[Lldb-commits] [lldb] [lldb] gdb rsp file error pass fix (PR #106950)

[Jason Molenda via lldb-commits]

================
@@ -3064,22 +3064,41 @@ static int gdb_errno_to_system(int err) {
 
 static uint64_t ParseHostIOPacketResponse(StringExtractorGDBRemote &response,
                                           uint64_t fail_result, Status &error) {
+  // The packet is expected to have the following format:
+  // 'F<retcode>,<errno>'
+
   response.SetFilePos(0);
   if (response.GetChar() != 'F')
     return fail_result;
+
   int32_t result = response.GetS32(-2, 16);
   if (result == -2)
     return fail_result;
-  if (response.GetChar() == ',') {
-    int result_errno = gdb_errno_to_system(response.GetS32(-1, 16));
-    if (result_errno != -1)
-      error = Status(result_errno, eErrorTypePOSIX);
-    else
-      error = Status(-1, eErrorTypeGeneric);
-  } else
+
+  if (response.GetChar() != ',') {
     error.Clear();
+    return result;
+  }
+
+  // Response packet should contain a error code at the end. This code
+  // corresponds either to the gdb IOFile error code, or to the posix errno.
----------------
jasonmolenda wrote:

One background info: `lldb/docs/lldb-platform-packets.txt` was written by me as I was reading the lldb/debugserver/lldb-server implementations of these packets, and trying to cross-reference with the official gdb remote serial protocol documentation.  Any disagreement between lldb-platform-packets.txt and the official docs should be assumed to be a mistake on my part, probably reflecting how lldb was currently handling things.  As you say, if we wanted to relay a new error type, we'd need to get gdb to add that to the "errno" table in their docs.  Otherwise the only safe error type we can use is EUNKNOWN.

https://github.com/llvm/llvm-project/pull/106950