What is GitHub Copilot?

Description

It's time to take a break, stretch our legs, grab a drink, and maybe even join in some interesting conversations around the water cooler as Michael goes off script, Joe is very confused, and Allen insists that we stay on script.

The full show notes for this episode are available at https://www.codingblocks.net/episode163. Stop by, check it out, and join the conversation.

Survey Says

News

We really appreciate the latest reviews, so thank you!
- iTunes: EveryNickIsTaken2858, Memnoch97
Allen finished his latest ergonomic keyboard review: Moonlander Ergonomic Keyboard Long Term Review (YouTube)
Sadly, the .http files tip from episode 161 for JetBrains IDEs is only application for JetBrains' Ultimate version.
- Test RESTful Web services (JetBrains)

Meantime, at the watercooler….

GitHub Copilot (GitHub)

In short, it's a VS Code Extension that leverages the OpenAI Codex, a product that translates natural language to code, in order to … wait for it … write code. It's currently in limited preview.

What's the value?

Is the code correct? Github says ~40-50% in some large scale test cases
It works best with small, documented functions
Does having the code written for you steer you towards solutions?
Could this encourage similar bugs/security holes across multiple languages by people importing the same code?
Is this any different from developers using the same common solutions from StackOverflow?
Could it become a crutch for new developers?
Better for certain kinds of code? (Boiler plate, common accessors, date math)
- Boiler Plate (like angular / controller vars)
- Common APIs (Twitter, Goodreads)
- Common Algorithms, Design Patterns
- Less Familiar Languages
But is it useful? We'll see!

Is this the future?

We see more low, no, and now co-code solutions all the time, is this where things are going?
This probably won't be "it", but maybe we will see things like this more commonly – in any case it's different, why not give it a shot?

Is it Ethical?

The "AI" or whatever has been trained on "billions of lines" of open-source code…but not strictly permissive licenses. This means a dev using this tool runs the risk of accidently including proprietary code
- Quake Engine Source Code Example (GPLv2) (Twitter)
From an article in VentureBeat:
- 54 million public software repositories hosted on GitHub as of May 2020 (for Python) 179GB of unique Python files under 1MB in size. Some basic limitations on line and file length, sanitization: The final training dataset totaled 159GB.
- There is problem with bias, especially in more niche categories
Is it ethical to use somebody else's data to train an AI without their permission?
Can it get you sued?
Would your thoughts change if the data is public? License restricted?
Would your thoughts change if the product/model were open-sourced?

Abstractions… how far is too far?

Services should communicate with datastores and services via APIs that hide the details, these provide for a nice indirection that allows for easier maintenance in the future
Do you abstract at the service level or the feature level?
Are ORMs a foregone conclusion?
What about services that have a unique communication pattern, or assist with cross cutting concerns for things like microservices (We are looking at you hear Kafka!)

The 10 Best Practices for Remote Software Engineering

From article: The 10 Best Practices for Remote Software Engineering (ACM)
- Work on Things You Care About
- Define Goals for Yourself
- Define Productivity for Yourself
- Establish Routine and Environment
- Take Responsibility for Your Work
- Take Responsibility for Human Connection
- Practice Empathetic Review
- Have Self-Compassion
- Learn to Say Yes, No, and Not Anymore
- Choose Correct Communication Channels

Terminal Tricks (CodeMag.com)

Some of Michael's (Linux/macOS) favorites from the article:

Abbreviate your directories with tab completion when changing directories, such as cd /v/l/a, and assuming that that abbreviated path can uniquely identify something like, /var/logs/apache, tab completion will take care of the rest.
Use nl to get a numbered list of some previous command's output, such as ls -l | nl.
- ERRATUM: During the episode, Michael mentioned that the output would first list the total lines, but that just happened to be due to output from ll and was unrelated to the output from nl.
On macOS, you can use the powermetrics command to gain access to all sorts of metrics related to the internals of your computer, such as the temperature at various sensors.
Use !! to repeat the last command. This can be especially helpful when you want to do something like prepend/append the previous command, such as sudo !!.
- ERRATUM: Wow, Michael really got this one wrong during the episode. It doesn't repeat the "last sudo command" nor does it leave the command in edit mode. Listen to Allen's description. /8)
Awesome keyboard shortcuts:
- CTRL+A takes you to the start of the line and CTRL+E takes you to the end.
- No need to type clear any longer as CTRL+L will clear your screen.
- CTRL+U deletes the content to the left of the cursor and CTRL+K deletes the content to the right of the cursor.
- Made a mistake in while typing your command? Use CTRL+SHIFT+- to undo what you last typed.
Using the history command, you can see your previous commands and even limit it with a negative number, such as history -5 to see only the last five commands.

Tip of the Week

Partial Diff is a VS Code extension that makes it easy to compare text. You can right click to compare files or even blocks of text in the same file, as well as in different files. (Visual Studio Marketplace)
StackBlitz is an online development environment for full stack applications. (StackBlitz.com)
Microcks, an open source Kubernetes native tool for API mocking and testing. (Microcks.io)
Bridging the HTTP protocol to Apache Kafka (Strimzi.io)
Difference Between grep, sed, and awk (Baeldung.com)
As an alternative to the ruler hack mentioned in episode 161, there are several compact, travel ready laptop stands. (Amazon)

Listen