Many commits squashed; turns out that with the combination of:
* Python v2.7,
* Sphinx v1.8.5, and
* Pygments v2.3.1
versions (old!) used in the CI, the only viable approach is:
* to use `code-block` directives with explicit language specification,
* to provide no file-local default using `highlight`, and
* to set language as `none` for grammar specifications.
Underlying are the following issues (again, for the old versions
listed above):
* Generic RST `code` doesn't work when language is `none`:
Warning, treated as error:
/root/project/docs/yul.rst:430:Cannot analyze code. No Pygments lexer found for "none".
Additionally, it might be trying to fall back to the default
(Solidity) if left unspecified.
* If a file-local default is specified using `highlight`, then
`code-block` _must_ also provide a language:
Warning, treated as error:
/root/project/docs/yul.rst:185:Error in "code-block" directive:
1 argument(s) required, 0 supplied.
* Sphinx seems to try the file-local default "yul" (specified with
`highlight`) on `code` marked having language `json`:
Warning, treated as error:
/root/project/docs/yul.rst:130:Could not lex literal_block as "yul". Highlighting skipped.
* The only well-lexed highlighter for two of the three grammar
specifications is `peg`, but it was added in Pygments v2.6.
One of the grammars - in the "Formal Specification" section,
the one after "We will use a destructuring notation for the
AST nodes." - _must_ be left unhighlighted, with language set
to `none`: all lexers do really poorly.
... And one should never, ever start treating warnings as mere
warnings, without having exhausted all other options.
Otherwise, it's a slippery slope, - and look where that brought
Gandhi: to being a strawman in every lousy argument to be had!..
- Long chromosomes in the intial population are better. Set minimum and maximum to 100.
- The classic algorithm does not work well without elite. 50% performed better but I think it might be too large. Let's set it to 25%.
- Switch to uniform crossover since this is what was used in most experiments and performed well.
- We want to start accepting abbreviation sequences on the command line and parentheses would always have to be escaped in that context.
- There wasn't any important reason behind choosing () rather than [] or {} and it still isn't too late to switch.
- The script was leaving hundreds of loose `tmp.XXXXXX` and `tmp.XXXXXX.bak` files in `/tmp` after each run
- There's a trap handler that removes them but it's being registered multiple times in a loop and only the last one actually runs when the script exits. It's still useful because it removes the remaining files from the most recent iteration but on its own it's not enough to clean up everything.
