wip: lua metaprogram rework

scripts/passes/offsets.lua

@@ -0,0 +1,43 @@
+-- passes/offsets.lua
+--
+-- Generate <module>/gen/<basename>.offsets.h with branch offset
+-- immediates for every atom_offset(F, T) reference in atom bodies.
+-- Ported from scripts/tape_atom.offset_gen.meta.lua:148-389.
+--
+-- The branch offset regression we just fixed in commit 98e27c2 must
+-- NOT return. The fix was in duffle.lua's split_top_level_commas +
+-- tape_atom_annotation_pass.lua's compute_component_word_count.
+-- word_count_eval.count_token_words (Task 1) preserves the fix.
+--
+-- Coding standard: tabs (1/level), EmmyLua annotations, no regex.
+
+--- @class M
+
+local M = {}
+
+--- @param ctx PassCtx
+--- @return PassResult
+function M.run(ctx)
+	-- PORT NOTE: copy find_atoms, scan_atom_body, compute_offsets,
+	-- generate_header, process_source from tape_atom.offset_gen.meta.lua
+	-- lines 245-389.
+	-- Adaptations:
+	--   - Replace local word_count_of_token with word_count_eval.count_token_words
+	--   - Take (ctx, src) instead of (source_path, word_counts) — read
+	--     word counts from ctx.shared.word_counts
+	--   - Return { outputs, errors, warnings } shape
+	--   - Convert indentation from 8-space to tabs
+	--
+	-- For each src in ctx.sources:
+	--   1. Find MipsAtom_ declarations in src.text (find_atoms)
+	--   2. For each atom body, scan_atom_body to count words + find
+	--      atom_label/atom_offset markers
+	--   3. compute_offsets (target_word - branch_word - 1 per pair)
+	--   4. generate_header emits "#define atom_offset__F__T  (N)" +
+	--      "enum { atom_offset__F__T = N, ... }" to <src.dir>/gen/<src.basename>.offsets.h
+	--   5. Add {offsets_h = path} to result.outputs
+	error("passes.offsets.run: implement by porting from " ..
+		"tape_atom.offset_gen.meta.lua:245-389")
+end
+
+return M