(Back to Shell Autocompletion)

1. Completing the shell language itself. The knowledge to do this is all in the shell binary.
   • Completing variable names for ${v<TAB>}.
   • Completing keywords like for, while, etc.
   • Completing the shell's own builtins like set -o <TAB>.
   • In zsh: completing various glob modes.
2. Completing flags names and arguments. This info is inherently "upstream".
   • ... for common tools, which can be grouped:
     • GNU coreutils and family tend to use the same syntax. bash_completion treats them all the same.
     • tools in Go tend to use the flags package, which has a consistent syntax.
   • ... for unusual tools
     • Diversity in Command Line Syntax -- bash itself has a nonstandard flag syntax, python -c, test, etc.
   • Very large tools -- it would be nicer to reuse the work from upstream.
     • git
     • clang
     • npm?
   • NOTE: Completing file system paths is a special case of a flag argument. The behavior here may be configurable in the shell, e.g. case-sensitivity.
3. Completing languages/DSLs within arguments.
   • This is generally a harder problem than completing args, because it may require recursive parsing.
   • sed -e -- completing all the one letter commands with descriptions, completing the one letter flags with descriptions (from Oliver Kiddle)
   • ssh -- this is equivalent to parsing all of shell! i.e. the OSH parser. (from Oliver Kiddle)
   • awk
   • Regexes? I always forget how spell character classes like [:space:](:space:.html).

Notes:

• Shellac Protocol Proposal handles #2 and #3. #1 is up to the shell binary.
• It is not clear to me (andychu) if grammar-based approaches can handle #3 (in general).