On 23.04.2025 18:50, Martin Storsjö wrote:
On Wed, 23 Apr 2025, Martin Storsjö wrote:
On Wed, 23 Apr 2025, Jacek Caban wrote:
Instead of for specific architectures. Fixes setjmp on ARM64EC.
---
v2: Reorder checks to simplify the code a bit. We could probably
simplify it further
by removing defined(__SEH__) checks.
mingw-w64-headers/crt/setjmp.h | 16 +++++++---------
1 file changed, 7 insertions(+), 9 deletions(-)
diff --git a/mingw-w64-headers/crt/setjmp.h
b/mingw-w64-headers/crt/setjmp.h
index 2dc7d8739..8bfab5c58 100644
--- a/mingw-w64-headers/crt/setjmp.h
+++ b/mingw-w64-headers/crt/setjmp.h
@@ -229,16 +229,14 @@ void * __cdecl __attribute__ ((__nothrow__))
mingw_getsp (void);
# define longjmp __mingw_longjmp
int __cdecl __attribute__ ((__nothrow__,__returns_twice__))
__mingw_setjmp(jmp_buf _Buf);
__MINGW_ATTRIB_NORETURN __attribute__ ((__nothrow__)) void
__mingw_longjmp(jmp_buf _Buf,int _Value);
-# elif defined(__SEH__) && !defined(__USE_MINGW_SETJMP_NON_SEH)
-# if defined(__aarch64__) || defined(_ARM64_) ||
defined(__arm__) || defined(_ARM_)
-# define setjmp(BUF) _setjmp((BUF), __builtin_sponentry())
-# elif (__MINGW_GCC_VERSION < 40702) && !defined(__clang__)
-# define setjmp(BUF) _setjmp((BUF), mingw_getsp())
-# else
-# define setjmp(BUF) _setjmp((BUF), __builtin_frame_address (0))
-# endif
-# else
+# elif !defined(__SEH__) || defined(__USE_MINGW_SETJMP_NON_SEH)
# define setjmp(BUF) _setjmp((BUF), NULL)
+# elif __has_builtin(__builtin_sponentry)
+# define setjmp(BUF) _setjmp((BUF), __builtin_sponentry())
+# elif (__MINGW_GCC_VERSION < 40702) && !defined(__clang__)
+# define setjmp(BUF) _setjmp((BUF), mingw_getsp())
+# else
+# define setjmp(BUF) _setjmp((BUF), __builtin_frame_address (0))
Sorry, I think I was unclear; the condition which I find most
unwieldy is the one above this block entirely. These ones aren't all
that bad. The one I found problematic was this one:
# elif ((defined(_ARM_) || defined(__arm__) || defined(_ARM64_) ||
defined(__aarch64__)) && (!defined(__SEH__) ||
!__has_builtin(__builtin_sponentry) ||
defined(__USE_MINGW_SETJMP_NON_SEH)))
The annoying thing here is that it repeats many of the conditions
that we have below...
It would at least be less annoying if we'd get rid of the duplicate
_ARM_ vs __arm__ checks everywhere. But even then, I'm not sure how
much we can simplify it? As we do need to decide if we want to use
the SEH-requiring _setjmp or __mingw_setjmp.
As far as I see it, we have these cases we've considered:
- x86_64, with __SEH__ defined, with mingw_getsp() or
__builtin_frame_address(), with GCC or Clang
- Old aarch64 clang, with __SEH__ defined but without
__builtin_sponentry(). We probably don't need to support this any
longer - this was for Clang before version 9.
- Old armv7 clang, without __SEH__ defined. (This was Clang before
version 15.) Not very important, but I think we added __SEH__ in sync
with adding __builtin_sponentry() there, so I think this case may be
handled essentially for free.
- New aarch64 and armv7 clang, with __SEH__ defined, with
__builtin_sponentry() available
- Aarch64 and armv7 clang, with __SEH__ defined, but the user defined
__USE_MINGW_SETJMP_NON_SEH
- Aarch64 GCC, probably without __SEH__ defined, with no
__builtin_sponentry()
I forgot a couple other cases:
- Aarch64 and armv7 clang, but with -fdwarf-execptions; this makes
__SEH__ no longer be predefined. Not a very important case, but I
think we do need to check for __SEH__ to cover many of the odd
configurations anyway.
Yes, I assumed you were referring to that part, but I didn’t immediately
see a way to simplify it. Even if we drop support for older Clang
versions, the GCC case still requires the target macro checks. That
said, your point about __SEH__ and __builtin_sponentry() does help
reduce some duplication, so I’ll send an updated version.
Thanks,
Jacek
_______________________________________________
Mingw-w64-public mailing list
Mingw-w64-public@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mingw-w64-public
