The pragma around set_parents() is no longer needed: the base branch
(#5585) creates a value before setting its type. Before, the type of a new
string was stored first, so GCC had to assume that operator new could
change it again, kept the object branch of the inlined set_parents() alive,
and checked it against the string's allocation. Rewriting set_parents()
itself (if/else instead of switch) does not help.
Verified with Docker GCC 12.5, 13, 14, 15.3 and 16.2: the regression test
builds and passes without the pragma, and fails to build on develop.
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
With JSON_DIAGNOSTICS, GCC 13 to 15 report a false -Warray-bounds error
at -O3 when set_parents() is inlined right after a non-container value
(e.g., a string) was created: the std::map access in the object branch
is checked against the string allocation although m_type rules that
branch out. GCC 15 fixed the pattern from #4819 but not this one.
Suppress -Warray-bounds around set_parents() only, and add a regression
test that is compiled with -O3 -Werror=array-bounds on GCC.
Fixes#5742
Signed-off-by: Niels Lohmann <mail@nlohmann.me>
Niels Lohmann