On 4/5/19 2:25 PM, Martin Sebor wrote:
On 4/5/19 10:45 AM, Martin Sebor wrote:
The change committed yesterday to handle the C++ 2a string literals
as non-type template arguments treats empty string initializers for
trailing pointer array elements as zero. That's of course wrong
because addresses of literals are non-zero.
The attached tweak constrains the trimming of trailing zero
initializers to either non-pointer initializers or non-pointer
types.
The patch passes x86_64 bootstrap (but so did the change yesterday).
The tests are still running. Is it okay to commit if the tests pass?
I'm also separately bootstrapping using the configure options from
the report where apparently it caused the x86_64 bootstrap to fail.
This bootstrap also passed.
Attached is a slightly updated patch that, besides fixing this
bug, also reverts the use of TYPE_HAS_TRIVIAL_DFLT() in favor
of trivial_type_p(). Using the macro turned out not to be
quite right -- it caused cpp0x/constexpr-63241.C to be rejected.
I missed that among all the other test failures.
I'm still not sure what happened with my yesterday's build that
it doesn't show the bootstrap failure.
OK.
Jason
