ed/pikuma_ps1
1
0
Fork 0
mirror of https://github.com/Ed94/pikuma_ps1.git synced 2026-08-25 02:20:33 +00:00
Code Issues Packages Projects Releases Wiki Activity

WIP: reviewing lua, some upgrades and fixes along the way.

Browse Source
This commit is contained in:
ed
2026-08-19 21:11:09 -04:00
parent cf78cfa120
commit bde829bf59
14 changed files with 1487 additions and 1121 deletions
Download Patch File Download Diff File Expand all files Collapse all files
scripts/duffle_emit.lua
+60 -33
View File
@@ -210,7 +210,7 @@ local function _project_emission_inner(root_body_entry, ctx_table)
local prefix = reg_use_param .. "."
if operand:sub(1, #prefix) ~= prefix then return nil end
local member_path = operand:sub(#prefix + 1)
local slot = reg_use_schema.alias_to_slot[member_path]
local slot = reg_use_schema.alias_to_slot[member_path]
if not slot then return nil, member_path end
return "reguse:" .. atom_name .. ":" .. slot, nil, slot
end
@@ -223,9 +223,7 @@ local function _project_emission_inner(root_body_entry, ctx_table)
return ids
end
local function emit_word(encoder, args, line, word_call_text,
def_source_now, def_line_now,
immediate_call_text, root_call_text_w, sub_map)
local function emit_word(encoder, args, line, word_call_text, def_source_now, def_line_now, immediate_call_text, root_call_text_w, sub_map)
local inv_ids = open_invocation_ids_snapshot()
local outermost = inv_ids[1] or 0
-- For words emitted at the root atom body, `immediate_call_text` is nil and the walker's `word_call_text` (the word's own token, e.g. "nop") becomes the effective call_text.
@@ -233,7 +231,7 @@ local function _project_emission_inner(root_body_entry, ctx_table)
-- The call that triggered the body expansion we're currently walking.
local eff_call_text = immediate_call_text or word_call_text
local eff_root_call_text = root_call_text_w
local gpr_keys = nil
local gpr_keys = nil
if reg_use_schema or sub_map then
gpr_keys = {}
for pos, arg in ipairs(args or {}) do
@@ -268,17 +266,17 @@ local function _project_emission_inner(root_body_entry, ctx_table)
if not reg_use_schema then
gpr_keys = nil
end
local isa = M.instr(encoder)
local isa_kind = isa and isa.kind or "unknown"
local isa = M.instr(encoder)
local isa_kind = isa and isa.kind or "unknown"
local nop_words = (encoder == "nop" and 1) or (encoder == "nop2" and 2) or 0
local is_yield = (encoder == "mac_yield" or encoder == "mac_yield_tail")
local is_yield = (encoder == "mac_yield" or encoder == "mac_yield_tail")
local gp0_shape = type(encoder) == "string"
and encoder:match("^mac_format_([%w_]+)_color$")
or nil
local is_load = (isa_kind == "load")
local is_branch = (isa_kind == "branch")
local is_load = (isa_kind == "load")
local is_branch = (isa_kind == "branch")
local is_unconditional_jump = (encoder == "jump" or encoder == "call_addr")
local is_terminal_jump = (encoder == "jump_reg" or encoder == "call_reg" or encoder == "jump_link")
local is_terminal_jump = (encoder == "jump_reg" or encoder == "call_reg" or encoder == "jump_link")
items[#items + 1] = {
kind = "word",
encoder = encoder,
@@ -335,14 +333,18 @@ local function _project_emission_inner(root_body_entry, ctx_table)
-- `consuming_encoder` + `consuming_arg_pos` carry the surrounding control-transfer instruction context
-- (e.g. `branch_le_zero` consuming its 3rd argument, or `jump` / `call_addr` consuming their only argument).
-- `passes/offsets.lua` reads these to dispatch per-consuming-instruction offset encoding.
-- nil for top-level markers (where the marker is the entire token — no surrounding consuming instruction).
-- Offset markers require a real consuming encoder. A lone top-level `atom_offset` is not emitted.
-- Label and delay markers may have a nil encoder (they are not consumed as immediates).
if kind == "offset" and (consuming_encoder == nil or consuming_encoder == "") then
return
end
local it = {
kind = kind,
name = name,
line = line,
word_index = word_idx,
invocation_ids = inv_ids,
outermost_invocation_id = outermost,
kind = kind,
name = name,
line = line,
word_index = word_idx,
invocation_ids = inv_ids,
outermost_invocation_id = outermost,
}
if target ~= nil then it.target = target end
if consuming_encoder then it.consuming_encoder = consuming_encoder end
@@ -450,15 +452,14 @@ local function _project_emission_inner(root_body_entry, ctx_table)
-- Commit: label takes 1 arg, offset takes 2.
-- For embedded markers, propagate the consuming_encoder + the marker's arg position
-- (1-based) so `passes/offsets.lua` can dispatch per-consuming-instruction offset encoding.
-- Top-level markers (no consuming_encoder) get nil for both — the offsets pass treats
-- them as branch-equivalent for backward compatibility.
-- Offset markers are emitted only when a consuming encoder is present.
local arg_pos = nil
if consuming_encoder and consuming_paren then
arg_pos = count_top_level_commas(tok, consuming_paren + 1, pos) + 1
end
local args = split_call_args(inner)
if ident == "atom_label" then emit_marker("label", args[1] or "", nil, tok_line, nil, nil, consuming_encoder, arg_pos)
else emit_marker("offset", args[1] or "", args[2] or "", tok_line, nil, nil, consuming_encoder, arg_pos)
elseif consuming_encoder then emit_marker("offset", args[1] or "", args[2] or "", tok_line, nil, nil, consuming_encoder, arg_pos)
end
pos = after_paren
::continue_loop::
@@ -545,7 +546,7 @@ local function _project_emission_inner(root_body_entry, ctx_table)
local wc = ctx_table.word_counts
if wc and wc[ident] then return wc[ident] end
local canon = M.gte_canon(ident)
if canon ~= ident and wc and wc[canon] then return wc[canon] end
if canon ~= ident and wc and wc[canon] then return wc[canon] end
warnings[#warnings + 1] = {
kind = "uncounted",
line = tok_line,
@@ -572,6 +573,18 @@ local function _project_emission_inner(root_body_entry, ctx_table)
-- "opaque word" emit handles direct encoders + mac_X-without-component.
local function process_token(bt)
local tok = M.trim(bt.tok or "")
-- Substituted MipsCode args can carry // comments from the call site.
while tok ~= "" do
if tok:sub(1, 2) == "//" then
local nl = tok:find("\n")
tok = M.trim(nl and tok:sub(nl + 1) or "")
elseif tok:sub(1, 2) == "/*" then
local close = tok:find("*/", 3, true)
tok = M.trim(close and tok:sub(close + 2) or "")
else
break
end
end
if tok == "" then return end
local ident, after = M.read_ident(tok, 1)
if not ident then ident = "?" end
@@ -582,10 +595,16 @@ local function _project_emission_inner(root_body_entry, ctx_table)
local rest = tok:sub(after or (#tok + 1))
while true do
rest = M.trim(rest)
if rest:sub(1, 2) ~= "/*" then break end
local close = rest:find("*/", 3, true)
if not close then rest = ""; break end
rest = rest:sub(close + 2)
if rest:sub(1, 2) == "//" then
local nl = rest:find("\n")
rest = nl and rest:sub(nl + 1) or ""
elseif rest:sub(1, 2) == "/*" then
local close = rest:find("*/", 3, true)
if not close then rest = ""; break end
rest = rest:sub(close + 2)
else
break
end
end
if rest ~= "" then
process_token({ tok = rest, rel = bt.rel })
@@ -601,12 +620,19 @@ local function _project_emission_inner(root_body_entry, ctx_table)
emit_embedded_markers(tok, tok_line, consuming_encoder_for_markers)
end
-- atom_label / atom_offset: terminal markers, no further descent.
-- Top-level markers (the marker IS the entire token) have no consuming instruction;
-- nil for both `consuming_encoder` and `consuming_arg_pos`.
-- The offsets pass treats these as branch-equivalent for backward compatibility.
-- TODO(Ed): Review this don't want legacy cruft here..
if ident == "atom_label" then emit_marker("label", args[1] or "", nil, tok_line); return
elseif ident == "atom_offset" then emit_marker("offset", args[1] or "", args[2] or "", tok_line); return
-- A lone top-level atom_label is an anchor and is emitted with no consuming encoder.
