debugger: make attach to running process stop the process (again)

src/plugins/debugger/gdb/attachgdbadapter.cpp

@@ -78,10 +78,11 @@ void AttachGdbAdapter::setupInferior()
 void AttachGdbAdapter::runEngine()
 {
     QTC_ASSERT(state() == EngineRunRequested, qDebug() << state());
-    m_engine->notifyEngineRunAndInferiorStopOk();
-    m_engine->continueInferiorInternal();
     m_engine->showStatusMessage(tr("Attached to process %1.")
         .arg(m_engine->inferiorPid()));
+    m_engine->notifyEngineRunAndInferiorStopOk();
+    GdbMi data;
+    m_engine->handleStop0(data);
 }
 
 void AttachGdbAdapter::handleAttach(const GdbResponse &response)

src/plugins/debugger/gdb/gdbengine.cpp

@@ -1215,6 +1215,34 @@ void GdbEngine::handleStopResponse(const GdbMi &data)
         notifyInferiorStopOk();
     }
 
+    // FIXME: Replace the #ifdef by the "target" architecture.
+    // FIXME: Is this needed at all anymore?
+#ifdef Q_OS_LINUX
+    if (!m_entryPoint.isEmpty()) {
+        if (frame.findChild("addr").data() == m_entryPoint) {
+            // There are two expected reasons for getting here:
+            // 1) For some reason, attaching to a stopped process causes *two* SIGSTOPs
+            //    when trying to continue (kernel i386 2.6.24-23-ubuntu, gdb 6.8).
+            //    Interestingly enough, on MacOSX no signal is delivered at all.
+            // 2) The explicit tbreak at the entry point we set to query the PID.
+            //    Gdb <= 6.8 reports a frame but no reason, 6.8.50+ reports everything.
+            // The case of the user really setting a breakpoint at _start is simply
+            // unsupported.
+            if (!inferiorPid()) // For programs without -pthread under gdb <= 6.8.
+                postCommand("info proc", CB(handleInfoProc));
+            continueInferiorInternal();
+            return;
+        }
+        // We are past the initial stop(s). No need to waste time on further checks.
+        m_entryPoint.clear();
+    }
+#endif
+
+    handleStop0(data);
+}
+
+void GdbEngine::handleStop0(const GdbMi &data)
+{
 #if 0 // See http://vladimir_prus.blogspot.com/2007/12/debugger-stories-pending-breakpoints.html
     // Due to LD_PRELOADing the dumpers, these events can occur even before
     // reaching the entry point. So handle it before the entry point hacks below.
@@ -1244,27 +1272,8 @@ void GdbEngine::handleStopResponse(const GdbMi &data)
     }
 #endif
 
-    // FIXME: Replace the #ifdef by the "target" architecture
-#ifdef Q_OS_LINUX
-    if (!m_entryPoint.isEmpty()) {
-        if (frame.findChild("addr").data() == m_entryPoint) {
-            // There are two expected reasons for getting here:
-            // 1) For some reason, attaching to a stopped process causes *two* SIGSTOPs
-            //    when trying to continue (kernel i386 2.6.24-23-ubuntu, gdb 6.8).
-            //    Interestingly enough, on MacOSX no signal is delivered at all.
-            // 2) The explicit tbreak at the entry point we set to query the PID.
-            //    Gdb <= 6.8 reports a frame but no reason, 6.8.50+ reports everything.
-            // The case of the user really setting a breakpoint at _start is simply
-            // unsupported.
-            if (!inferiorPid()) // For programs without -pthread under gdb <= 6.8.
-                postCommand("info proc", CB(handleInfoProc));
-            continueInferiorInternal();
-            return;
-        }
-        // We are past the initial stop(s). No need to waste time on further checks.
-        m_entryPoint.clear();
-    }
-#endif
+    const GdbMi frame = data.findChild("frame");
+    const QByteArray reason = data.findChild("reason").data();
 
     // This was seen on XP after removing a breakpoint while running
     //  >945*stopped,reason="signal-received",signal-name="SIGTRAP",

src/plugins/debugger/gdb/gdbengine.h

@@ -278,6 +278,7 @@ private: ////////// Gdb Output, State & Capability Handling //////////
     void handleResponse(const QByteArray &buff);
     void handleStopResponse(const GdbMi &data);
     void handleResultRecord(GdbResponse *response);
+    void handleStop0(const GdbMi &data);
     void handleStop1(const GdbResponse &response);
     void handleStop1(const GdbMi &data);
     StackFrame parseStackFrame(const GdbMi &mi, int level);