Remove Clone requirement for argument types
#353
Conversation
bnaecker
force-pushed
the
no-clone-closure
branch
7 times, most recently
from
9f35399 to
43b7c86
Compare
- Rework construction of the internal type-checking function to remove the clone requirement. Construct and call the type-check-function right at the site we already expand the probe argument closure, rather than outside. This means we don't need to expand things twice, and don't need to clone the closure, and so don't need the argument types to be cloneable either. - Fixup tests for new method, add some comments. - Update trybuild tests - Closes #136
bnaecker
force-pushed
the
no-clone-closure
branch
from
43b7c86 to
0477fd0
Compare
| unsafe { | ||
| ::std::arch::asm!( | ||
| "990: nop", | ||
| #probe_rec, | ||
| #in_regs | ||
| options(nomem, nostack, preserves_flags) | ||
| options(nostack) |
There was a problem hiding this comment.
my read of nomem and preserves_flags is that they're appropriate here; is there a reason you removed them?
There was a problem hiding this comment.
I was being cautious. The flags might be preserved, I think, but I'm not sure about memory accesses. The raw assembly we've written doesn't read or write any. But let's suppose there's a char * in one of the registers, because we're passing a string as a probe argument. DTrace will need to access that memory, but I don't really know if that counts here.
There was a problem hiding this comment.
I was being cautious. The flags might be preserved, I think, [...]
Yeah. It's really the nop or the mov so flags are preserved.
but I'm not sure about memory accesses. The raw assembly we've written doesn't read or write any. But let's suppose there's a
char *in one of the registers, because we're passing a string as a probe argument. DTrace will need to access that memory, but I don't really know if that counts here.
I don't think that's going to count. Say a debugger attaches to this process when the %rip is on this specific instruction; it then does loads and stores. That's fine.
The asm! block does not read from or write to any memory accessible outside of the asm! block. This allows the compiler to cache the values of modified global variables in registers across the asm! block since it knows that they are not read or written to by the asm!. The compiler also assumes that this asm! block does not perform any kind of synchronization with other threads, e.g. via fences.
Yeah, we're not going to screw up cached loads for example...
There was a problem hiding this comment.
Ok, I'll put these flags back.
There was a problem hiding this comment.
If I'm understanding correctly, then the nomem option is also fine in the first ASM block too, right? That's the clr %rax bit, which does potentially touch the flags, but does not access any memory at all.
Clonerequirement #136