This is a work in progress rewrite of the web framework the ideas of which I am using in multiple non-opensource personal projects.
I started working on the ideas for this a couple of years ago, when the choice of frameworks was pretty scarce, and also the ones that were there, were requiring nightly rust.
This project has a few general principles behind:
- Allow for purely server-based logic.
- No unsafe {} blocks.
- Only stable Rust.
- Completely stateless on the server side.
- Clear separation between Access, Logic and Presentation layers.
- Allow to customize each layer, but give sane defaults.
DISCLAIMER: this is work in progress and interfaces are subject to change.
Before running the examples, make sure your setup has the necessary prerequisites, on Ubuntu 18.04 it is:
sudo apt-get install build-essential libsqlite3-dev libpq-dev
If you do not have these installed, the compilation will fail. When you have them installed, clone this repository and then run the example:
ubuntu@host:~/rsp10$ cargo run --example simple
Finished dev [unoptimized + debuginfo] target(s) in 0.07s
Running `target/debug/examples/simple`
HTTP server for Simple Example starting on 127.0.0.1:4480
then connect the browser to http://127.0.0.1:4480/ - or, if you are running the example on a different machine, bind to all addresses:
ubuntu@host:~/rsp10$ BIND_IP=0.0.0.0 cargo run --example simple
Finished dev [unoptimized + debuginfo] target(s) in 0.08s
Running `target/debug/examples/simple`
HTTP server for Simple Example starting on 0.0.0.0:4480
You will be prompted to login (user "user" and "pass") and then you will see the example 'interactive' page which has a few input elements and allows to get the idea of what this is all about.
Did it work ? Interested to know how ? Here's some more to it....
The basic idea is that each web page can be represented by three data elements:
- State Key: This is a (maybe optional) set of arguments passed via the query string that define the initial state for the page when it is being loaded. A very simple key is an Option being the optional ID of an entity to edit.
- Initial State: This is a record containing the state of the page, which is initially retrieved based on State Key before being sent to renderer.
- State: This is the current state, which may or may not be different from the Initial State, the difference is expected to be due to the user changing it by typing in the text elements, selecting different dropdowns, etc.
However, what happens if someone modifies the data in question in the background ? For this reason we need a fourth data field: Current Initial State - this is the "Initial State", however, freshly recalculated before each pass of business logic.
In order to perform any business logic we also need a fifth component, and that is Event.
Kept together, these five elements allow to perform any business logic in a completely stateless manner - which is a very useful property. It allows less logic for load balancing, as well as allows to survive the service restarts and (potentially) upgrades.
-
The browser performs the GET request, optionally supplying parameters for the State Key.
-
The server converts the query string arguments into State Key. It uses two parts for it: the definition of the page key, and optionally the method for converting the querystring arguments into the key. The state key for a page can be defined as follows (in this example the key is a simple integer):
#[derive(Debug, Clone, Serialize, Deserialize, Default)]
pub struct KeyI32 {
id: Option<i32>,
}Note, that with the above definition, passing the "id=XXX" (where XXX is something that parses as i32) will automatically populate the state key, however it is possible to have the custom method as well, for example like this:
fn get_key(
auth: &MyPageAuth,
args: &HashMap<String, Vec<String>>,
maybe_state: &Option<PageState>,
) -> Option<KeyI32> {
Some(KeyI32 {
id: args.get("id").map_or(None, |x| x[0].parse::<i32>().ok()),
})
}- Initial request: The server uses the key to retrieve the Initial State, which again has two parts: a struct holding it and method that populates it. Here is a sample struct:
#[derive(Debug, Clone, Serialize, Deserialize, Default)]
pub struct PageState {
message: String,
dd_testing: i32,
txt_text_message: String,
cbTestCheck: bool,
ddMyDropdown: i32,
}The method that returns the Initial State, based on the State Key must be idempotent:
fn get_state(req: &mut Request, auth: &MyPageAuth, key: KeyI32) -> PageState {
println!("default state for PageState with key: {:?}", &key);
PageState {
dd_testing: -1,
txt_text_message: "test".to_string(),
ddMyDropdown: key.id.unwrap_or(-1),
cbTestCheck: true,
..Default::default()
}
}If we are processing an update, then the form data will contain the Initial State, as well as State, which will be filled in from the form data. The Current Initial State will be still freshly filled as above.
