Tuesday, January 26, 2021

The Awkwardness of Fundamental Definitions

A pragmatic description a language (like Python) from axiomatic -- or really axiom-like -- foundations is exasperatingly hard. I don't think I have any answers, but I sure do have a lot of challenges. I think there's a path that involves a lot of "don't look down -- just edge along the face of this cliff a few more feet and it's going to be okay."

The actual language foundations, and a more useful conceptual foundation don't always match up.

My specific example is trying to use a subset of Python to get started with.

One of the reasons we write software is to see useful results. So you need output, i.e., the print() function. On the other end of this, you might want to accept inputs. But. But... You could finesse that by simply assigning literal values instead of reading something. 

This leads to the print() function and expressions is a kind of minimal language to see the essence of programming. (I know this isn't an original thought. I'm setting up the conflict.)

Do we explain functions in general when we're explaining this subset of Python? Or do we stick with arithmetic? 

print("Hello, world!")
print("355/113=", 355/113)

Maybe the F-string?  Assignment? Or do we really need assignment? (Spoiler alert, no.)

What about other functions? Python has a bunch of built-ins. The math and random modules let us build small games without much intellectual overhead.

And how much can we explain about functions? The whole mathematical notion of function as mapping from domain to range: does that count? Or nah? What about Python's flexible argument and parameter handling? 

Exceptions? Do we have to explain them? Or do we shrug at division by zero and kind of ignore it? If we talk about exceptions, do we have to talk about stack frames and tracebacks?

At the foundation a language is all variable bindings and function evaluation. But. Do you explain any of that? And if so, how?

The additional complication is multiple authors, technical reviews, editors, and everyone else involved. There's a complex web of varying opinions on what's foundational and how much explanation is required. It's new to one person, so it should be up font. But, a lot of up-front material is boring. Everything in programming involves a tricky nuance somewhere; how much of that nuance is relevant now and how much is later and how much is digressive?


To an extent, there's a trust relationship between writer and reader. It may help to build this if the writer can provide a "trust me on this," scenario that (eventually) results in a more complete explanation. The logical conclusion here is that there's no place in the book for "too advanced, buy my next book." Instead, every difficult and nuanced thing would need to (eventually) be explained. That seems impossible, though.

With editors, co-authors, and reviewers, the trust relationship is exactly wrong. Everything needs to be challenged and clarified.

tl;dr Writing is sometimes hard. And that's the expectation. It's a narrow fairway surrounded by shoals. 

Tuesday, January 5, 2021

Diagrams and UML notation

When I started in this business I was given a flowcharting template.

See https://americanhistory.si.edu/collections/object-groups/flowcharting-templates. I'm pretty sure I had one of these: https://americanhistory.si.edu/collections/search/object/nmah_690078.

Since then, things have changed a little.

I fondly recall using the Rational Rose (and an earlier tool that did Rumbaugh OMT diagrams) to create object models.

But these were expensive.

After much searching, I found ArgoUML. This was my go-to-diagrammer of choice for many years. It's available here: https://argouml-tigris-org.github.io.

Then, I wound up using yuml at https://yuml.me/diagram/scruffy/class/draw. This was very nice because there was a source text version of the diagram. It was a high-level code-like description that would lead to a handy picture you could include in documentation.


Recently, I spent time using draw.io. Start here https://draw.io. You have plain-text source version of the diagram that's Git-friendly. I liked that. It has a lot of UML features, which is very nice, also.

But now. 

I'm using plantUML, and I think it's pretty handy. https://plantuml.com. It's a big-old JAR file that converts text to a diagram. There's no GUI component to this. You describe the image in a source-code like way. Run it through the tool, and you get a picture you can paste into documentation. Like yuml, it has an easy-to-understand high-level text description. I strongly suspect I could walk a Python AST and emit plantUML source as an intermediate language from which pictures can be created.

The Python-Markup tool (https://python-markdown.github.io/extensions/) has a third-party PlantUML plug-in. PyCharm can leverage this to draw while you're editing in the markdown window.

The Fiddly Bits

It's a little fiddly to get all the parts organized properly, but, it really, really does work. You can write technical documentation, with pictures.

  • Add the Markdown tool to PyCharm.
  • In the preferences for the PyCharm Markdown tool, install and enable PlantUML.
  • You can usually use conda to install graphviz as well as installing the plantuml-markdown tools. You can manually run the markdown_py application to create the HTML copies of the .md files. 
  • Update your OS environment settings to set the GRAPHVIZ_DOT environment variable to name the conda virtual environment where graphviz was installed. For macOS and Linux users update the ~/.zshrc (or ~/.bashrc) file, depending on which shell is in use. Windows users have my heartfelt sympathy; maybe set the system environment variables.
  • You may also need to create a plantuml shell script that's on your PATH. I put it in /usr/local/bin

See https://github.com/mikitex70/plantuml-markdown for details on installation.

After all this fussing around, it worked delightfully. I'm a convert to PlantUML.

I suggest the following in each diagram.

skinparam monochrome true
skinparam handwritten false
skinparam shadowing false
hide class circle

You may want to set a more global configuration, but I sometimes want to change the handwritten parameter to true for "draft" diagrams, separate from final. 


You can integrate plantUML with PyCharm to draw pictures while you're editing in the markdown window.

You do have to trust plantUML to draw more-or-less what you want. There are limits, and if you don't like what plantUML is doing, switch to draw.io. If you are flexible, however, it's really, really good.