Polymer 2.4: Paving the way for 3.0 and TypeScript support(polymer-project.org)
polymer-project.org
Polymer 2.4: Paving the way for 3.0 and TypeScript support
https://www.polymer-project.org/blog/2018-01-25-polymer-2-4.html
6 comments
It may be true that JSON inherently doesn't have types, however unless your backend generates random JSON for your front end to consume, specifying types for the JSON that various endpoints return is extremely helpful.
In my experience giving your JSON API type information is one of the most beneficial things you can do to your web codebase.
To some extent, but there is a difference between giving each object a type property (which is open to interpretation by the client app) vs forcing the client to adhere to rigid type definitions/constraints (which make it more difficult for the client to actually use the API). Developers who have used SOAP before will remember what that's like.
Typescript is as rigid as you want it to be, you're free to loosen the types as your api demands. It really is nowhere near as heavyweight as SOAP.
The client has to adhere to the shape of the JSON that the API is generating anyway. Defining those types just makes it easier to avoid mistakes. There's no reason that need to be any more rigid than the API would otherwise be.
But the schema can change over time with different API versions. So you need different schemas on the front end for different versions of the API? It just seems like a lot of hassle when ultimately all you need to do is get a string property from the API and put it in a textbox.
The simple thing would be to avoid making incompatible changes to the public API. If you keep the same schema and loosen the types a bit (accepting more kinds of requests) then there's no need to have more than one schema.
But you're describing a fairly complicated scenario where you have multiple public API's with different request types, internal data (hopefully canonicalized at the border) and response types.
How do you keep track of all that data conversion without types?
But you're describing a fairly complicated scenario where you have multiple public API's with different request types, internal data (hopefully canonicalized at the border) and response types.
How do you keep track of all that data conversion without types?
If the schema is changing you'll need to document it and change the client code anyway. If you're able to auto-generate a schema then that's a hell of a lot easier.
Microsoft addressed this and says Typescript is most beneficial to manage large code bases
I've worked with TypeScript on large code bases. I found that the compile step becomes really slow and it slows down the debug cycle.
On my last large project, I had to wait 10 to 15 seconds for the code to build every time I saved the file (and yes the compilation step was optimized to only include relevant code; a large number of dependencies don't help).
So every time I made a change to the code to check something, I had to wait... You can't use console.log() to quickly check stuff anymore; the compile overhead is so significant that you're forced to pull up the debugger and manually step through the relevant code every single time you want to check something (you want to get as much info as possible from each debug cycle because it's so slow).
When writing plain JavaScript, I only pull up the debugger when it's a particularly difficult bug (e.g. a race condition). For most bugs, I can get all the information I need with just a couple of calls to console.log() in my code. The debugger is just unwieldy.
On my last large project, I had to wait 10 to 15 seconds for the code to build every time I saved the file (and yes the compilation step was optimized to only include relevant code; a large number of dependencies don't help).
So every time I made a change to the code to check something, I had to wait... You can't use console.log() to quickly check stuff anymore; the compile overhead is so significant that you're forced to pull up the debugger and manually step through the relevant code every single time you want to check something (you want to get as much info as possible from each debug cycle because it's so slow).
When writing plain JavaScript, I only pull up the debugger when it's a particularly difficult bug (e.g. a race condition). For most bugs, I can get all the information I need with just a couple of calls to console.log() in my code. The debugger is just unwieldy.
I used to be a c++ dev so a 15 sec compile time doesn't seem that bad. But what if you're compiler told you what was wrong faster than your console.log() could help you find it. I think is a trade off a lot of people want to make.
I agree, that makes sense for C/C++ because working with pointers and dynamic memory is a lot more complicated - Types make sense for C/C++ because the focus is execution speed and flexibility (not development productivity).
The case you describe very rarely happens to me with JavaScript; having fewer primitives to work with greatly reduces these kinds of issues... Sure it opens up the possibility that you'll do something stupid like trying to add an object with a string; but that's a pretty silly thing to do. If you name your variables properly then that never happens.
With C/C++, there are many primitives so you might get errors when you try to assign an int64 to an int8 (for example)... With JS, however, because there are fewer primitives so the odds of that happening are much much smaller.
Because there are so few primitives in JS, it's easy to remember how they all interact with each other. The only exception is the boolean truth table for '==' (which is a mess); that's why it's usually recommended to use '===' instead.
