https://gcc.gnu.org/bugzilla/show_bug.cgi?id=92793
Bug ID: 92793
Summary: Fortran Location Data for Diagnostic lacks the column
number – when passing on to ME
Product: gcc
Version: 10.0
Status: UNCONFIRMED
Keywords: diagnostic
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: burnus at gcc dot gnu.org
Target Milestone: ---
Currently, all location_t locations (such as gfc_current_locus->lb->location)
always point to the beginning of the line, i.e. LOCATION_COLUMN(...) == 0.
For front-end diagnostic, this is corrected in gcc/fortran/error.c's
gfc_format_decoder using:
unsigned int offset = loc->nextc - loc->lb->line;
int loc_num = text->get_location (0) == UNKNOWN_LOCATION ? 0 : 1;
location_t src_loc
= linemap_position_for_loc_and_offset (line_table,
loc->lb->location,
offset);
But this correction does not happen when setting the location for 'tree', e.g.:
build_decl (ns->proc_name->declared_at.lb->location
or
tmp = fold_build1_loc (code->expr1->where.lb->location, LOOP_EXPR
or
…
The question is how to fix this best. My feeling is that it should already be
fixed in the scanner such that loc->lb->location is always correct and we
can get rid off the offset calculation in error.c
