Closed userrand closed 1 year ago
Hi @userrand,
This is true. I am planning to commit a huge update this week, which will bring the most important features including sort of Iconise[]
. It is automatically called, when the output string is too long (>1000 symbols) and iconizes the object.
Regarding the installation issues, well, the current version of the project was more like a concept proof, now it is getting more serious. 127.0.0.01:8090
means it runs a local web-server at your localhost and 8090
is a default port chosen in a way not to conflict with some other applications running on the default 80s port. I general the installlation steps will look like
git clone https://github.com/JerryI/wolfram-js-frontend
wolframscript start.wls
that's all. Everything else is for developers.
might try this in the future. I am currently learning Javascript and I might try to understand the code and add features in the future. Is the current framework limited to 3D graphics via threejs or is it possible to include other features related to Dynamic?
Back at the time I was fed up with glitchy Mathematica's Frontend and wanted more control over it and flexibillity. To check the limitations of the Graphics3D
, one can check this page. Sure you can display any Mathematica's object as Plot[x, {x,0,1}]//SVGForm
, however, there is a lack of interaction. Therefore, to bring it back one had to write an interpreter in JS of Wolfram Language expressions, which would translate the commands and recreate objects using three.js
. It was done already will be developed further. There is no limitations from three.js
side, it just requires some time and efforts ;) The same can be done with all Graphics[]
objects as well, using, for instance d3.js
. I personally recreated only ListPlot
and LinePlot
using ploty.js
, which was quite easy.
All interactive objects are wrapped as
FrontEndExecutable["some-id to the association which holds the data"]
which actually means that the content will be executed by the simplified JS interpreter of Wolfram Language. The content of the function can determine if the results will be displayed as HTML block or not. There are a lot of flexibility on that.
The whole binding between JS (or DOM in general) is easily scalable. Since any Wolfram symbol/function ca be expressed as
core.NameThisFunction = function(args, env) {
console.log(interpretate(args[0]))
}
and we can call it from Wolfram Engine as
WebSocketPublish[server, NameThisFunction["Hello World!]]
And that's all! No exports
or some tricky bindings/bridging.
In principle it opens up a new branch of mixing two technologies. In the same manner the whole Graphics3D
. One can go even further and embed any DOM object in this fashion
HTMLForm["<script>alert('Hi!')</script>"]
or add some new definitions to the JS interpreter
HTMLForm["<script>core.MyFunction = function(args, env) { } </script>"]
For the JS side we can also call any Wolfram's functions
socket.send('Print["This has to popped up in your Wolfram console"];')
Moreover, since we are using Wolfram Script Pages (a hypertext post processor, similar to PHP by the syntax), it allows to make dynamic pages, generate HTML content and so on.
Taking everything into consideration it is possible to make a nice analog of Dynamic
and Manipulate
with even more features compared to the Mathematica's Frontend. I am planing to implement sliders and other simple stuff to show the potential of this combo.
I am using this stack of technologies in my research a lot and in sort of production, thus I would say, this has some good potential.
Nice, thank you for the detailed explanation. The update sounds like it is going to be quite cool (:
answer to multiple coding languages: copy pasted from https://github.com/JerryI/wolfram-js-frontend/issues/5#issuecomment-1415258015
CodeMirror supports mixed and multicode syntax highlighting. However for the autocomplete (or autosuggestions) it might be tricky, there is a packet from Wolfram called Language Server, which provides Lang docs onfly via TCP, which can be redirected to websockets.
One has to be careful, since the front-end and the cell structure highly depend on Wolfram backend and designed for that. Then, any other kernels (js, python) have to be connected to the Wolfram Engine first via Link library
My reply on multiple coding languages :
One has to be careful, since the front-end and the cell structure highly depend on Wolfram backend and designed for that. Then, any other kernels (js, python) have to be connected to the Wolfram Engine first via Link library
I am not referring to direct communication between mathematica and other languages via library link. I meant that notebook cells belonging to different languages would be completely independent, no library link involved. On top of that the community can include translators if they want and add it to a separate repository but the default behavior would be independent kernels and independent languages just in the same notebook for the convenience of copy-pasting results.