The case you describe very rarely happens to me with JavaScript; having fewer primitives to work with greatly reduces these kinds of issues... Sure it opens up the possibility that you'll do something stupid like trying to add an object with a string; but that's a pretty silly thing to do. If you name your variables properly then that never happens.
With C/C++, there are many primitives so you might get errors when you try to assign an int64 to an int8 (for example)... With JS, however, because there are fewer primitives so the odds of that happening are much much smaller.
Because there are so few primitives in JS, it's easy to remember how they all interact with each other. The only exception is the boolean truth table for '==' (which is a mess); that's why it's usually recommended to use '===' instead.
This. It's also simple to generate JSON Schema from TypeScript types (and vice versa).
> It makes no sense at all whatsoever that the front end logic should be more strict than the API!
It does to people who appreciate type information. To each his own. I can see the appeal of dynamic typing but I am pretty sure obvious errors (obvious if the compiler has type info) go uncaught.
It does to people who appreciate type information. To each his own. I can see the appeal of dynamic typing but I am pretty sure obvious errors (obvious if the compiler has type info) go uncaught.
Polymer will not require you to use TS - it just provides you with a way to use it if you want.
I wish they actually focused on 3.0 more. This "auto generated from 2.x" thing is pretty bad — paper elements that depended on neon-animation just don't work in 3.0:
https://github.com/PolymerElements/paper-dialog/issues/156
https://github.com/PolymerElements/paper-dialog/issues/156
It will get fixed - modulizer is still being worked on, and some elements are not being converted correctly as far as I can tell.
Can anyone reccomend a good state management scheme for building large scale Polymer apps? My team has gone this route of building small custom elements for everything, but getting them to communicate has been a pain. AngularJS solved that really well I think with $broadcast and $emit, but we'd rather stay away from monolithic frameworks.
Well you can do `polymer-redux`, `polymer-uniflow` or mixins (https://www.captaincodeman.com/2017/07/06/managing-state-in-...).
I went with redux.
You could I guess technically you could implement $broadcast - but that would be slow same as with angular, as for $emit you can just emit standard events.
I went with redux.
You could I guess technically you could implement $broadcast - but that would be slow same as with angular, as for $emit you can just emit standard events.
This could also be of benefit to your research: https://gist.github.com/kevinpschaaf/995c9d1fd0f58fe021b174c...
I've already upgraded some of my extensions to the tagged template literal syntax. It's so much nicer than what I was working with before, and is tiding me over for IE11 support in lit-html. Pretty big wins all around. Next, I'm gonna have to take a deeper look at those TS decorators!
I don't really work in Polymer (usually React) but just the same you might appreciate my favourite and most-abused decorator:
https://github.com/andreypopp/autobind-decorator
https://github.com/andreypopp/autobind-decorator
That's a great one. It's a big part of my TC39 reading: https://github.com/tc39/proposal-decorators#decorators
Very exciting.
Very exciting.
I like the idea of writing Web Components with TypeScript. Are Web Components (or the frameworks pushing it forward) good enough for production use? Any issues worth considering before diving in?
I've used web components in production in two different companies and in a number of personal projects, so I'd say so. The move to official TS support is brand new, so you might catch some turbulence in that area. Polymer generally is about to make the move from 2.x to 3.x, so there could be some issues there. However, you can work with 3.x code now, and the official release should be in the next month or two, so getting started now might be just the right time.
Thanks for the advice.
This is great news for Polymer - I can't wait to try out 3.0 when it's ready :) Typescript is indeed a favorable target for their existing audience.
TypeScript is not suitable for building the vast majority of web UIs; it just makes things way more complicated (with the added build step, source-mapping issues, missing type definitions, outdated type definitions...) and adds no value beyond improved code completion and compiler warnings (and those two are conveniences at best).
Web applications are centred around APIs. APIs use JSON... JSON doesn't have types; so already there is an obvious impedance mismatch!
It makes no sense at all whatsoever that the front end logic should be more strict than the API!
As far as the front end is concerned, the API is the one source of truth; so unless you're willing to go back to the old days of SOAP and also start adding type annotations to your transport protocol then it makes no sense whatsoever to use TypeScript.
The main exception I can think of are web-based games; in this case it's possible that most of the complexity is actually on the front end itself (and not in the data integration aspect).