| commit | da0f20df4f4f736f6d0ce2088cdc14e99ecf5f59 | |
| author | Douglas Katzman <dougk@google.com> | |
| Wed, 14 Jan 2015 18:12:56 +0000 (14 13:12 -0500) | ||
| committer | Douglas Katzman <dougk@google.com> | |
| Wed, 14 Jan 2015 18:41:12 +0000 (14 13:41 -0500) | ||
| tree | 6a427d7359510bd8a92b0c9c0a0da441e8526712 | treesnapshot (tar.gz zip) |
| parent | f9d204ecd1c85654f5b7fd07d16408862eedb3e6 | commitdiff |
Fix INVALID-ARG-COUNT-ERROR on x86-64 with ud2-breakpoints (Darwin).
To my surprise, the "wtf?" branch of INTERNAL-ERROR was taken
because INTERNAL-ERROR-ARGS was made to return one hardcoded location
- dubiously named arg-count-sc instead of arg-count-sc-offset -
only when the trap kind was invalid-arg-count. But VERIFY-ARG-COUNT
didn't emit the new trap kind for UD2, nor use the older technique of
emitting RCX's sc-offset into the instruction stream after the break.
To my surprise, the "wtf?" branch of INTERNAL-ERROR was taken
because INTERNAL-ERROR-ARGS was made to return one hardcoded location
- dubiously named arg-count-sc instead of arg-count-sc-offset -
only when the trap kind was invalid-arg-count. But VERIFY-ARG-COUNT
didn't emit the new trap kind for UD2, nor use the older technique of
emitting RCX's sc-offset into the instruction stream after the break.
| src/compiler/x86-64/call.lisp | diffblobblamehistory | |
| src/compiler/x86-64/macros.lisp | diffblobblamehistory | |
| tests/compiler.pure.lisp | diffblobblamehistory |