- Handle the event. The event handler is called on each page update, including the initial load. Based on the event, as well as the three state records, it may alter the State to its liking, as well as return Action to perform: e.g. render the page, redirect to a different URL, etc.
fn event_handler(ri: RspInfo<Self, KeyI32, MyPageAuth>) -> RspEventHandlerResult<Self, KeyI32> {
let mut action = rsp10::RspAction::Render;
let mut initial_state = ri.initial_state;
let mut state = ri.state;
if ri.event.event == "submit" {
state.message = "".to_string();
if !ri.state_none {
let ev = ri.event;
let tgt = &ev.target[..];
match tgt {
"_eq" => {
state.txt_text_message =
format!("Pressed eq when state is {}", state.dd_testing);
}
"_lt" => {
state.dd_testing = state.dd_testing - 1;
}
"_gt" => {
if state.dd_testing == -1 {
state.message = format!("Select a value from the right dropdown first");
} else {
state.dd_testing = state.dd_testing + 1;
}
}
_ => {}
}
}
}
RspEventHandlerResult {
initial_state,
state,
action,
}
}- Now the server can populate the data that will be used to render the template. If the page does not contain any interactive elements, then it is not necessary to define it, but since most of the pages actually do interact, you will define it, something along these lines:
fn fill_data(ri: RspInfo<Self, KeyI32, MyPageAuth>) -> RspFillDataResult<Self> {
let mut modified = false;
let mut ri = ri;
let mut gd = RspDataBuilder::new();
let real_key = ri.key.id.unwrap_or(-1);
println!("{:?}", &ri.state);
rsp10_button!(btnTest, "Test button" => gd);
rsp10_select!(dd_testing, dbh_get_dropdown(ri.state.dd_testing), ri => gd, modified);
rsp10_select!(ddMyDropdown, dbh_get_dropdown(real_key), ri => gd, modified);
rsp10_text!(txt_text_message, ri => gd, modified);
rsp10_check!(gd, cbTestCheck, ri, modified);
rsp10_data!(modified => gd);
Self::fill_data_result(ri, gd)
}This method effectively translates the higher-level abstractions of the state into more visual data for the template rendering - dropdowns, checkboxes, text elements, and just simple data.
You will notice most of the operations are hidden behind macros - this is to minimize the clutter, because behind the scenes the "state.SomeElement" value, which may be an i32, for example, is rendered into a "SomeElement" Rc<RefCell>, which can be modified within the fill_data() code, if the complex UI element interactions require it.
After finishing the preparation the RspDataBuilder object is passed to the Self::fill_data_result() along with the "RspInfo" structure (which contains a lot of interesting data about the request), which compiles the Mustache data builder and returns the RspFillDataResult, which is used to render the templates.
- Compile the Mustache template file. The file name is normally derived automatically, but you can override it on a per-page basis. Also - for simplicity of debugging the compile currently happens on each page load, but it is trivial to compile the templates once upon the start. The option to do so will may be some time in the future.
The typical template file will contain HTML forms, with the template looking as follows:
<form method="post">
{{#btnTest}} {{> html/submit}} {{/btnTest}}
{{#ddMyDropdown}} {{>html/select}} {{/ddMyDropdown}}
{{#dd_testing}} {{>html/select}} {{/dd_testing}}
{{#cbTestCheck}} {{> html/checkbox }} {{/cbTestCheck}}
{{#txt_text_message}} {{> html/text }} {{/txt_text_message}}
<input type="hidden" name="initial_state_json" value="{{initial_state_json}}">
<input type="hidden" name="state_json" value="{{state_json}}">
<input type="submit" name="submit_lt" value="<">
<input type="submit" name="submit_eq" value="=">
<input type="submit" name="submit_gt" value=">">
</form>
Notice the '{{#foo}} ... {{/foo}}' pairs, with {{> html/something}} inside. While it looks like some cool markup - this is simply Mustache syntax to "dive in" one level into the element. It allows to have a very uniform yet easily customizable look and feel for the elements. As you see, you can also code regular HTML with no template data whatsoever.
However, note the two fields "initial_state_json" and "state_json" - they are essential for the correct functioning, and carry the state information about the page.
-
The rendered page is sent to the user.
-
User performs some manipulations, the client side code potentially does something as well, and eventualy a changed data is being submitted. At this point the cycle repeats from the beginning.
You will notice that current implementation is completely Javascript-free: this is obviously not the final state of affairs, but one of the goals of this framework was graceful fallback, and javascript-free operation with the server side completely controlling the data flow.
In the future more client-side functionality will be added.
I have completely omitted discussing the question of access control, however a curious reader might have noticed the "MyPageAuth" type.
Authentication is implemented via a trait, which returns Result<AuthObject, String> - with the successful result being the auth object, and the error containing a string with the URL to redirect to. The simplest authentication is no authentication:
pub struct NoPageAuth {}
impl rsp10::RspUserAuth for NoPageAuth {
fn from_request(_req: &mut iron::Request) -> Result<NoPageAuth, String> {
Ok(NoPageAuth {})
}
}In case the authentication layer returns the error, the processing of the request stops and a redirect to the supplied login URL is issued. This way, once you specify the auth type in the resource, you do not have to worry about it - you simply get the auth object that you can query.