lang::typecheck::subst
lang/src/typecheck/subst.mo
De Bruijn shift/substitution for the canonical Term type — the two
primitives the macro-expansion phase's TERM-producing macro
application (defmacro name params := <term>) needs.
Standard, well-known PL theory (Pierce, Types and Programming
Languages, §6/§7) — no macro-specific hygiene/gensym-renaming
needed here, unlike the Rust reference's own subst_macro
(rust/core/src/eval/macro_expand.rs), which is NAME-based (walks a
pre-de-Bruijn-resolution Term matching NameRefs) and needs
alpha_rename_body's gensym counter specifically to stop two
DIFFERENT macro expansions' own introduced binders from colliding
on the same name. Self-hosted's canonical Term is de-Bruijn-
indexed from the parser itself — there is no "same name" for two
binders to collide on, so this whole class of bug is inherently
impossible here by construction. See
plans/bootstrapping/self-hosted-compiler.md's macro-expansion-phase
section for the full reasoning.
NOTE: this module is intentionally NOT used for the DECL-generating
macro form (defmacro name params := decls { ... }) — a decl
template's own top-level decl_list have no enclosing lambda binder, so
a macro param referenced inside one is an ordinary FREE/qualified
Term.var sentinel (DebugName.named X) reference, not a bound de
Bruijn index. That form needs a separate, NAME-based substitution
(mirroring the reference's own subst_macro/subst_decl_var much
more directly) — a different module, not this one.
pub def term_shift_above (cutoff : I64) (d : I64) (t : Term) : Termterm_shift restricted to variables at or above cutoff: the OUTER
references of a term being placed under new binders, leaving that term's
references to the binders just introduced (indices below cutoff) alone.
A constructor's field types reference its EARLIER fields at indices
0..i-1 and the enclosing context at i and above, so placing field i
under i new binders -- what the Elim minor's Pi chain and lambda chain
both do -- is exactly term_shift_above i i.