Why Sponsor Oils? | blog | oilshell.org

Blog Retrospective #2

In the first blog retrospective, written in November, I organized blog posts into five themes:

1. Project Updates
2. Shell Trivia / WTFs
3. Oil Language Design
4. Static Parsing of Bash
5. Parsing Algorithms and Techniques

Here are the entries I've written since, also organized by theme:

1. Oil Language

I validated the Oil language design by Translating Shell to Oil. Part Two goes into more detail.

The requirement to automatically translate shell to Oil will have a large effect on the language design. I've come to view that as a good thing.

2. Shell: The Good Parts

• Shell Has a Forth-Like Quality. Bernstein chaining is a powerful mechanism of composition that most languages don't have.
• Pipelines Support Vectorized, Point-Free, and Imperative Style. Pipelines are another powerful mechanism that most languages don't have.
• Bernstein chaining of ssh and su

I plan at least two more posts in this series, but they'll have to wait until after the initial OSH release.

3. Language Implementation Techniques

(a) Zephyr ASDL. It took me about two months to replace the core of the interpreter with ASDL, so it was one of the biggest topics.

• Before implementation:
  • What is Zephyr ASDL?
  • Analogies for ASDL
• After implementation:
  • Success With ASDL
  • ASDL Implementation and Use Cases
  • What is OHeap? Although OHeap is on hold because of the new direction with OPy, it may play a role later in the project.
  • Pretty Printing ASTs with ASDL, and Metaprogramming
  • The Thinner Waist of the Interpreter

(b) Other Implementation Techniques

• Pratt Parsing Index and Updates. An index of popular posts on algorithms for parsing expressions.
• From AST to Lossless Syntax Tree. A pattern for representing code. Also see the wiki page.
• Questions About Superoptimization is a diversion into code generation. It was fun, but likely won't be used, especially with the the new OPy architecture. But I learned things about C compilers that will prove useful.

4. Programming Language Design

Implementing Oil has shed light on a few language design issues. Though I only wrote about one tradeoff, I have many more notes.

• A Problem with Type Checking. Python 3's static type checker wasn't worth it for the OSH codebase.
• Type Checking vs. Metaprogramming; ML vs. Lisp. Addendum: For OSH, dynamic typing with the ML-like model of ASDL is better than static typing with a Java-like model.

5. Project Status Updates

Writing these status updates helps me organize the project. In chronological order:

• November 2016
  • Roadmap #3: Two Top Priorities
  • Roadmap #3 Continued
• January 2017
  • On the Wiki: Project Goals and Related Projects
• February
  • Review of Roadmap #3
  • Project Roadmap #4
• March
  • Oil-dev is Alive!. I hope that oil-dev will be more active after an initial release of OSH.
  • Measuring Progress with Tests
• April
  • The Riskiest Part of the Project
  • An Unexpected Solution: Cobbling Together a Python Interpreter. As mentioned yesterday, this solution will evolve.

6. Blog Meta

• Using Reddit for Blog Comments. I'm happy with how this turned out. I'm not the only one to have problems with popular comment services.

These entries enumerate topics I wanted to address, but didn't have the time to. If you're interested in a particular topic, feel free to leave a comment.

• January: Blog TODO Stack
• February: Status Update and Blog Backlog
  • Notably, Shell: The Bad Parts got dropped. I will return to this topic when I've made more progress on the Oil language, because it's designed to address them.
• April: Switching to "Twitter Mode"

Reminder

I put the blog into Twitter mode in order to concentrate on coding. After the OSH 0.1 release, I should have time to write more substantive posts.