Merge pull request #52 from monzo/stack-string-causal-chain

Add causal chain to the stack string

errors.go

@@ -158,13 +158,45 @@ func (p *Error) StackTrace() []uintptr {
 	return out
 }
 
-// StackString formats the stack as a beautiful string with newlines
+// StackString formats the stacks from the terror chain as a string. If we
+// encounter more than one terror in the chain with a stack frame, we'll print
+// each one, separated by three hyphens on their own line.
 func (p *Error) StackString() string {
-	stackStr := ""
-	for _, frame := range p.StackFrames {
-		stackStr = fmt.Sprintf("%s\n  %s:%d in %s", stackStr, frame.Filename, frame.Line, frame.Method)
+	// 32,000 seems like a reasonable limit for a stack trace. Otherwise, we risk
+	// overwhelming downstream systems.
+	return StackStringWithMaxSize(p, 32000)
+}
+
+func StackStringWithMaxSize(p *Error, sizeLimit int) string {
+	// if we run into this many causes, we've likely run into something absurd. Like
+	// a self causing error.
+	const maxCausalDepth = 1024
+	var buffer strings.Builder
+	terr := p
+	var causalDepth int
+outer:
+	for terr != nil {
+		if buffer.Len() != 0 && len(terr.StackFrames) > 0 {
+			fmt.Fprintf(&buffer, "\n---")
+		}
+		for _, frame := range terr.StackFrames {
+			// 10 seems like a reasonable estimate of how large the rest of the line would be.
+			estimatedLineLen := len(frame.Filename) + len(frame.Method) + 16
+			if estimatedLineLen+buffer.Len() > sizeLimit {
+				break outer
+			}
+			fmt.Fprintf(&buffer, "\n  %s:%d in %s", frame.Filename, frame.Line, frame.Method)
+		}
+
+		if tcause, ok := terr.cause.(*Error); ok && causalDepth < maxCausalDepth {
+			terr = tcause
+			causalDepth += 1
+		} else {
+			break outer
+		}
 	}
-	return stackStr
+
+	return buffer.String()
 }
 
 // VerboseString returns the error message, stack trace and params

errors_test.go

@@ -583,3 +583,36 @@ func TestSetIsUnexpected(t *testing.T) {
 	err.SetIsUnexpected(false)
 	assert.False(t, *err.IsUnexpected)
 }
+
+func failyFunction() error {
+	return InternalService("halp", "I'm in trouble", nil)
+}
+
+func TestStackStringChasesCausalChain(t *testing.T) {
+	err := Augment(failyFunction(), "something may be up", nil)
+	terr := err.(*Error)
+	ss := terr.StackString()
+	t.Log(ss)
+	assert.Contains(t, ss, "failyFunction")
+}
+
+func TestCircularErrorProducesFiniteOutputWithStackFrames(t *testing.T) {
+	orig := failyFunction()
+	err := Augment(orig, "something may be up", nil)
+	terr := err.(*Error)
+	terr.cause = terr
+	terr.StackFrames = orig.(*Error).StackFrames
+	ss := terr.StackString()
+
+	// The default field size limit used in elastic-slog. It's kind of arbitrary, but it'll do for now.
+	assert.Less(t, len(ss), 32000)
+	assert.GreaterOrEqual(t, len(ss), 32000-1000)
+}
+func TestCircularErrorProducesFiniteOutputWithoutStackFrames(t *testing.T) {
+	err := Augment(failyFunction(), "something may be up", nil)
+	terr := err.(*Error)
+	terr.cause = terr
+	ss := terr.StackString()
+	// There's no actual stack in the causal cycle, so we don't render anything here.
+	assert.Empty(t, ss)
+}