The way direct mode check was done in mdb_condition_handler() was to check if the frame
immediately previous to the current frame is a direct-mode frame (SFT_DM) or if we are in compile
time and if so check if the current frame is a direct-mode frame. This meant that if an error
occurs inside an indirection usage (which translates to an arbitrary number of uncounted frames
underneath the SFT_DM frame), we would end up treating that error as coming from a non-direct-mode
action which is inconsistent since the same error outside of the indirection will be treated as if
it happened in direct-mode. This inconsistency is fixed by checking all frames back from the current
one until we see either a counted frame or a SFT_DM frame. And using that to decide whether we
are in direct mode or not.
Note that this change does not discard the special handling previously done for TREF(compile_time).
In that case, we start from "frame_pointer" instead of "frame_pointer->old_frame_pointer". I don't see
a reason why that is needed but did not want to change it without reason so letting it stay as is.
The way direct mode check was done in mdb_condition_handler() was to check if the frame immediately previous to the current frame is a direct-mode frame (SFT_DM) or if we are in compile time and if so check if the current frame is a direct-mode frame. This meant that if an error occurs inside an indirection usage (which translates to an arbitrary number of uncounted frames underneath the SFT_DM frame), we would end up treating that error as coming from a non-direct-mode action which is inconsistent since the same error outside of the indirection will be treated as if it happened in direct-mode. This inconsistency is fixed by checking all frames back from the current one until we see either a counted frame or a SFT_DM frame. And using that to decide whether we are in direct mode or not.
Note that this change does not discard the special handling previously done for TREF(compile_time). In that case, we start from "frame_pointer" instead of "frame_pointer->old_frame_pointer". I don't see a reason why that is needed but did not want to change it without reason so letting it stay as is.