Kasper,

There's a way to run PlantUML to process in a pipe (stdin/out) which gives pretty good performance for rendering:

java -Djava.awt.headless=true -jar plantuml.jar -pipe -tsvg -charset UTF-8��

It takes a PlantUML source as input and generates the format of the file (e.g. SVG) on stdout. The VSCode extension called��markdown-preview-enhanced��runs it this way. Because of the pipe complexity (and the fact that your machine needs a JVM, GraphViz and a properly configured path, etc.), I never tried to use it this way in Pharo.��

The markdown-preview-enhanced extension for VSCode does a lot of things (too many?) but I think of pillar when I use it. It shows an HTML version in real-time as you type, which includes PlantUML diagrams. You can use pandoc to get to LaTeX/PDF as final output. I'm using it for my course notes because of this, but also because it does reveal output (useful for presenting), too. If I had a Mac, I would use pillar :) -- I struggled with using it in Windows 10 with MikTeX (which I use very well with LyX).

If my memory serves me, Pavel K. and/or Julien D. had done a visitor in Pharo that would generate PlantUML script from (moose?) (meta)models, but I can't remember the precise details. I didn't aim for the sexy visitor version when I did the Moose support because I wanted people to be able to choose easily what classes to include in the diagram (because there's usually some context for it to be useful). A list with classes inside a hierarchy (the way we did it in Moose) seemed to be adequate. I'm pretty sure you could make an intuitive browser (or reuse the current one) with filters to select the classes for the diagram.��

As for using UML to make diagrams, I also have toyed with the idea of making an editor on top of PlantUMLs SVG, such that you could change the source script intuitively. For example, if you don't want to see the 30+ messages a class supports, you can right click on the class in the SVG and choose a "hide" option. Sadly, the SVG that is generated by PlantUML is more for drawing (it was ported from the PNG code I think) and so things like a bounding box of a class or the IDs for classes don't exist. Classes are "drawn" as a rectangle for the title, a rectangle for the attributes, etc. However, those are all things that could be coded back into PlantUML (it's open source). SVG (in a modern browser at least) can be pretty powerful.��

These days, the only UML diagrams I "draw" are on whiteboards or an electronic sketch pad (Surface pro). Pushing rectangles and lines around in a graphics editor is something that only makes me anxious (I have used some Visio templates a few years ago, but they're all now requiring too many clicks when you're used to PlantUML). Plus, PlantUML also supports many other formats (including DOT, just put��@startdot/@enddot).��

On Fri, 20 Dec 2019 at 14:31, Kasper ��sterbye <kasper.osterbye@gmail.com> wrote:

On 20 December 2019 at 19.59.05, tbrunz (wild.ideas@gmail.com) wrote:
What I would find useful (and what I think Kasper has started down the road��
to) is the ability to generate PlantUML script from a specified (set of)��
packages or classes in Pharo. That would make it easy not only to examine��
the design (in a more synoptic way -- something I've seen several other��
newcomers express an interest in), but also to embed diagrams in��
documentation such as PDFs & web pages.��
I expect that to be done over the hollidays. I expect to have a working copies of
- Generate object diagram from a root (I will filter on category names and subclasses)
- Generate a sequence diagram by analysing a MessageTally from a block. Filter on categories, subclasses and methods
Short of that, and probably much easier to implement, is to just add��
PlantUML script to Pharo class comments as you write them -- but have a��
simple way to trigger the system to parse the comment and render the script��
into PlantUML diagrams. This could be in the form of opening a web browser��
in the host and having the diagram appear via the PlantUML server. But more��
useful perhaps would be to have it create PNG file(s) in the host filesystem��
(and open them).��
I have that working in my image. The��https://github.com/kasperosterbye/PillarRichTextRender��allow pillar to be used and rendered in class and package comments. Again, in my image I have an extension which allow the [[[ �� ]]] syntax of pillar (and ```of github) to use a language called ��plantuml��, which renders the diagram in the documentation. I hope to package it all up over the holidays or early January. The same thing allow latex math to be rendered in those same comments.
Best,
Kasper