This is a bit odd having properties space being checked for in UI_block_begin,

Why not pass the search string in to ED_region_panels_layout_ex ?

Side-note, contexts[], category_override, search_string could be wrapped into a struct if the arguments are getting a bit much.

Comment who owns the data, is it allocated etc.

Also, shouldn't this be const? If it can be written to, it should have a size variable to avoid buffer overrun.

These should use translated strings too. See rna_translate_ui_text.

Should use RNA_property_enum_items_gettexted, same for other descriptions.

"later used to deactivate it" - I find that a bit confusing. Isn't it more like "later used to determine which items to keep enabled, and which to disable"?

No need for void * cast.

Would avoid the double negated logic below and use search_enabled.

This doesn't seem like something that should be in a UI_block_layout_resolve() function, it should be done by the caller I guess. Given that UI_block_layout_resolve() is called in quite some places, I'd suggest adding a wrapper, say UI_block_layout_end() that resolves the layout and does the search post-processing (disable buttons).

Number seems a bit arbitrary, maybe use UI_MAX_NAME_STR, or something else that is defined.
I'd also suggest putting this into a runtime only struct that is NULLed on file read. I don't see too great use in keeping the search string when loading files, not worth the memory added to .blends.

Good point, that's not great. I passed it to ed_panel_draw from ED_region_panels_layout_ex which can get it from the area with an existing helper function. That's nice because it stays generalized.

Ah right, the meaning of this flag was reversed before and I didn't update the comment.

I'm a bit confused here, I thought these functions accounted for that. For example:

const char *RNA_property_ui_name(const PropertyRNA *prop)
{
  return CTX_IFACE_(prop->translation_context, rna_ensure_property_name(prop));
}

Would this not do the translation?

It seems we do need a cast here to avoid a warning freeing the const items_array. That's how it's done everywhere else at least. But I changed it to a (EnumPropertyItem *) cast, that's more clear about the reasoning.

I refactored this area a bit. Now you run UI_block_apply_search_filter() right before UI_block_layout_resolve(), and the two are completely separate processes. Should make more sense now.

I thought it might be nice to keep it stored because I figured people might want to keep it on more permanently in some cases, but I don't really care much, so I moved it to a new runtime struct.

Note that the size definition is just in a comment because this is in DNA.

Better add a comment on why this is disabled.

Use rna_area_from_space(ptr) here, avoid context.

You have to NULL-check if the runtime data already exists. The init() callback is called multiple times. E.g. on every area size change, window size change or window move. That should also be the source of the memory leak in D8859.