consider an IR for words and ${}

[andychu]

These are confusing to disambiguate from the LST:

• ${!ref-default} -- prefix_op and unary suffix op
• ${!prefix@} -- prefix_op and nullary suffix_op
• ${!array_keys[@]} -- prefix op and bracket op

We could compile it to something like:

braced_var_sub =   Ref(...) | Prefix(...) | Keys(...)

instead of prefix_op / suffix_op / bracket_op

It could be a step that's analogous to BraceExpand. It could be done at parse time, but we don't print the LST with it.

• So maybe store functions with the more compact representation? But top-level code does it at runtime. Hm.

---

For words

• osh/word_compile.py
  • $'c style \n strings' are evaluated at runtime, but they don't have to be
  • triple_quoted has the same issue? Has to strip leading space
• We don't really need to do detection and expansion of brace substitutions separately. That was more for the LST invariant ...

[andychu]

NOTE: sh_lhs_expr is sort of an IR because it's used for both:

• (( x = 42 )) and (( a[i] = 42 ))
• x=42 and a[i]=42 (no spaces allowed)

edit: Actually it's not an IR because we don't make separate syntax tree nodes first. It's a rare case of losing information! We could make it more like one.

[andychu]

Note: This is done for arithmetic, e.g. for #640 and also for #651

Not sure about ${}, that may not happen for awhile.