Basically, I would like to know if there are parts in the code that are strongly hard coded that would not allow adding more web sockets to other languages independently of the presence of Mathematica.
It seems you can already do this with JSRun but I would like syntax highlighting and documentation. The same for python. I can look into this but I wanted to know if the present code has parts that prevent adding another language or that make it difficult
reply by owner on multiple coding languages copied from : https://github.com/JerryI/wolfram-js-frontend/issues/5#issuecomment-1415410110
I see. In principle, the master kernel, which serves the page, takes care about saving notebook and so on has no idea, what was evaluated. It feeds the string to so called xEvaluator with a callback on creating a new child cell, which can be a remote kernel, any kind of process in theory. In this sense, I would not suggest to get rid of Wolfram kernel fully, because someone would need to do all this supporting stuff. Also the code for the making cells depends on WSP extension for Wolfram Engine.
Expanding the frontend to any languages I can see the following problem pop up
Graphics3D and other possible implementations of interactive objects from Mathematica will have to be recreated in Python or any other languages. I follows the paradigm of Wolfram Language, that everything is a symbol, which is not a part of all programming languages. TableForm and others now is a part of frontend, of course, one can get rid of them and move to the sort of extension, but in the end the resulting frontend will be the simplest REPL machine with cells and no other fancy stuff.
Summering all of this, at this stage this project might be still too small and young to pretend to be a multitool, as you described. Bringing Dynamics might involve some architectural changes, but I will keep everyone up to date with it.
1.
The rest you listed in other opened issues, so, I would like to close this one. It is also quite outdated.
Nice !
List of feature requests
To avoid using multiple separate feature requests (I have more I just don't want to bother you with so many at once) I will regroup here some of the feature requests that were already answered.
Outline:
1. Would it be possible to implement Iconize with this framework?
Rest of the message (about installation, plans to understand the code and maybe contribute, question about Dynamic):
Hi, this project looks quite interesting. I have not tried it yet. What do the numbers in "open browser with http://127.0.0.1:8090" represent? Also, maybe consider including the command line for downloading a GitHub project for convenience.
I might try this in the future. I am currently learning Javascript and I might try to understand the code and add features in the future. Is the current framework limited to 3D graphics via threejs or is it possible to include other features related to Dynamic?
2. Multiple coding languages
Is the code currently built in such a way that one could develop a notebook that has different coding languages (python, wolfram language, javascript) in different cells ? Mathematica already does this but it does not have syntax highlighting and short documentation when hovering over functions in other languages.
This question has already been asked for jupyterlab here : https://github.com/jupyterlab/jupyterlab/issues/2815. The author here: https://github.com/jupyterlab/jupyterlab/issues/2815#issuecomment-349773751 claims to have made an implementation that does that but I did not test it. There is also another claim on the same thread here https://github.com/jupyterlab/jupyterlab/issues/2815#issuecomment-956313608.
This is an ambitious task I am just exploring the potential of the project at this moment.
3. In place evaluation
Is it possible to do in place evaluation in the notebook cell ? (for example in an expression 4+2+6, select 4+2, then use ctrl+shift+enter and the 4+2 evaluates to 6 without evaluating the entire cell).
Someone already tried doing this with code mirror 6 here: https://observablehq.com/@tophtucker/eval-in-place (or google in place evaluation codemirror). Why ?
I like in place evaluation as it allows checking the definition or full definition of a function by using Definition or FullDefinition with in place evaluation which allows me to check the entire function definition while remaining in the same notebook cell (rather than creating a new cell that I then have to delete or searching for the function in the notebook).
In place evaluation can also be used to track a variable x during the evaluation by using Dynamic and placing the variable in a comment like ( Dynamic[x] ) then select Dynamic[x] and ctrl+shift+ enter then evaluate the notebook cell to see x changing inside the comment. Example:
( Dynamic[x] ) x = 0; Do[x++; Pause[1], 4]
4 Community Plugins/ Packages
https://github.com/JerryI/wolfram-js-frontend/issues/21