-- A lone top-level atom_offset has no consuming encoder and is not emitted.
if ident == "atom_label" then emit_marker("label", args[1] or "", nil, tok_line); return
elseif ident == "atom_offset" then return
end
-- MipsCode formals (nop_slot1, …): the ident is a sub_map key.
-- Re-process the replacement token so load_word(...) becomes a real encoder.
if sub_map and type(sub_map[ident]) == "string" and sub_map[ident] ~= ident then
local repl = M.trim(sub_map[ident])
if repl ~= "" then
process_token({ tok = repl, rel = bt.rel })
return
end
end
if ident:sub(1, 4) == "mac_" then
local bare = ident:sub(5)
@@ -738,7 +764,8 @@ end
--- * Metadata-backed N-word tokens (`nop2`, `mask_upper`, ...): N `word` items, all sharing the same encoder + word_count = 1.
--- `nop2` is normalized to encoder `nop` (per the spec).
--- * `atom_label(F)` markers: one `label` item with `name = "F"`, `word_index = current word_idx`; zero-width (does NOT advance word_idx).
--- * `atom_offset(B, T)` markers: one `offset` item with `name = "B"`, `target = "T"`, `word_index = current word_idx`; zero-width.
--- * `atom_offset(B, T)` nested in a consuming instruction: one `offset` item with `name = "B"`, `target = "T"`, `word_index = current word_idx`; zero-width.
--- A lone top-level `atom_offset` is not emitted.
--- * Delay markers (`GteDelay_` / `LdSlot_` / `BdSlot_` / `DmaSlot_`): one `delay` item; zero-width. The following encoder is the next token.
--- * `mac_X(...)` calls: emit `invoke_begin` (zero-width), recurse into the component body, emit `invoke_end` (zero-width).
--- The component body's words land between the begin/end pair; one invocation record is allocated per call (monotonic ID per atom).
scripts/elf_dwarf.lua
+56 -41
View File
@@ -13,7 +13,7 @@ local lfs = require("lfs")
-- scripts/elf32.lua contains format-constant tables + the byte-level walker.
-- The this file re-exports `read_u32_le` / `read_u16_le` (and the DWARF32 terminator).
-- TODO(Ed): Remove re-export.
-- read_u32_le is this module's reader; implementation in elf32.lua.
local E = require("elf32")
local M = {}
@@ -113,7 +113,6 @@ M.MIPS_BYTES_PER_WORD = 0x04
--- spec: DWARF4 spec §7.4 — 32-bit DWARF initial-length terminator
M.dw_dwarf32_terminator = E.dw_dwarf32_terminator
-- TODO(Ed): Remove re-export.
-- ----------------------------------------------------------------------------
-- DWARF4 .debug_aranges (per DWARF5 spec §7.4 — Address Range Table)
@@ -372,48 +371,70 @@ end
-- For DW_FORM_strp we return the inline string resolved from `str_buf`.
-- For DW_FORM_ref4 we return the absolute CU-relative offset.
-- The caller decides whether to interpret that as a section offset.
local function read_form_value(buf, str_buf, pos, form)
if form == M.DW_FORM.addr then
local FORM_READERS = {
[M.DW_FORM.addr] = function(buf, _, pos)
return M.read_u32_le(buf, pos), pos + 4
elseif form == M.DW_FORM.string then
end,
[M.DW_FORM.string] = function(buf, _, pos)
local s = read_c_string_at(buf, pos)
return s, pos + #s + 1
elseif form == M.DW_FORM.strp then
end,
[M.DW_FORM.strp] = function(buf, str_buf, pos)
-- DW_FORM_strp: 4-byte offset into .debug_str.
local strp_off = M.read_u32_le(buf, pos)
return read_c_string_at(str_buf, strp_off), pos + 4
elseif form == M.DW_FORM.udata then return M.read_uleb128_at(buf, pos)
elseif form == M.DW_FORM.data1 then return buf:byte(pos + 1), pos + 1
elseif form == M.DW_FORM.data2 then return M.read_u16_le(buf, pos), pos + 2
elseif form == M.DW_FORM.data4 then return M.read_u32_le(buf, pos), pos + 4
elseif form == M.DW_FORM.ref4 then return M.read_u32_le(buf, pos), pos + 4
elseif form == M.DW_FORM.sec_offset then
end,
[M.DW_FORM.udata] = function(buf, _, pos)
return M.read_uleb128_at(buf, pos)
end,
[M.DW_FORM.data1] = function(buf, _, pos)
return buf:byte(pos + 1), pos + 1
end,
[M.DW_FORM.data2] = function(buf, _, pos)
return M.read_u16_le(buf, pos), pos + 2
end,
[M.DW_FORM.data4] = function(buf, _, pos)
return M.read_u32_le(buf, pos), pos + 4
end,
[M.DW_FORM.ref4] = function(buf, _, pos)
return M.read_u32_le(buf, pos), pos + 4
end,
[M.DW_FORM.sec_offset] = function(buf, _, pos)
-- DW_FORM_sec_offset: 4-byte offset (size depends on DWARF version;
-- on DWARF5 32-bit it's always 4 bytes).
return M.read_u32_le(buf, pos), pos + 4
elseif form == M.DW_FORM.flag_present then
end,
[M.DW_FORM.flag_present] = function(_, _, pos)
return 1, pos
elseif form == M.DW_FORM.exprloc then
end,
[M.DW_FORM.exprloc] = function(buf, _, pos)
-- DW_FORM_exprloc: ULEB byte count + that many bytes of DW_OP_*.
local len, ne = M.read_uleb128_at(buf, pos)
local len, ne = M.read_uleb128_at(buf, pos)
if not len then return nil, pos end
return nil, ne + len
elseif form == DW_FORM_implicit_const then
end,
[DW_FORM_implicit_const] = function(_, _, pos)
-- The constant is declared in the abbrev; no value bytes in the DIE.
return nil, pos
elseif form == M.DW_FORM.ref_sig8 then
end,
[M.DW_FORM.ref_sig8] = function(buf, _, pos)
-- DW_FORM_ref_sig8 (DWARF5 §7.4.2): An 8-byte value identifying a type by signature.
-- The low 4 bytes (LE) are the type signature (content hash);
-- The high 4 bytes (LE) are a CU-relative offset into the matching type unit.
-- Consumers use the low 4 to look up the type unit (see M.find_type_unit_by_signature)
-- then the high 4 to resolve the specific type within it.
-- Consumers use the low 4 to look up the type unit (see M.find_type_unit_by_signature)
-- then the high 4 to resolve the specific type within it.
-- Return the low 4 as the primary value to preserve the (value, next_pos) shape;
-- the high 4 is exposed via M.read_ref_sig8 (which returns both halves).
-- the high 4 is exposed via M.read_ref_sig8 (which returns both halves).
local _, _, next_pos = M.read_ref_sig8(buf, pos)
return M.read_u32_le(buf, pos), next_pos
else
end,
}
local function read_form_value(buf, str_buf, pos, form)
local r = FORM_READERS[form]
if not r then
return nil, pos
end
return r(buf, str_buf, pos)
end
--- Read a `DW_FORM_ref_sig8` value at 0-based offset `pos` from `buf`.
@@ -426,9 +447,7 @@ end
--- @return integer -- low 4 bytes (LE), the type signature
--- @return integer -- high 4 bytes (LE), the offset within the matching type unit
--- @return integer -- cursor after the 8-byte value
function M.read_ref_sig8(buf, pos)
return M.read_u32_le(buf, pos), M.read_u32_le(buf, pos + 4), pos + 8
end
function M.read_ref_sig8(buf, pos) return M.read_u32_le(buf, pos), M.read_u32_le(buf, pos + 4), pos + 8 end
--- DWARF5 §7.5.6 (Type Entries).
