On Tue, 2024-07-02 at 18:02 -0400, David Malcolm wrote:
> On Tue, 2024-07-02 at 16:17 +0000, Qing Zhao wrote:
[...snip...]
>
> > + path.add_event (cur_ch->condition, NULL_TREE, 0,
> > + "when the condition is evaluated to %s",
> > + cur_ch->is_true_path ? "true" : "false");
>
> add_event's 2nd param is the function in which the event occurs,
> which
> you may want to set to cfun, and set the stack depth to 1, rather
^^^^
cfun's decl, that is, sorry.
[...snip...]
>
> Or, all those
>
> if (for_array_bound)
> warned = warning_at (richloc, OPT_Warray_bounds_,
> /* etc */);
>
> could become:
>
> if (for_array_bound)
> {
> copy_history_diagnostic_path path (location, stmt);
> warned = warning_at (richloc, OPT_Warray_bounds_,
> /* etc */);
> }
Or rather:
if (for_array_bound)
{
copy_history_location richloc (location, stmt);
warned = warning_at (&richloc, OPT_Warray_bounds_,
/* etc */);
}
where the copy_history_location is a new rich_location subclass, and it
only populates the path if it's asked for one (when we're actually
emitting the diagnostic).
> or somesuch, explicitly deferring creation of the path until we're
> actually about to emit a warning, and putting the logic in
> copy_history_diagnostic_path's ctor.
[...snip...]
Dave
