z0-9_] with _, and prefix a leading digit — makes it robust.

2. _iter_misc_props — use the list cast. It does prop_val = node.propval(prop_name) then len(prop_val). An untyped propval() can return a bare scalar (not a list), on which len() raises. Everywhere else in lopper the pattern is node.propval(name, list); using it here guarantees safe iteration for int / string / empty / list values alike.

3. Minor — de-duplicate the name transform. The var_name = … line is copy-pasted in both gen_misc_props_macros and gen_misc_props_cmake_vars; yielding the sanitized name from _iter_misc_props keeps one source of truth (and folds in the fix for #1). The isinstance(val, bool) branches are also effectively dead — DT propvals surface as int/str, not Python bool.

4. Tests. Nothing exercises the misc_props path — a small fixture with a misc_props node covering int, string, list, and empty properties, asserting the emitted macros/CMake vars, would lock it in and catch #1.

The red CI is the cpu-cluster-prune bug that was on master (fixed now) — a rebase clears it; onkarharsh's LGTM is informal (it doesn't set a formal review verdict). Otherwise the approach is sound.

[sathishkumar-amd]

@zeddii Thanks for the review, all points are addressed in the latest push:

1. Identifier sanitization
Added misc_prop_var_name() with a general scrub: strip xlnx,/amd, prefixes, replace any non-[A-Za-z0-9] with _, and prefix _ when the name starts with a digit. This covers vendor prefixes (arm,foo → ARM_FOO), dots (foo.bar → FOO_BAR), and leading digits (2nd-x → _2ND_X).

2. Safe propval iteration
_iter_misc_props now uses node.propval(prop_name, list) so scalar values are always list-wrapped before len() / iteration.

3. Single source of truth for name transform
_iter_misc_props yields (prop_name, var_name, value) with var_name from the shared helper; both gen_misc_props_macros and gen_misc_props_cmake_vars
consume it. Removed the dead isinstance(val, bool) branches — lopper surfaces booleans as empty flags or string "true"/"false" values, which are handled explicitly.