--- Walk all units in `info` and return the 0-based offset of the first unit whose `DW_AT_type_signature`
@@ -459,23 +478,23 @@ function M.find_type_unit_by_signature(info, target_sig_lo, target_sig_hi)
return nil, nil -- malformed
end
-- Per DWARF5 §7.5.6, the type_unit (DW_UT_type = 0x02) body layout is:
-- 0: version (2)
-- 2: unit_type (1) -- DW_UT_type = 0x02
-- 3: address_size (1)
-- 0: version (2)
-- 2: unit_type (1) -- DW_UT_type = 0x02
-- 3: address_size (1)
-- 4: debug_abbrev_offset (4)
-- 8: type_signature (8)
-- 16: type_offset (4)
-- 8: type_signature (8)
-- 16: type_offset (4)
-- 20: <children>
if body_end - body_start >= 20 then
-- read_ref_sig8 / write_u32_le / etc. are 1-indexed (string:byte);
-- pos / body_start / body_end are 0-based wire offsets, so the 1-indexed byte at 0-based wire offset X is string:byte(X + 1).
-- Per DWARF5 §7.5.6, the type_unit body is laid out as:
-- byte 0-1: version (2)
-- byte 2: unit_type (1) -- DW_UT_type = 0x02
-- byte 3: address_size (1)
-- byte 0-1: version (2)
-- byte 2: unit_type (1) -- DW_UT_type = 0x02
-- byte 3: address_size (1)
-- byte 4-7: debug_abbrev_offset (4)
-- byte 8-15: type_signature (8)
-- byte 16-19: type_offset (4)
-- byte 8-15: type_signature (8)
-- byte 16-19: type_offset (4)
local unit_type = info:byte(body_start + 2 + 1) -- 0-based +2 = unit_type in 1-indexed
if unit_type == 0x02 then -- DW_UT_type
local sig_lo, sig_hi, _ = M.read_ref_sig8(info, body_start + 8) -- 0-based +8 = type_signature in 1-indexed
@@ -626,14 +645,10 @@ function M.read_nm(elf_path)
local addrs = {}
-- Existence check first; an empty or missing ELF returns an empty map.
if lfs.attributes(elf_path, "mode") ~= "file" then
return addrs
end
if lfs.attributes(elf_path, "mode") ~= "file" then return addrs end
local f = io.open(elf_path, "rb")
if not f then
return addrs
end
if not f then return addrs end
-- Build the file adapter for E.*.
local file_size
@@ -775,7 +790,7 @@ end
--- @return string
function M.sleb128(n)
local bytes = {}
local more = true
local more = true
while more do
local b = n % (LEB_DATA_MASK + 1) -- extract low 7 bits
n = (n - b) / (LEB_DATA_MASK + 1) -- arithmetic shift right by 7
scripts/passes/annotation.lua
+38 -57
View File
@@ -45,14 +45,12 @@ local duffle = dofile(_bootstrap_dir .. "../duffle_paths.lua")
--- @field warnings table[]
--- @class AtomAnnotation
--- @field line integer -- Source line of the atom_info call
--- @field macro string -- Macro name (always "atom_info" in the new shape)
--- @field name string -- Atom name
--- @field kind string -- Always "info"
--- @field binds string|nil -- Binds_X name if any
--- @field reads string[] -- R_* names (read targets)
--- @field writes string[] -- R_* names (write targets)
--- @field errors string[]|nil -- Parse-time errors from scan_source (atom_info body malformed)
--- @field atom_name string -- Atom name (scan.atom_infos row)
--- @field info_line integer -- Source line of the atom_info call
--- @field binds string|nil -- Binds_X name if any
--- @field reads string[] -- R_* names (read targets)
--- @field writes string[] -- R_* names (write targets)
--- @field errors string[]|nil -- Parse-time errors from scan_source (atom_info body malformed)
--- @class DebugSkipMarker -- Sub-shape of scan_source.lua's @class DebugSkipMarker
--- @field marker_kind string -- Exact marker ident read from source. Only "atom_dbg_skip" (bare) is positive.
@@ -74,7 +72,7 @@ local duffle = dofile(_bootstrap_dir .. "../duffle_paths.lua")
--- @field info Finding[]
--- @class PipeCtx
--- @field atom_index table<string, AtomAnnotation> -- Name -> AtomAnnotation (only kind=="atom")
--- @field atom_index table<string, AtomEntry> -- raw_name or name -> scan.atoms row (kind atom/atom_proc)
--- @field binds_index table<string, BindsStruct> -- Name -> BindsStruct
--- @field annot_counts table<string, integer> -- Name -> annotation count (for unique_annotation check)
--- @field types table<string, RegTypeDefault> -- From scan_source
@@ -99,14 +97,14 @@ local duffle = dofile(_bootstrap_dir .. "../duffle_paths.lua")
--- `macro_word_drift` writes errors[] for missing or mismatched metadata and info[] for a match.
--- Check: Every annotated atom must have a matching MipsAtom_(name) declaration.
--- @param a AtomAnnotation
--- @param info AtomAnnotation
--- @param pipe_ctx PipeCtx
--- @param findings Findings
local function check_atom_decl_exists(a, pipe_ctx, findings)
if not pipe_ctx.atom_index[a.name] then
local function check_atom_decl_exists(info, pipe_ctx, findings)
if not pipe_ctx.atom_index[info.atom_name] then
findings.errors[#findings.errors + 1] = {
line = a.line,
msg = string.format("annotation for '%s' has no matching MipsAtom_(%s) { ... }", a.name, a.name),
line = info.info_line,
msg = string.format("annotation for '%s' has no matching MipsAtom_(%s) { ... }", info.atom_name, info.atom_name),
}
end
end
@@ -128,17 +126,17 @@ end
--- Check: BIND atoms must reference a real Binds_* struct.
--- I keep this as a warning so the annotation pass can report the common test-fixture case; `check_abi_handoff` in static analysis supplies the build-stopping error.
--- @param a AtomAnnotation
--- @param info AtomAnnotation
--- @param pipe_ctx PipeCtx
--- @param findings Findings
local function check_binds_struct_exists(a, pipe_ctx, findings)
if not a.binds then return end
if pipe_ctx.binds_index[a.binds] then return end
local function check_binds_struct_exists(info, pipe_ctx, findings)
if not info.binds then return end
if pipe_ctx.binds_index[info.binds] then return end
findings.warnings[#findings.warnings + 1] = {
line = a.line,
line = info.info_line,
msg = string.format("'%s' binds '%s' but no Struct_(%s) { ... } "
.. "declaration found (also flagged as an error by check_abi_handoff in the static-analysis pass)"
, a.name, a.binds, a.binds),
, info.atom_name, info.binds, info.binds),
}
end
@@ -315,7 +313,7 @@ end
--- 6. unsupported target_kind -> marker precedes an unrelated declaration
--- Valid markers stamp `debug_skip` on whole-atom, bare-component, and proc-component declaration records in scan_source.lua.
--- @param marker DebugSkipMarker
--- @param _pipe_ctx PipeCtx -- Unused; kept for consistency with per_annot // TODO(Ed): Remove?
--- @param _pipe_ctx PipeCtx -- Unused; kept for consistency with per_annot
--- @param findings Findings
local function check_skip_marker(marker, _pipe_ctx, findings)
local kind = marker.marker_kind
@@ -400,7 +398,7 @@ end
-- ════════════════════════════════════════════════════════════════════════════
--
-- Each rule entry picks one of four "shapes" of dispatch:
-- per_annot(annot, pipe_ctx, findings) -- runs once per AtomAnnotation
-- per_annot(info, pipe_ctx, findings) -- runs once per scan.atom_infos row
-- post(pipe_ctx, findings) -- runs once after all per_annot calls complete (full-corpus aggregation)
-- per_macro(macro, wc, findings) -- runs once per TAPE_WORDS / _Pragma macro declaration
-- per_skip_marker(marker, pipe_ctx, findings) -- runs once per src.scan.debug_skip_markers entry
@@ -430,7 +428,7 @@ local CHECK_RULES = {
--- @param ctx PassCtx
--- @return PipeCtx
local function build_corpus_pipe_ctx(ctx)
local view = duffle.corpus_view(ctx)
local view = duffle.corpus_view(ctx)
local annot_counts = {}
for _, info in ipairs(view.atom_infos) do
if info and info.atom_name then
@@ -452,29 +450,6 @@ local function validate(ctx, src, corpus_pipe_ctx)
corpus_pipe_ctx = corpus_pipe_ctx or build_corpus_pipe_ctx(ctx)
local scan = src.scan
-- Project the pre-scanned atoms to the AtomEntry shape this pass needs.
local atoms = {}
for _, a in ipairs(scan.atoms) do
if a.kind == "atom" or a.kind == "atom_proc" then
atoms[#atoms + 1] = { line = a.line, name = a.raw_name or a.name }
end
end
-- Project the pre-scanned atom_infos to AtomAnnotation shape.
