Blog Backlog

2018-04-10

This post outlines topics I won't have time to write about before attending Recurse Center. If you're interested any of them, please [leave a comment][comments_url].

Questions about Python internals

pgen2 algorithm
why does lexer depend on await?
PYTHONHASHSEED, existence of a .pyc file
Oddities: _fixuprange() has a bug! The variable 'l' is never defined.
Why don't classes have a __code__ object like functions? I see
- TODO: Test Python 3.
Why is there types.BuiltinMethodType and types.BuiltinFunctionType?
Observations: signatures are LOST at runtime!!!! inspect.getsourcefile() and inspect.getargspec does alot of acrobatics.

Ideas

Will OSH Have Fewer Bugs than Bash?
- main points:
  - 8800 lines of code (cloc). Higher comment/blank ratio than bash.
  - This is unusual and valuable because it's REAL DATA
Project Retrospective: How I Rewrote Bash in 8800 Lines of Code
- April to October: pre-blog period, getting it working, convincing myself it could work
- October to July -- enough for a first release. It ran one real program
  - research on Opy, shipping the prototype
- July to January -- fleshing it out, and then it runs real programs, some optimization.
- January- OPy, hopefully working on compelling efatures
Preview of Oil Language
- fill out quickref
- honestly people wanted this, but I'm going to punt. I can call the project successful if I deliver a bash clone. I "front-loaded" the project so I wouldn't feel bad about walking away.
  - If I took 2 years to design a new incopmatible shell that nobody uses, I would feel bad.

Outline

Project Retrospective: How I Rewrote Bash in 8800 Lines of Code

Caveats:

It's in Python, which was slower than I expected. (I generally think large C and C++ codebases are slow due to lack of understanding and internal inconsistency, but even I have to admit that OSH is much slower than native code.)
non-comment, non-blank lines
only the programming language parts, not interactive parts

However, I claim:

the architecture of Oil will not change.
The architecture is significantly different than that of any other shell, due to static parsing.

The rest of the work on OSH will be:

Implementing the small features using the existing architecture
Making it faster. (Althoug this is a major architectral change.)
Fixing bugs and generally making it more robust. Polishing error messages.
techniques for code compression
- Python
- metaprogramming: core/id_kind has lots of lookup tables.
  - id /kind distinction
- ASDL
- re2c!
Sharing code! echo -e etc.

Drafts

Lexing Posts
- many drafts
The Evaluation Model of the OSH Interpreter? Kind of necessary to explain things.
Escaping vs. Quoting, Code vs. Data
- background for backslash
Backslash Meanings
- helped me sort out the OIl langauge
The Else-Whatever Anti-Pattern I'm just going to coin this term, and give a few examples
- \0800 through \0FFF
- \t should be t outside
- \t should be \t
- regex: [^a-b], [^ab-] should be [^ab-] [ab^] should be [ab^]
The Oil Project FAQ, Part 2
- What is OSH? -- This might be in the FAQ
- I have a draft of this, should publish
Later: shell techniques, for Oil language
- yeah I have a good one
A List of Things Academics Can Help With on Oil
- I want to invite people to help and scrutinize my code
A Plan to Make OSH Smaller
A list of system calls: Python/C/ Shell
- this is related to the plan to make OSH smaller
Github Help Wanted
- Maybe making a unix.c or libc.c module
  - kernel.c
  - can be replaced with stubs in the browser?
- osh2oil requires a lot of labor? fixing bugs?
Theme: theory vs. practice
- accidentally turing complete
- taking vector transposes seriously
  - looks into the history of mathematics, when vector algebra and linear algebra were separate
  - householder notation
  - lessons for Julia's type system
- I like pointing out good work, sort of like CommonMark
- maybe go back and tag:
  - pratt parsing
  - bash undecidable