https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69126
--- Comment #8 from David Malcolm <dmalcolm at gcc dot gnu.org> ---
Looking at calls to linemap_compare_locations, the crucial comparison is the
comparison of the location of the 2nd diagnostic with that of the "ignore"
pragma:
Breakpoint 2, linemap_compare_locations (set=0x7ffff7ffb000, pre=226930,
post=post@entry=2147483645)
at ../../src/libcpp/line-map.c:1311
(gdb) call inform (226930, "pre, the location of the ignore")
/tmp/test.cc:17:24: note: pre, the location of the ignore
MACRO;
^
(gdb) call inform (2147483645, "post, the location of the diagnostic")
/tmp/test.cc:13:9: note: post, the location of the diagnostic
int x;
^
/tmp/test.cc:17:5: note: in expansion of macro ‘MACRO’
MACRO;
^~~~~
The initial value of l1 == post (location of the diagnostic):
(gdb) p /x l1
$15 = 0x7ffffffd
linemap_compare_locations expands macro locations at LRK_MACRO_EXPANSION_POINT
and obtains:
(gdb) p l1
$19 = 226308
(gdb) call inform (226308, "")
/tmp/test.cc:17:5: note:
MACRO;
^~~~~
and hence treats the location of "post" (the diagnostic) as the "M" of the
macro expansion point.
However, the location of the "ignore" within the macro definition is 226930,
which is *after* that "M":
(gdb) call inform (226930, "pre, the location of the ignore")
/tmp/test.cc:17:24: note: pre, the location of the ignore
MACRO;
^
Hence that strange-looking synthesized location of the "ignore" directive
within the tokens synthesized by _cpp_do__Pragma is effectively *after* that of
the macro expansion point. Hence the "ignore" is treated as occurring after
the diagnostic, hence the "ignore" doesn't affect the diagnostic, and the
warning is emitted.
FWIW, there's also some logic in linemap_compare_locations for handling
locations in the same macro expansion, but this isn't the case here as the
location of the pragma has been recorded as an ordinary location.