local annots = {}
for _, info in ipairs(scan.atom_infos) do
annots[#annots + 1] = {
line = info.info_line,
macro = "atom_info",
name = info.atom_name,
kind = "info",
binds = info.binds,
reads = info.reads or {},
writes = info.writes or {},
errors = info.errors,
}
end
-- Build a per-source pipe_ctx: shared lookups come from `corpus_pipe_ctx`, while declarations, bodies, types, views, defaults, and occurrences come from `src.scan`.
local seen_defaults = {}; for reg, _ in pairs (scan.types or {}) do seen_defaults[reg] = (seen_defaults[reg] or 0) + 1 end
local atom_infos_list = {}; for _, ai in ipairs(scan.atom_infos or {}) do atom_infos_list[#atom_infos_list + 1] = ai end
@@ -493,7 +468,13 @@ local function validate(ctx, src, corpus_pipe_ctx)
register_alias_registry = corpus_pipe_ctx.register_alias_registry,
type_name_registry = corpus_pipe_ctx.type_name_registry,
}
for _, a in ipairs(atoms) do pipe_ctx.atom_index [a.name] = a end
local atoms = {}
for _, a in ipairs(scan.atoms) do
if a.kind == "atom" or a.kind == "atom_proc" then
atoms[#atoms + 1] = a
pipe_ctx.atom_index[a.raw_name or a.name] = a
end
end
for _, b in ipairs(scan.binds) do pipe_ctx.binds_index[b.name] = b end
-- Findings live in a single struct with three lists (errors / warnings / info).
@@ -501,20 +482,20 @@ local function validate(ctx, src, corpus_pipe_ctx)
local findings = { errors = {}, warnings = {}, info = {} }
-- Lift parse-time errors already recorded in scan_source's atom_info payload into this pass's findings list.
for _, a in ipairs(annots) do
if a.errors then
for _, msg in ipairs(a.errors) do
for _, info in ipairs(scan.atom_infos) do
if info.errors then
for _, msg in ipairs(info.errors) do
findings.errors[#findings.errors + 1] = {
line = a.line,
msg = string.format("'%s': %s", a.name, msg),
line = info.info_line,
msg = string.format("'%s': %s", info.atom_name, msg),
}
end
end
end
-- THE per-annotation pipeline. ONE loop. CHECK_RULES dispatches per_annot rules.
for _, a in ipairs(annots) do
duffle.run_check_rules(CHECK_RULES, "per_annot", a, pipe_ctx, findings)
for _, info in ipairs(scan.atom_infos) do
duffle.run_check_rules(CHECK_RULES, "per_annot", info, pipe_ctx, findings)
end
-- Post-loop rules (one-shot checks that need full-corpus aggregation in pipe_ctx).
@@ -541,12 +522,12 @@ local function validate(ctx, src, corpus_pipe_ctx)
findings.info[#findings.info + 1] = {
line = 0,
msg = string.format("scanned: %d atom(s), %d annotation(s), %d macro-word-decl(s), %d binds struct(s)"
, #atoms, #annots, #scan.macros, #scan.binds),
, #atoms, #scan.atom_infos, #scan.macros, #scan.binds),
}
return {
atoms = atoms,
annots = annots,
annots = scan.atom_infos,
macros = scan.macros,
binds = scan.binds,
errors = findings.errors,
@@ -576,7 +557,7 @@ function M.run(ctx)
-- Build the shared pipe_ctx once for this run; every validate() call sees the same cross-source registries.
-- The corpus owns the canonical cross-source registries; per-source scans retain body / declaration ownership.
local corpus_pipe_ctx = build_corpus_pipe_ctx(ctx)
local corpus = ctx.shared.corpus
local corpus = ctx.shared.corpus
-- Group `corpus.sources_by_dir` by module, validate every source in each bucket, and emit one errors.h per directory.
local by_dir = (corpus and corpus.sources_by_dir) or {}
scripts/passes/atoms_source_map.lua
+3 -5
View File
@@ -526,13 +526,11 @@ function M.render_atom_provenance(atom, wc, rel_path)
local lines = {}
lines[#lines + 1] = string.format("ATOM %s %d", (atom.raw_name or atom.name), total)
for _, entry in ipairs(entries) do
local inv = entry.invocation
local inv = entry.invocation
local macro_count = inv and wc and wc["mac_" .. inv.component_name]
if inv and macro_count ~= nil then
lines[#lines + 1] = string.format(
'WORD %d CALL %s:%d MACRO %s "%s:%d" BODY %d',
entry.pos, rel_path, entry.line, inv.component_name,
inv.def_path or "", inv.def_line or 0, entry.body_line)
lines[#lines + 1] = string.format('WORD %d CALL %s:%d MACRO %s "%s:%d" BODY %d'
, entry.pos, rel_path, entry.line, inv.component_name, inv.def_path or "", inv.def_line or 0, entry.body_line)
else
lines[#lines + 1] = string.format(
"WORD %d CALL %s:%d RAW", entry.pos, rel_path, entry.line)
scripts/passes/auto_reg.lua
+26 -32
View File
@@ -9,8 +9,7 @@
--- These GPRs are unavailable to EVERY atom's source pool.
--- Carriers are preserved across atoms by context discipline and must never be reallocated.
--- Per-atom body parsing also catches alias references (R_<Alias>) and hardcoded R_Tn references,
--- so the user can write either `R_T4` or `R_ResolveScratch` in an atom body and the pass will
--- exclude R_T4 from that atom's pool.
--- so the user can write either `R_T4` or `R_ResolveScratch` in an atom body and the pass will exclude R_T4 from that atom's pool.
---
--- Conflict detection: If the user hardcodes `R_Tn` in an atom body that shares a phase with an auto-reg that picked `R_Tn`,
--- emit `phase_register_clash` as an info finding (no build stop).
@@ -27,30 +26,26 @@
local _bootstrap_dir = debug.getinfo(1, "S").source:match("^@?(.*[/\\])") or "./"
local duffle = dofile(_bootstrap_dir .. "../duffle_paths.lua")
-- ════════════════════════════════════════════════════════════════════════════
-- THE GPR ALLOCATION POOL — what's allocatable, and (more importantly) WHY
-- ════════════════════════════════════════════════════════════════════════════
--
-- The auto-reg pass picks physical GPRs for `atom_auto_reg(...)` / `phase_auto_reg(...)` markers.
-- The 24-register pool covers R2-R25 (the user/atom allocatable surface):
-- R_T0..R_T7, R_V0..R_V1, R_A0..A3, R_S0..S7, R_T8..T9.
-- Excluded (and never added to the pool):
-- R_0 (code 0) — hardwired zero. Cannot be written.
-- R_AT (code 1) — assembler temporary. Reserved by the MIPS O32 ABI.
-- R_A0..A3 — explicitly omitted above even though their integer codes
-- map to POOL entries; the pool-construction loop below
-- only references the POOL string literals, never the
-- integer codes, so they are NOT auto-allocated by default.
-- (A0-A3 become available when the user adds them to
-- POOL or hardcodes an R_A0 reference in the atom body.)
-- R_K0/K1 (codes 26-27) — kernel / interrupt handler reserves. Never touched by user code.
-- R_GP/SP/FP/RA (codes 28-31) — R_SP/R_FP/R_RA are tape-runtime carriers between
-- tape_enter and tape_exit; R_GP stays the host global pointer.
--- ════════════════════════════════════════════════════════════════════════════
--- THE GPR ALLOCATION POOL — what's allocatable, and (more importantly) WHY
--- ════════════════════════════════════════════════════════════════════════════
---
--- The auto-reg pass picks physical GPRs for `atom_auto_reg(...)` / `phase_auto_reg(...)` markers.
--- The 24-register pool covers R2-R25 (the user/atom allocatable surface):
--- R_T0..R_T7, R_V0..R_V1, R_A0..A3, R_S0..S7, R_T8..T9.
--- Excluded (and never added to the pool):
--- R_0 (code 0) — Hardwired zero. Cannot be written.
--- R_AT (code 1) — Assembler temporary. Reserved by the MIPS O32 ABI.
--- R_A0..A3 — Explicitly omitted above even though their integer codes map to POOL entries;
--- the pool-construction loop below only references the POOL string literals, never the integer codes, so they are NOT auto-allocated by default.
--- (A0-A3 become available when the user adds them to POOL or hardcodes an R_A0 reference in the atom body.)
