Print task backtraces on sigquit (#59389)

Author: gbaraldi

src/julia_internal.h

@@ -1527,6 +1527,7 @@ JL_DLLEXPORT jl_value_t *jl_backtrace_from_here(int returnsp, int skip);
 void jl_fprint_critical_error(ios_t *t, int sig, int si_code, bt_context_t *context, jl_task_t *ct);
 JL_DLLEXPORT void jl_raise_debugger(void) JL_NOTSAFEPOINT;
 JL_DLLEXPORT void jl_gdblookup(void* ip) JL_NOTSAFEPOINT;
+JL_DLLEXPORT void jl_print_task_backtraces(int show_done) JL_NOTSAFEPOINT;
 void jl_fprint_native_codeloc(ios_t *s, uintptr_t ip) JL_NOTSAFEPOINT;
 void jl_fprint_bt_entry_codeloc(ios_t *s, jl_bt_element_t *bt_data) JL_NOTSAFEPOINT;
 #ifdef _OS_WINDOWS_

src/signal-handling.c

@@ -641,6 +641,9 @@ void jl_fprint_critical_error(ios_t *s, int sig, int si_code, bt_context_t *cont
             jl_safe_fprintf(s, "\n[%d] signal %d (%d): %s\n", getpid(), sig, si_code, strsignal(sig));
         else
             jl_safe_fprintf(s, "\n[%d] signal %d: %s\n", getpid(), sig, strsignal(sig));
+        if (sig == SIGQUIT) {
+            jl_print_task_backtraces(0);
+        }
     }
     jl_safe_fprintf(s, "in expression starting at %s:%d\n", jl_atomic_load_relaxed(&jl_filename), jl_atomic_load_relaxed(&jl_lineno));
     if (context && ct) {

src/stackwalk.c

@@ -1397,11 +1397,19 @@ JL_DLLEXPORT void jl_print_backtrace(void) JL_NOTSAFEPOINT
 // Print backtrace for specified task to `s`
 JL_DLLEXPORT void jl_fprint_backtracet(ios_t *s, jl_task_t *t) JL_NOTSAFEPOINT
 {
-    jl_task_t *ct = jl_current_task;
-    jl_ptls_t ptls = ct->ptls;
-    ptls->bt_size = 0;
-    jl_bt_element_t *bt_data = ptls->bt_data;
-    jl_record_backtrace_result_t r = jl_record_backtrace(t, bt_data, JL_MAX_BT_SIZE, 0);
+    jl_bt_element_t *bt_data;
+    jl_task_t *ct = jl_get_current_task();
+    size_t max_bt_size;
+    if (ct && ct->ptls != NULL) {
+        jl_ptls_t ptls = ct->ptls;
+        ptls->bt_size = 0;
+        bt_data = ptls->bt_data;
+        max_bt_size = JL_MAX_BT_SIZE;
+    } else {
+        max_bt_size = 1024; //8kb of stack should be safe
+        bt_data = (jl_bt_element_t *)alloca(max_bt_size * sizeof(jl_bt_element_t));
+    }
+    jl_record_backtrace_result_t r = jl_record_backtrace(t, bt_data, max_bt_size, 0);
     size_t bt_size = r.bt_size;
     size_t i;
     for (i = 0; i < bt_size; i += jl_bt_entry_size(bt_data + i)) {

test/cmdlineargs.jl

@@ -74,6 +74,32 @@ let
     @test format_filename("%a%%b") == "a%b"
 end
 
+if Sys.isunix()
+    @testset "SIGQUIT prints task backtraces" begin
+        script = """
+            mutable struct RLimit
+                cur::Int64
+                max::Int64
+            end
+            const RLIMIT_CORE = 4 # from /usr/include/sys/resource.h
+            ccall(:setrlimit, Cint, (Cint, Ref{RLimit}), RLIMIT_CORE, Ref(RLimit(0, 0)))
+            write(stdout, "r")
+            wait()
+        """
+        exename = `$(Base.julia_cmd()) --startup-file=no --color=no`
+        errp = PipeBuffer()
+        # disable coredumps for this process
+        p = open(pipeline(`$exename -e $script`, stderr=errp), "r")
+        @test read(p, UInt8) == UInt8('r')
+        Base.kill(p, Base.SIGQUIT)
+        wait(p)
+        err_s = readchomp(errp)
+        @test Base.process_signaled(p) && p.termsignal == Base.SIGQUIT
+        @test occursin("==== Thread ", err_s)
+        @test occursin("==== Done", err_s)
+    end
+end
+
 @testset "julia_cmd" begin
     julia_basic = Base.julia_cmd()
     function get_julia_cmd(arg)