Save corefile on assert failures [WIP]

.github/workflows/testing.yml

@@ -26,4 +26,7 @@ jobs:
       - run: sudo apt install -y cmake libssl-dev libcurl4-openssl-dev ${{ matrix.aptpkg }}
       # Building
       - run: ./run-make-from-ci.sh --compiler "${{ matrix.compiler }}"
+      # Test prerequisites
+      - run: sudo apt install -y gdb
+      # Testing
       - run: ./run-tests-from-ci.sh

.github/workflows/testing_non_sse.yml

@@ -26,4 +26,7 @@ jobs:
       - run: sudo apt install -y cmake libssl-dev libcurl4-openssl-dev ${{ matrix.aptpkg }}
       # Building
       - run: ./run-make-from-ci.sh --compiler "${{ matrix.compiler }}" --extra-config "FMQ_NO_SSE=1"
+      # Test prerequisites
+      - run: sudo apt install -y gdb
+      # Testing
       - run: ./run-tests-from-ci.sh

FlashMQTests/fmq-generate-core-file.py

@@ -0,0 +1,35 @@
+import time
+import os
+
+
+class FMQGenerateCoreFile(gdb.Command):
+    """Generate a corefile with a dynamic name instead of core.PID"""
+
+    def __init__(self):
+        super().__init__("fmq-generate-core-file", gdb.COMMAND_USER)
+
+    def invoke(self, arg, from_tty):
+        # Are in in[MainTests::testAsserts]? If yes don't do anything (this is testing the framework on bootup)
+        frame = gdb.selected_frame()
+        while frame:
+            # The name is a bit compiler dependent, I've seen:
+            # MainTests::testAsserts
+            # MainTests::testAsserts()
+            frame_name = frame.name() or ''
+            if frame_name.startswith("MainTests::testAsserts"):
+                return False
+            frame = frame.older()
+
+        # This is something else
+        try:
+            runid = f".{os.environ['FMQ_RUN_ID']}."
+        except KeyError:
+            runid = "."
+        gdb.execute(
+            f"generate-core-file core{runid}{time.strftime('%Y-%d-%mT%H%M%S')}.{time.thread_time_ns()}"
+        )
+        issues = gdb.convenience_variable('issues') or 0
+        gdb.set_convenience_variable('issues', issues + 1)
+
+
+FMQGenerateCoreFile()  # this registers the CLI command

FlashMQTests/main.cpp

@@ -1,3 +1,4 @@
+#include <csignal>
 #include <iostream>
 #include <vector>
 
@@ -9,8 +10,15 @@ void printHelp(const std::string &arg0)
     std::cout << "Usage: " << arg0 << " [ --skip-tests-with-internet ] [ --skip-server-tests ] " << " <tests> " << std::endl;
 }
 
+void signal_noop(int signal)
+{
+    std::cout << "Signal NoOP (if you used save-core-on-sigvtalrm.gdb a core dump will now be created)" << std::endl;
+}
+
 int main(int argc, char *argv[])
 {
+    std::signal(SIGVTALRM, signal_noop);
+
     bool skip_tests_with_internet = false;
     bool skip_server_tests = false;
     std::vector<std::string> tests;

FlashMQTests/run-tests-from-ci.sh

@@ -2,13 +2,27 @@
 
 STDERR_LOG=$(mktemp)
 
+
+export FMQ_RUN_ID=$EPOCHSECONDS  # this is only granular to the second but that should be fine
+
 cd buildtests || exit 1
 
-if ./flashmq-tests 2> "$STDERR_LOG" ; then
+if gdb --batch -x ../fmq-generate-core-file.py -x ../save-core-on-sigvtalrm.gdb --args ./flashmq-tests 2> "$STDERR_LOG" ; then
   echo -e '\033[01;32mSUCCESS!\033[00m'
 else
   echo -e '\033[01;31mBummer\033[00m'
   echo -e "\n\nTail of stderr:\n\n"
   tail -n 200 "$STDERR_LOG"
+
+  for i in "core.${FMQ_RUN_ID}."*
+  do
+	  [[ "$i" == "core.${FMQ_RUN_ID}.*" ]] && break
+	  echo ""
+	  echo ",#############################################,"
+    echo "| $i |"
+    echo "'#############################################'"
+    gdb --batch --ex "bt -full" "./flashmq-tests" "$i"
+  done
+
   exit 1
 fi

FlashMQTests/save-core-on-sigvtalrm.gdb

@@ -0,0 +1,16 @@
+# Less verbose gdb please
+set print thread-events off
+set print inferior-events off
+
+# fmq-generate-core-file will increase $issues when it dumps
+set $issues = 0
+
+catch signal SIGVTALRM
+commands
+  fmq-generate-core-file
+  continue
+end
+
+run
+
+exit $issues

FlashMQTests/testhelpers.cpp

@@ -1,6 +1,7 @@
 #include "testhelpers.h"
 
 #include <iostream>
+#include <csignal>
 #include <cstring>
 
 int assert_count;
@@ -15,6 +16,8 @@ bool fmq_assert(bool b, const char *failmsg, const char *actual, const char *exp
     {
         assert_fail_count++;
 
+        raise(SIGVTALRM); // to create a corefile using GDB (if attached)
+
         if (asserts_print)
         {
             // There are two types of failmsg: unformatted ones and formatted ones.