--- R_K0/K1 (codes 26-27) — Kernel / interrupt handler reserves. Never touched by user code.
--- R_GP/SP/FP/RA (codes 28-31) — R_SP/R_FP/R_RA are tape-runtime carriers between tape_enter and tape_exit; R_GP stays the host global pointer.
---
local POOL = {
"R_V0", "R_V1",
"R_T0", "R_T1", "R_T2", "R_T3",
"R_T4", "R_T5", "R_T6", "R_T7",
"R_V0", "R_V1",
"R_A0", "R_A1", "R_A2", "R_A3",
"R_S0", "R_S1", "R_S2", "R_S3",
"R_S4", "R_S5", "R_S6", "R_S7",
@@ -131,13 +126,13 @@ local function build_user_pins(corpus)
return user_pinned, alias_to_gpr
end
-- Find every physical GPR referenced in the atom body, via EITHER:
-- (a) A hardcoded physical GPR ident (R_T\d+|R_V\d+|R_A\d+|R_S\d+) — the existing regex;
-- (b) An alias ident (R_<Alias>) resolved via alias_to_gpr back to its physical GPR ident.
-- Returns { [physical_gpr_ident] = count }. Clash-detection and source-pool-exclusion logic
-- only needs the presence of each GPR (boolean test), but keeping count preserves the
-- original find_hardcoded_rn shape so callers can switch without churn.
-- The alias pattern is sorted lexicographically to keep the regex deterministic.
--- Find every physical GPR referenced in the atom body, via EITHER:
--- (a) A hardcoded physical GPR ident (R_T\d+|R_V\d+|R_A\d+|R_S\d+) — the existing regex;
--- (b) An alias ident (R_<Alias>) resolved via alias_to_gpr back to its physical GPR ident.
--- Returns { [physical_gpr_ident] = count }.
--- Clash-detection and source-pool-exclusion logic only needs the presence of each GPR (boolean test),
--- but keeping count preserves the original find_hardcoded_rn shape so callers can switch without churn.
--- The alias pattern is sorted lexicographically to keep the regex deterministic.
local function find_used_gprs(body_text, alias_to_gpr)
local found = {}
-- (a) Hardcoded physical GPRs (R_T0..R_T7, R_V0..R_V1, R_A0..R_A3, R_S0..R_S7).
@@ -265,8 +260,7 @@ function M.run(ctx)
end
local source_pool = {}
for _, gpr in ipairs(POOL) do
-- Exclude (a) prior commitments, (b) USER-PINNED GPRs (wave-context carriers
-- declared via atom_reg + _Code defs, preserved across atoms globally).
-- Exclude (a) prior commitments, (b) USER-PINNED GPRs (wave-context carriers declared via atom_reg + _Code defs, preserved across atoms globally).
if not used[gpr] and not user_pinned[gpr] then
source_pool[#source_pool + 1] = gpr
end
@@ -297,7 +291,7 @@ function M.run(ctx)
-- For each resolved (scope, sym) -> R_Tn mapping, scan the atom body source for used GPRs.
for atom_scope, decls in pairs(atom_allocations) do
local atom = corpus.atoms_by_name and corpus.atoms_by_name[atom_scope]
if atom and atom.body then
if atom and atom.body then
local used_in_body = find_used_gprs(atom.body, alias_to_gpr)
for sym, allocated_gpr in pairs(decls) do
if used_in_body[allocated_gpr] and used_in_body[allocated_gpr] > 0 then
scripts/passes/components.lua
+72 -107
View File
@@ -100,9 +100,8 @@ local M = {}
--- Returns the args string (e.g., `"U4 off, U4 code, U1 r, U1 g, U1 b"`) or nil if no function declaration is found.
---
--- After the `sym` arg was dropped from MipsAtomComp_Proc_, the component name
--- and the args both come from the preceding `FI_ Slice_MipsCode ac_X(args)`
--- declaration. The shared `duffle.find_function_decl_for` helper does the
--- backward walk; this function returns just the args.
--- and the args both come from the preceding `FI_ Slice_MipsCode ac_X(args)` declaration.
--- The shared `duffle.find_function_decl_for` helper does the backward walk; this function returns just the args.
---
--- @param source string
--- @param name string (retained for signature stability; unused — the walk derives the name)
@@ -336,11 +335,11 @@ end
--- @param tok string
--- @return string
local function strip_leading_delay_marker(tok)
local ident = duffle.read_ident(tok, 1)
local ident = duffle.read_ident(tok, 1)
if not ident or not duffle.DELAY_MARKERS[ident] then return tok end
local rest = tok:sub(#ident + 1):match("^%s*(.*)$") or ""
while rest:sub(1, 2) == "/*" do
local close = rest:find("*/", 3, true)
local close = rest:find("*/", 3, true)
if not close then return "" end
rest = rest:sub(close + 2):match("^%s*(.*)$") or ""
end
@@ -428,115 +427,87 @@ end
-- Always walk the original `MipsAtomComp_` body via `cc.body_tokens`.
-- ═══════════════════════════════════════════
--- (internal) Recursive cycle-cost derivation. Sum `latency[ident]` per emitted instruction in the component body,
--- recursing through nested `mac_*` calls (so `mac_format_g4_color`'s cost = 4 × `mac_pack_color_word`'s cost).
--- Special rule: `mac_yield`'s cost = 0 (per `lottes_tape.h:125-130` "the runtime cost lands in the next atom's prologue").
--- (internal) One walk of a component body that fills both `cycle_cost` and `gp0_contrib`.
--- Cycle: sum `isa.cycles` / `gte.cycles` / `latency[ident]` / 1 per leaf, recurse `mac_*`.
--- `mac_yield` cycle_cost is 0 (runtime cost lands in the next atom's prologue); its gp0 still comes from the token walk.
--- GP0: count `gte_sw` and `store_word` / `store_half` / `store_byte` that target `R_PrimCursor` / `O_(Poly_` / `r_prim_cursor` / `r_primitive_cursor` / `r_base`.
--- `insert_ot_tag*` gp0_contrib is 0; cycle still comes from the body walk.
--- Missing component: cycle 1, gp0 0.
--- @param name string -- component bare name (e.g. "yield", "pack_color_word")
--- @param comp_by_name table<string, Component>
--- @param latency table<string, integer>
--- @param cache table<string, integer> -- shared memoization; `-1` sentinel detects cycles
--- @return integer
local function cycle_cost_rec(name, comp_by_name, latency, cache)
--- @param cache table<string, {cycle_cost=integer, gp0_contrib=integer}>
--- @return {cycle_cost=integer, gp0_contrib=integer}
local function component_meta_rec(name, comp_by_name, latency, cache)
if cache[name] ~= nil then return cache[name] end
cache[name] = -1
cache[name] = { cycle_cost = -1, gp0_contrib = -1 }
local cc = comp_by_name[name]
local n
local cycle_cost
local gp0_contrib
if cc then
if name == "yield" then
-- mac_yield's cost is 0 by convention (the runtime cost lands in the next atom's prologue).
n = 0
else
n = 0
local skip_cycle = (name == "yield")
local skip_gp0 = name:match("^insert_ot_tag") ~= nil
cycle_cost = 0
gp0_contrib = 0
if not skip_cycle or not skip_gp0 then
local tokens = cc.body_tokens
for _, t in ipairs(tokens) do
local trimmed = t.tok
if trimmed ~= "" then
local ident = duffle.read_ident(trimmed, 1)
if ident and ident:sub(1, MAC_PREFIX_LEN) == MAC_PREFIX then
-- Nested `mac_X(...)` call: recurse.
if ident and ident:sub(1, MAC_PREFIX_LEN) == MAC_PREFIX then
local nested = ident:sub(MAC_PREFIX_LEN + 1)
n = n + cycle_cost_rec(nested, comp_by_name, latency, cache)
local nested_meta = component_meta_rec(nested, comp_by_name, latency, cache)
if not skip_cycle then
cycle_cost = cycle_cost + nested_meta.cycle_cost
end
if not skip_gp0 then
gp0_contrib = gp0_contrib + nested_meta.gp0_contrib
end
else
-- Leaf instruction or pseudo-macro.
local isa = duffle.instr(ident)
local gte = duffle.gte(ident)
n = n + ((isa and isa.cycles) or (gte and gte.cycles) or latency[ident] or 1)
if not skip_cycle then
local isa = duffle.instr(ident)
local gte = duffle.gte(ident)
cycle_cost = cycle_cost + ((isa and isa.cycles) or (gte and gte.cycles) or latency[ident] or 1)
end
if not skip_gp0 then
if ident == "gte_sw" then
gp0_contrib = gp0_contrib + 1
elseif ident == "store_word" or ident == "store_half" or ident == "store_byte" then
if trimmed:find("R_PrimCursor", 1, true)
or trimmed:find("O_(Poly_", 1, true)
or trimmed:find("r_prim_cursor", 1, true)
or trimmed:find("r_primitive_cursor", 1, true)
or trimmed:find("r_base", 1, true)
then
gp0_contrib = gp0_contrib + 1
end
end
end
end
end
end
end
else
n = 1
cycle_cost = 1
gp0_contrib = 0
end
cache[name] = n
return n
end
--- (internal) Recursive GP0 prim-buffer contribution. Count `store_word` / `store_half` / `store_byte`
--- calls in the component body that target `R_PrimCursor` (these are the RAM-side prim-buffer words the macro contributes), recursing through nested `mac_*` calls.
