fix quickfix list when test duration exceeds timeout

Do not treat a panic caused by a test timeout as an error that occurred
within a test's callstack. When go test panics due to a test timeout,
the running goroutine's stacktrace only contains entries for the
standard library, and there is one goroutine whose stacktrace contains
entries for the standard library and the test binary. None of the other
goroutines should be considered to contain an error, either, because the
panic was not due to any code executing within those goroutines' stacks.
Therefore, treat panics caused by a test timeout as an error without a
file location.

autoload/go/test-fixtures/test/src/timeout/timeout_test.go

@@ -0,0 +1,47 @@
+// Run a few parallel tests, all in parallel, using multiple techniques for
+// causing the test to take a while so that the stacktraces resulting from a
+// test timeout will contain several goroutines to avoid giving a false sense
+// of confidence or creating error formats that don't account for the more
+// complex scenarios that can occur with timeouts.
+
+package main
+
+import (
+	"testing"
+	"time"
+)
+
+func TestSleep(t *testing.T) {
+	t.Parallel()
+	time.Sleep(15 * time.Second)
+	t.Log("expected panic if run with timeout < 15s")
+}
+
+func TestRunning(t *testing.T) {
+	t.Parallel()
+	c := time.After(15 * time.Second)
+Loop:
+	for {
+		select {
+		case <-c:
+			break Loop
+		default:
+		}
+	}
+
+	t.Log("expected panic if run with timeout < 15s")
+}
+
+func TestRunningAlso(t *testing.T) {
+	t.Parallel()
+	c := time.After(15 * time.Second)
+Loop:
+	for {
+		select {
+		case <-c:
+			break Loop
+		default:
+		}
+	}
+	t.Log("expected panic if run with timeout < 15s")
+}

autoload/go/test.vim

@@ -318,6 +318,10 @@ function! s:errorformat() abort
 
   " set the format for panics.
 
+  " handle panics from test timeouts
+  let format .= ",%+Gpanic: test timed out after %.%\\+"
+
+  " handle non-timeout panics
   " In addition to 'panic', check for 'fatal error' to support older versions
   " of Go that used 'fatal error'.
   "
@@ -368,6 +372,11 @@ function! s:errorformat() abort
   " library line in stack traces.
   let format .= ",%-G" . readyaddress
 
+  " a blank line is the end of a multi-line message. This is only relevant
+  " when tests panic; the goroutine stacktrace sections are separated by an
+  " empty line.
+  let format .= ",%Z%^%$"
+
   " Match and ignore exit status lines (produced when go test panics) whether
   " part of a multi-line message or not, because these lines sometimes come
   " before and sometimes after panic stacktraces.

autoload/go/test_test.vim

@@ -47,6 +47,16 @@ func! Test_GoTestCompilerError() abort
   call s:test('compilerror/compilerror_test.go', expected)
 endfunc
 
+func! Test_GoTestTimeout() abort
+  let expected = [
+        \ {'lnum': 0, 'bufnr': 0, 'col': 0, 'valid': 1, 'vcol': 0, 'nr': -1, 'type': '', 'pattern': '', 'text': 'panic: test timed out after 2s'}
+      \ ]
+
+  let g:go_test_timeout="2s"
+  call s:test('timeout/timeout_test.go', expected)
+  unlet g:go_test_timeout
+endfunc
+
 func! s:test(file, expected, ...) abort
   if has('nvim')
     " nvim mostly shows test errors correctly, but the the expected errors are