[Lldb-commits] [lldb] Reapply "[lldb] Implement basic support for reverse-continue (#125242)" (again) (PR #128156)

Tue Mar 18 00:11:58 PDT 2025

rocallahan wrote:

These are the test log packets between the reverse-exec proxy to the debugserver for the forward singlestep operation:
```
2025-03-17 15:49:10,842 INFO     Sending packet vCont;s:4870aa
2025-03-17 15:49:10,843 INFO     Received reply T05thread:4870aa;threads:4870aa;thread-pcs:100e0ff94;00:0100000000000000;01:1037ff6e01000000;02:2037ff6e01000000;03:b037ff6e01000000;04:74444e8101000000;05:d87b80db01000000;06:636570745f746162;07:a006000000000000;08:0200000000000000;09:b834ff6e01000000;0a:000000003e196aff;0b:c002000000000000;0c:0080000000000000;0d:0000000000000010;0e:0400000000000000;0f:0080000000000000;10:e034ff6e01000000;11:e034ff6e01000000;12:0000000000000000;13:401c220101000000;14:34ffe00001000000;15:e034ff6e01000000;16:1019220101000000;17:6035ff6e01000000;18:a035ff6e01000000;19:bb85538101000000;1a:0000000000000000;1b:0000000000000000;1c:0000000000000000;1d:c034ff6e01000000;1e:68ffe00001000000;1f:b034ff6e01000000;20:94ffe00001000000;21:00100080;a1:0000000000000000;a2:220000cb;a3:00000000;metype:6;mecount:2;medata:1;medata:0;memory:0x16eff34c0=f036ff6e01000000e0904b8101000000;memory:0x16eff36f0=000000000000000000000000008045ff;
```
That reports metype:6, i.e. EXC_BREAKPOINT. I'm not sure if that is the right code for a watchpoint firing on this system. Whether it is or not, LLDB is definitely interpreting this as "instruction step into" and is not reporting to the user that a watchpoint fired. The proxy code should not be modifying this T packet so it would be surprising if the reverse-exec test code is causing this problem.
So I wonder whether, in normal operation on the test system, singlestepping an instruction which triggers a watchpoint actually reports the watchpoint being hit.

https://github.com/llvm/llvm-project/pull/128156