--- Only `R_PrimCursor`-targeting stores count. Stores targeting other registers (e.g. `R_OtBase`, heap pointers) are not prim-buffer contributions.
--- @param name string
--- @param comp_by_name table<string, Component>
--- @param cache table<string, integer>
--- @return integer
local function gp0_contrib_rec(name, comp_by_name, cache)
if name:match("^insert_ot_tag") then
cache[name] = 0
return 0
end
if cache[name] ~= nil then return cache[name] end
cache[name] = -1
local cc = comp_by_name[name]
local n
if cc then
n = 0
local tokens = cc.body_tokens
for _, t in ipairs(tokens) do
local trimmed = t.tok
if trimmed ~= "" then
local ident = duffle.read_ident(trimmed, 1)
if ident and ident:sub(1, MAC_PREFIX_LEN) == MAC_PREFIX then
-- Nested `mac_X(...)` call: recurse.
local nested = ident:sub(MAC_PREFIX_LEN + 1)
n = n + gp0_contrib_rec(nested, comp_by_name, cache)
elseif ident == "gte_sw" then
n = n + 1
elseif ident == "store_word" or ident == "store_half" or ident == "store_byte" then
if trimmed:find("R_PrimCursor", 1, true)
or trimmed:find("O_(Poly_", 1, true)
or trimmed:find("r_prim_cursor", 1, true)
or trimmed:find("r_primitive_cursor", 1, true)
or trimmed:find("r_base", 1, true)
then
n = n + 1
end
end
end
end
else
n = 0
end
cache[name] = n
return n
cache[name] = { cycle_cost = cycle_cost, gp0_contrib = gp0_contrib }
return cache[name]
end
--- Compute `cycle_cost` + `gp0_contrib` for every component in `components` in a single pass.
--- Memoization cache is built ONCE (per source) and shared across both helpers so that
--- a nested `mac_Y` reference inside a `mac_X` body computes its values once.
--- One memoization cache; a nested `mac_Y` inside a `mac_X` body computes both fields once.
--- @param components Component[]
--- @param latency table<string, integer>
--- @return table<string, {cycle_cost=integer, gp0_contrib=integer}>
local function compute_components_metadata(components, latency)
local comp_by_name = {}
for _, cc in ipairs(components) do comp_by_name[cc.name] = cc end
local cc_cache = {}
local gc_cache = {}
local out = {}
local cache = {}
local out = {}
for _, c in ipairs(components) do
out[c.name] = {
cycle_cost = cycle_cost_rec(c.name, comp_by_name, latency, cc_cache),
gp0_contrib = gp0_contrib_rec(c.name, comp_by_name, gc_cache),
}
out[c.name] = component_meta_rec(c.name, comp_by_name, latency, cache)
end
return out
end
@@ -589,29 +560,23 @@ local function strip_trailing_continuation(lines)
end
end
--- Classify a token as a "pure delay marker token" (a delay-marker identifier
--- with no following instruction — only whitespace and/or block comments).
--- Classify a token as a "pure delay marker token" (a delay-marker identifier with no following instruction — only whitespace and/or block comments).
--- Examples that match:
--- * `GteDelay_` → marker alone
--- * `GteDelay_ /* RT diagonal: D1 = a.x... */` → marker + block comment
--- * `GteDelay_ /* RT diagonal: ... */\n\t` → marker + comment + trailing whitespace
--- Examples that DO NOT match (these contain a real instruction after the marker
--- and must be preserved verbatim so the instruction still gets emitted):
--- Examples that DO NOT match (these contain a real instruction after the marker and must be preserved verbatim so the instruction still gets emitted):
--- * `GteDelay_ nop2`
--- * `GteDelay_ add_si(r.dst_ptr, r.scratch, dst_offset)`
---
--- Why this classification matters: the metaprogram emits tokens separated by `,`
--- and joins them with `\<newline>` line continuations. After C preprocessor
--- Why this classification matters: the metaprogram emits tokens separated by `,` and joins them with `\<newline>` line continuations. After C preprocessor
--- phase 2 (line splicing), the macro body collapses to a single logical line.
--- Each delay-marker identifier expands to empty (its definition
--- `#define GteDelay_ // ...` consumes the `//` line comment during preprocessing
--- of the definition itself, leaving an empty replacement list). When a token
--- is purely a delay marker with only a trailing comment, the `,` the metaprogram
--- normally adds before each token-after-the-first brackets empty content and
--- produces the syntax error `,,` (`expected expression before ',' token`) at
--- C compile. The metaprogram therefore emits such tokens WITHOUT the leading
--- `,` (see `token_skips_leading_comma`) — but the marker + trailing comment
--- are still emitted verbatim so the annotation is preserved in `gen/macs.h`.
--- Each delay-marker identifier expands to empty (its definition `#define GteDelay_ // ...` consumes the `//` line comment during preprocessing
--- of the definition itself, leaving an empty replacement list).
--- When a token is purely a delay marker with only a trailing comment, the `,` the metaprogram normally adds before
--- each token-after-the-first brackets empty content and produces the syntax error `,,` (`expected expression before ',' token`) at C compile.
--- The metaprogram therefore emits such tokens WITHOUT the leading `,` (see `token_skips_leading_comma`) —
--- but the marker + trailing comment are still emitted verbatim so the annotation is preserved in `gen/macs.h`.
--- @param tok string -- a single token from split_top_level_commas (already trimmed at the start, may contain trailing whitespace + block comment)
--- @return boolean
local function is_pure_delay_marker_token(tok)
@@ -655,17 +620,14 @@ end
--- followed by whitespace + optional block comment and NOTHING ELSE) expand
--- to empty at C preprocessor time. Emitting them WITHOUT the leading `,`
--- separator that the metaprogram normally adds before each token after the
--- first keeps exactly one `,` between the surrounding real expressions in
--- the spliced macro body:
---
--- first keeps exactly one `,` between the surrounding real expressions in the spliced macro body:
--- * before this rule: `<tok1> ,\t<gdelay> ,\t<tok3>` → after expansion
--- `<tok1> , /* comment */ , <tok3>` → `,,` syntax error.
--- * after this rule: `<tok1> \t<gdelay> ,\t<tok3>` → after expansion
--- `<tok1> /* comment */ , <tok3>` → `<tok1>, <tok3>` — valid.
---
--- Tokens like `GteDelay_ nop2` keep the leading `,` (the marker is followed
--- by a real instruction, so the marker + instruction together need the
--- separator on the LEFT to land between two real expressions).
--- Tokens like `GteDelay_ nop2` keep the leading `,`
--- (the marker is followed by a real instruction, so the marker + instruction together need the separator on the LEFT to land between two real expressions).
--- @param tok string
--- @return boolean -- true if the token needs NO leading `,` separator.
local function token_skips_leading_comma(tok)
@@ -675,7 +637,10 @@ end
--- Emit the `#define mac_X(sig) \<newline>\t<tok1> \<newline>,\t<tok2> ...` block.
--- Converts `//` line comments to `/* */` block comments in each token so they don't break the C macro `\` line continuations.
---
--- Pure delay-marker tokens (`GteDelay_` / `LdSlot_` / `BdSlot_` / `DmaSlot_` with only a trailing block comment, no real instruction) are emitted WITHOUT a leading `,` separator; the annotation IS preserved in the generated header (so the comment + marker remain visible to anyone reading `gen/macs.h`), but the C preprocessor expands the marker to empty, so leaving the `,` separator out is what stops the `,,` syntax error. See `token_skips_leading_comma` for the contract.
