Low cost stacktrace walking

What is the problem this feature will solve?

Public APIs that allow for stack collection/walking are quite heavy, doing memory allocation and symbol resolution in the background.
All the necessary features are already available - but exported as local symbols, making theme very hard to use even from native code.

What is the feature you are proposing to solve the problem?

Below is example code, that used to work fine in node 20 and 22, but will not work on 24+ because the necessary symbols are not global anymore (so in a way it is regression) . I do not mind that the private API are not stable and require recompilation specific for v8. I do not need new public API - it would be nice of course, but I can easily maintain multiple version of my code. Having symbols local though, makes the project extremely complex:

struct FrameId {
    uint32_t script_id;
    uint32_t start_position;
    uint32_t end_position;
    uint32_t filler;
    v8::internal::Tagged<v8::internal::SharedFunctionInfo> sfi;
};

size_t get_stack_frame_ids(v8::Isolate* isolate, FrameId* buffer, size_t max_len) {
    v8::internal::JavaScriptStackFrameIterator it(reinterpret_cast<v8::internal::Isolate*>(isolate));
    size_t count = 0;
    while (!it.done() && count < max_len) {
        v8::internal::JavaScriptFrame* frame = it.frame();
        v8::internal::Tagged<v8::internal::JSFunction> function = frame->function();
        v8::internal::Tagged<v8::internal::SharedFunctionInfo> sfi = function->shared();
        v8::internal::Tagged<v8::internal::Script> script = v8::internal::Cast<v8::internal::Script>(sfi->script());

        buffer[count].script_id = script->id();
        buffer[count].start_position = sfi->StartPosition();
        buffer[count].end_position = sfi->EndPosition();
        buffer[count].sfi = sfi;

        count++;
        it.Advance();
    }
    return count;
}

size_t describe_frame(v8::Isolate* isolate, FrameId* frame_id, char* buffer, size_t buffer_size) {
    // Fetch function name and script name
    std::unique_ptr<char[]> name = frame_id->sfi->DebugNameCStr();
    size_t offset = 0;
    size_t i=0;
    while(offset < buffer_size - 3 && name[i] != '\0') {
        buffer[offset++] = name[i++];
    }
    buffer[offset++] = '(';
    v8::internal::Tagged<v8::internal::Script> script = v8::internal::Cast<v8::internal::Script>(frame_id->sfi->script());
    v8::internal::Tagged<v8::internal::Object> raw_name = script->GetNameOrSourceURL();
    if (IsString(raw_name)) {
        name = v8::internal::Cast<v8::internal::String>(raw_name)->ToCString();
        i=0;
        while(offset < buffer_size - 2 && name[i] != '\0') {
            buffer[offset++] = name[i++];
        }
    } else {
        if(offset < buffer_size - 11) {
            memcpy(buffer+offset, "<unknown>", 9);
            offset += 9;
        }
    }
    buffer[offset++] = ':';

    // Format the final description into the buffer
    offset += snprintf(buffer+offset, buffer_size - offset, "%d:%d)", frame_id->start_position, frame_id->end_position);

    // If buffer was too small, null-terminate it safely
    if (offset >= buffer_size) {
        buffer[buffer_size - 1] = '\0'; // Null-terminate the buffer
        offset = buffer_size - 1; // Set offset to the last position
    }

    return offset;
}

What alternatives have you considered?

There do not seem to be alternative for "allocation free" stack walking.

Commencez par suivre les points d’entrée internes de V8 montrés dans l’exemple, en particulier JavaScriptStackFrameIterator, JavaScriptFrame et SharedFunctionInfo, et comparez la visibilité de leurs symboles entre Node 22 et Node 24. Identifiez ensuite la configuration de build ou d’exportation qui contrôle ces symboles. Le travail est terminé lorsqu’un parcours natif de la pile sans allocation reste utilisable sans nécessiter une nouvelle API publique.