On Mon, 16 Sep 2019, Liu Hao wrote:
在 2019/9/16 16:15, Martin Storsjö 写道:
On x86, mingw's 80 bit long double doesn't match MSVC's 64 bit long double.
This was done for a number of other functions in
d8ed328d4656cc467d692e1c5bd2dc7007695ef8, but the nanl function was
overlooked (in addition to all the complex functions that I've
omitted in my testing so far).
Signed-off-by: Martin Storsjö <[email protected]>
---
This patch looks good to me.
Thanks, pushed.
[Off topic:
Would you please try building Clang itself with the _FORTIFY_SOURCE
patch? As our CRT doesn't provide `__chk_fail()` and `__*_chk()`
functions I am suffering from undefined references when building GCC. I
have been tampering with GCC configure variables for quite a while but
it keeps failing sooner or later.
- end of Off topic]
I've built Clang with current mingw-w64 master without issues, as Clang's
build process doesn't set _FORTIFY_SOURCE anywhere. (And as noted
elsewhere, Clang doesn't really support __builtin_object_size through
inlined functions, so these inline wrappers essentially do nothing at all
in Clang.)
But if I do build with that flag set, I do need to link with -lssp, which
contains the __chk_fail() and corresponding functions. The only project I
regularly test building that uses _FORTIFY_SOURCE, VLC, already happens to
use -lssp (as it also builds with -fstack-protector-strong, which needs
the same.
(In my setups, even though I don't use anything else from gcc or binutils,
I do check out the libssp subdirectory from gcc and build that.)
// Martin
_______________________________________________
Mingw-w64-public mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