--- Pure delay-marker tokens (`GteDelay_` / `LdSlot_` / `BdSlot_` / `DmaSlot_` with only a trailing block comment, no real instruction) are emitted WITHOUT a leading `,` separator;
--- the annotation IS preserved in the generated header
--- (so the comment + marker remain visible to anyone reading `gen/macs.h`), but the C preprocessor expands the marker to empty, so leaving the `,`
--- separator out is what stops the `,,` syntax error. See `token_skips_leading_comma` for the contract.
local function emit_macro_body(lines, c, sig, tokens)
for tok_idx = 1, #tokens do
tokens[tok_idx] = convert_line_comments_to_block(tokens[tok_idx])
scripts/passes/dwarf_injection.lua
+92 -106
View File
@@ -1661,24 +1661,24 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
S.bytes[#S.bytes + 1] = s
S.next_offset = S.next_offset + #s
end
local FORM_WRITERS = {
string = function(emit, v) emit(v .. "\0") end,
data1 = function(emit, v) emit(string.char(v)) end,
udata = function(emit, v) emit(uleb128(v)) end,
addr = function(emit, v) emit(elf_dwarf.write_u32_le(v)) end,
ref4 = function(emit, v) emit(elf_dwarf.write_u32_le(v)) end,
sec_offset = function(emit, v) emit(elf_dwarf.write_u32_le(v)) end,
data2 = function(emit, v) emit(elf_dwarf.write_u16_le(v)) end,
exprloc = function(emit, v) emit(v) end,
}
local function emit_die(schema_name, values)
local row = DIE_SCHEMA[schema_name]
emit(uleb128(row.abbrev))
for _, attr in ipairs(row.attrs) do
local v = values[attr.key]
if attr.form == "string" then
emit(v .. "\0")
elseif attr.form == "data1" then
emit(string.char(v))
elseif attr.form == "udata" then
emit(uleb128(v))
elseif attr.form == "addr" or attr.form == "ref4" or attr.form == "sec_offset" then
emit(elf_dwarf.write_u32_le(v))
elseif attr.form == "data2" then
emit(elf_dwarf.write_u16_le(v))
elseif attr.form == "exprloc" then
emit(v)
end
local w = FORM_WRITERS[attr.form]
if not w then error("emit_die: unknown form " .. tostring(attr.form)) end
w(emit, v)
end
end
local function ref4_of(section_offset)
@@ -1737,55 +1737,35 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
-- then a single DW_TAG_pointer_type pointing at the structure_type. gdb's
-- `print *<var>` then expands the struct and lists its members (x, y, z, w for V4_S2; etc.).
--
-- Hardcoded member-info table for the 6 typed views the prototype uses (V2_S2 / V3_S2 / V4_S2 / V2_S4 / V3_S4 / V4_S4).
-- Each row is {byte_size, members}, where each member is {name, offset, byte_size, type_name}.
-- `type_name` is the base type name from `type_name_registry`; "S2" / "S4" need a signed base_type emit too.
--
-- The table is small + explicit — the prototype principle treats the typed-view struct layout as data, not derived state.
local STRUCT_MEMBER_TABLE = {
-- TODO(Ed): This hardcoding is brittle...
-- TODO(Ed): Better to just have a table for the fundamental types in duffle/dsl.h, we can derive the rest via typedef parsing...
-- 2-element signed short vector (rare; placeholder for future use).
V2_S2 = { byte_size = 4, members = {
{ name = "x", offset = 0, byte_size = 2 },
{ name = "y", offset = 2, byte_size = 2 },
}},
-- 4-element signed short vector (the dominant face-cursor type).
V4_S2 = { byte_size = 8, members = {
{ name = "x", offset = 0, byte_size = 2 },
{ name = "y", offset = 2, byte_size = 2 },
{ name = "z", offset = 4, byte_size = 2 },
{ name = "w", offset = 6, byte_size = 2 },
}},
-- 3-element signed short vector (the floor face-cursor type; includes explicit pad byte to match math.h's S2 pad member).
V3_S2 = { byte_size = 8, members = {
{ name = "x", offset = 0, byte_size = 2 },
{ name = "y", offset = 2, byte_size = 2 },
{ name = "z", offset = 4, byte_size = 2 },
{ name = "pad", offset = 6, byte_size = 2 },
}},
-- 2-element signed int vector.
V2_S4 = { byte_size = 8, members = {
{ name = "x", offset = 0, byte_size = 4 },
{ name = "y", offset = 4, byte_size = 4 },
}},
-- 3-element signed int vector.
V3_S4 = { byte_size = 16, members = {
{ name = "x", offset = 0, byte_size = 4 },
{ name = "y", offset = 4, byte_size = 4 },
{ name = "z", offset = 8, byte_size = 4 },
{ name = "pad", offset = 12, byte_size = 4 },
}},
-- 4-element signed int vector.
V4_S4 = { byte_size = 16, members = {
{ name = "x", offset = 0, byte_size = 4 },
{ name = "y", offset = 4, byte_size = 4 },
{ name = "z", offset = 8, byte_size = 4 },
{ name = "w", offset = 12, byte_size = 4 },
}},
}
local S2_TYPE_BYTE_SIZE = 2 -- S2 = signed 16-bit (see dsl.h)
local S4_TYPE_BYTE_SIZE = 4 -- S4 = signed 32-bit (see dsl.h)
-- Layout for V* / Rect_* / Reg_* / Slice / … comes from corpus.type_name_registry.
-- scan_source parses Struct_() in math.h, math.atom.h, memory.h and fills field offset + byte_size.
-- U1–U4 / S1–S4 / B1–B4 stay authored fundamentals (BUILTIN_BYTE_SIZES + base_type DIEs below).
local type_reg = registries.type_name_registry or {}
local function layout_from_registry(tn)
local entry = type_reg[tn]
if not entry or entry.kind ~= "struct" or not entry.fields or #entry.fields == 0 then
return nil
end
if entry.byte_size == nil then return nil end
local members = {}
for _, f in ipairs(entry.fields) do
if f.offset == nil or f.byte_size == nil then return nil end
members[#members + 1] = {
name = f.name,
offset = f.offset,
byte_size = f.byte_size,
type_name = f.type_name,
pointer_depth = f.pointer_depth or 0,
}
end
return { byte_size = entry.byte_size, members = members }
end
local function encoding_for_type(tn)
if type(tn) == "string" and tn:match("^S[124]$") then return 5 end
return 7
end
local S2_TYPE_BYTE_SIZE = 2
local S4_TYPE_BYTE_SIZE = 4
-- Pre-emit the signed base types (S2, S4) once if any typed view's member needs them.
-- We emit per-typed-view lazily below; build a member-base-type offset cache (idempotent).
@@ -1804,17 +1784,54 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
end
-- Always emit S2 + S4 (the v*_S2 / v*_S4 family base types) once before any typed-view,
-- even if no atom currently declares a v*_S2 field, so future atoms pick them up without rewiring.
ensure_member_base_type("S2", S2_TYPE_BYTE_SIZE, 5) -- DW_ATE_signed = 5
ensure_member_base_type("S2", S2_TYPE_BYTE_SIZE, 5)
ensure_member_base_type("S4", S4_TYPE_BYTE_SIZE, 5)
local type_chain_offsets = {}
local struct_die_offsets = {}
local function emit_struct_layout(tn)
if struct_die_offsets[tn] then return struct_die_offsets[tn] end
local type_info = layout_from_registry(tn)
if not type_info then return nil end
for _, m in ipairs(type_info.members) do
if (m.pointer_depth or 0) == 0 and layout_from_registry(m.type_name) then
emit_struct_layout(m.type_name)
end
end
local struct_offset = next_offset()
struct_die_offsets[tn] = struct_offset
emit_die("structure_type", {
name = tn,
byte_size = type_info.byte_size,
})
for _, m in ipairs(type_info.members) do
local member_type_off
if (m.pointer_depth or 0) > 0 then
member_type_off = type_chain_offsets[m.type_name .. "|" .. m.pointer_depth]
elseif layout_from_registry(m.type_name) then
member_type_off = struct_die_offsets[m.type_name]
else
member_type_off = ensure_member_base_type(
m.type_name, m.byte_size, encoding_for_type(m.type_name))
end
if not member_type_off then
member_type_off = ensure_member_base_type(
m.type_name or "U4", m.byte_size or U4_BYTE_SIZE, 7)
end
emit_die("member", {
name = m.name,
data_member_location = m.offset,
type = ref4_of(member_type_off),
})
end
emit(string.char(DIE_CHILDREN_TERMINATOR))
return struct_offset
end
for _, tn in ipairs(sorted_typed_types) do
if tn ~= "U4" then
local depth = used_typed_views[tn]
local type_info = STRUCT_MEMBER_TABLE[tn]
if not type_info then
-- Unknown typed view: fall back to a generic 4-byte unsigned base_type to keep the wire valid.
