Added "nop" to IM_DEBUG_BREAK macro on GCC to work around GDB bug (#7266)

There are two issues here - first, this macro uses AT&T specific syntax with $, which is not necessary. Also, some assemblers (nasm) emit different bytes for "int 3" and "int3", so it's better to use "int3" (cd 03 vs cc)

More importantly, GDB has some failing assertion whenever stepping after hitting an "int3" instruction. This makes it practically useless, as is. For some reason, putting a nop afterwards as a workaround is okay.

Related discussions:
https://sourceware.org/bugzilla/show_bug.cgi?id=31194
https://lists.sr.ht/~skeeto/public-inbox/%3C2d3d7662a361ddd049f7dc65b94cecdd%40disroot.org%3E
This commit is contained in:
Peter0x44
2024-01-25 10:16:27 +00:00
committed by ocornut
parent 8491cf36ad
commit f1960b60c1
2 changed files with 3 additions and 2 deletions

View File

@ -300,11 +300,11 @@ namespace ImStb
#elif defined(__clang__)
#define IM_DEBUG_BREAK() __builtin_debugtrap()
#elif defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__))
#define IM_DEBUG_BREAK() __asm__ volatile("int $0x03")
#define IM_DEBUG_BREAK() __asm__ volatile("int3;nop")
#elif defined(__GNUC__) && defined(__thumb__)
#define IM_DEBUG_BREAK() __asm__ volatile(".inst 0xde01")
#elif defined(__GNUC__) && defined(__arm__) && !defined(__thumb__)
#define IM_DEBUG_BREAK() __asm__ volatile(".inst 0xe7f001f0");
#define IM_DEBUG_BREAK() __asm__ volatile(".inst 0xe7f001f0")
#else
#define IM_DEBUG_BREAK() IM_ASSERT(0) // It is expected that you define IM_DEBUG_BREAK() into something that will break nicely in a debugger!
#endif