-- gdb renders as the typename but `print *ptr` only sees the first 4 bytes.
local struct_offset = emit_struct_layout(tn)
if not struct_offset then
local innermost_offset = next_offset()
emit_die("base_type", {
name = tn,
@@ -1824,40 +1841,12 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
local outermost_offset = next_offset()
emit_die("pointer_type", { type = ref4_of(innermost_offset) })
type_chain_offsets[tn .. "|" .. depth] = outermost_offset
elseif depth == 1 then
local outermost_offset = next_offset()
emit_die("pointer_type", { type = ref4_of(struct_offset) })
type_chain_offsets[tn .. "|" .. depth] = outermost_offset
else
-- Emit a proper structure_type DIE for this typed view.
local struct_offset = next_offset()
emit_die("structure_type", {
name = tn,
byte_size = type_info.byte_size,
})
for _, m in ipairs(type_info.members) do
local member_type_name, member_type_encoding
if m.byte_size == 2 then
member_type_name = "S2"
member_type_encoding = 5
elseif m.byte_size == 4 then
member_type_name = "S4"
member_type_encoding = 5
else
member_type_name = "U4"
member_type_encoding = 7
end
local member_base_off = ensure_member_base_type(member_type_name, m.byte_size, member_type_encoding)
emit_die("member", {
name = m.name,
data_member_location = m.offset,
type = ref4_of(member_base_off),
})
end
emit(string.char(DIE_CHILDREN_TERMINATOR))
if depth == 1 then
local outermost_offset = next_offset()
emit_die("pointer_type", { type = ref4_of(struct_offset) })
type_chain_offsets[tn .. "|" .. depth] = outermost_offset
else
error("typed-view: pointer_depth > 1 is not yet supported in this emission path")
end
error("typed-view: pointer_depth > 1 is not yet supported in this emission path")
end
end
end
@@ -1991,9 +1980,6 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
end
end
local atom_view = (registries.atom_views or {})[atom.name]
-- Build the atom-name lookup table once (cheap; O(atom_table)) so step (b) and step (d) can resolve rbind_atom names.
-- TODO(Ed): Bad assignment?
local atom_by_name = atom_by_name or (function() local m = {}; for _, a in ipairs(atom_table) do if a.name then m[a.name] = a end end; return m end)()
-- step (b) inputs: this atom's `atom_ctx(<rbind_atom>)` (resolved from the registries' atom_ctxs)
local this_ctx = registries.atom_ctxs and registries.atom_ctxs[atom.name]
if this_ctx and this_ctx.rbind_atom then
@@ -2073,9 +2059,9 @@ local function build_inserted_children(main_cu_offset, main_cu_end_excl, atom_ta
end,
-- (e) enum-site atom_type(<T>) default on the registry entry.
function(r_name, alias_code)
-- TODO(Ed): Bad definition?
if alias and alias.default_type and alias.default_depth and alias.default_depth > 0 then
return type_chain_offsets[alias.default_type .. "|" .. alias.default_depth]
local a = by_alias[r_name]
if a and a.default_type and a.default_depth and a.default_depth > 0 then
return type_chain_offsets[a.default_type .. "|" .. a.default_depth]
end
end,
}
scripts/passes/offsets.lua
+8 -12
View File
@@ -9,13 +9,6 @@
--- Per-directory aggregation: every source in the same directory contributes to the same `gen/offsets.h`.
--- The directory itself is the namespace; the filename does not repeat the module name.
---
--- (Task 12.16 note: atom-namespaced enum names — e.g., `atom_offset__normalize_v3s4__srav_path__aligned_done` —
--- were considered to prevent cross-atom label collisions, but the C-side `atom_offset(F, T)` macro in
--- `code/duffle/dsl.atom.h` doesn't know the current atom_name at expansion time, so any namespacing
--- on the metaprogram side breaks the C build. Reverted. The C-side would need a per-atom
--- `CURRENT_ATOM` #define (set by `MipsAtom_`/`MipsAtom_Proc_` macros) plus an updated `atom_offset`
--- macro that uses it. That's a coordinated refactor — deferred to a future track.)
---
--- The offset is `target_word - branch_word - 1` (the standard MIPS branch-immediate encoding: branch_offset = relative_pc_in_words - 1).
-- ════════════════════════════════════════════════════════════════════════════
@@ -124,9 +117,7 @@ end
--- For cross-module `j`/`jal` (atom body in one module, target in another), the linker emits a `R_MIPS_26` relocation against the lower 26 bits; the upper 4 bits come from the PC of the delay slot following the `j`.
--- The metaprogram doesn't know either at compile time, so the emitted value is the relative word offset that the duffle `enc_i` macro places in the immediate field; the toolchain handles the rest.
--- `jump_reg` / `call_reg` / `jump_link` -> ERROR. Register-form jumps have no offset field; `atom_offset` is invalid.
---
--- Top-level `atom_offset(F, T)` markers (where the marker is the entire token — `consuming_encoder` == nil) default to `branch_*` behavior (relative offset).
--- This preserves backward compatibility for any top-level marker that may exist outside a control-transfer instruction.
--- missing `consuming_encoder` -> ERROR. A lone top-level `atom_offset` is not a branch.
--- @param labels table<string, integer>
--- @param branches table[]
--- @param errors table[]
@@ -142,14 +133,19 @@ local function compute_offsets(labels, branches, errors)
}
else
local consuming = br.consuming_encoder
if consuming == "jump_reg" or consuming == "call_reg" or consuming == "jump_link" then
if consuming == nil or consuming == "" then
errors[#errors + 1] = {
line = br.line or 0,
msg = "atom_offset requires a consuming encoder (branch_*, jump, call_addr); top-level atom_offset is invalid; at word " .. br.branch_word,
}
elseif consuming == "jump_reg" or consuming == "call_reg" or consuming == "jump_link" then
errors[#errors + 1] = {
line = br.line or 0,
msg = "atom_offset cannot be used with " .. consuming
.. " (register-form jumps have no offset field); at word " .. br.branch_word,
}
else
-- All other consuming instructions (including `branch_*`, `jump`, `call_addr`, and nil for top-level markers) use the same relative offset value.
-- Consuming instructions with an offset field (`branch_*`, `jump`, `call_addr`) use the same relative offset value.
-- The MIPS encoding differs per opcode but the duffle `enc_i` macro handles the truncation to the immediate-field width.
results[#results + 1] = {
target = br.target,
scripts/passes/report.lua
+4 -4
View File
@@ -1,8 +1,9 @@
--- passes/report.lua — Per-MODULE annotation report renderer + project-wide summary writer.
---
--- Two output files per build:
--- - `build/gen/<dir_basename>.annotations.txt` — one per source-directory containing atoms; aggregates across all sources in the directory.
--- - `build/gen/annotation_validation.txt` — the project summary.
--- Per-module markdown plus one project summary:
--- - `build/<dir_basename>.atom_meta_report.md` — one per source-directory containing atoms
--- - `build/<dir_basename>.atoms.md` — verbose source map
--- - `build/atom_meta_report.summary.md` — the project summary
---
--- The canonical `corpus.sources_by_dir` projection groups sources by directory.
--- This pass builds one ModuleView per directory and walks SECTION_RENDERERS.
@@ -305,7 +306,6 @@ local function build_module_view(dir, dir_sources, corpus)
decls = decls,
schemas = schemas,
findings = sa.findings or {},
sa = sa,
corpus = corpus,
}
end
scripts/passes/scan_source.lua
+290 -193
View File
@@ -179,7 +179,7 @@ end
local function read_parens_after(source, ident_end, fallback)
local open_paren = duffle.skip_ws_and_cmt(source, ident_end)
if source:sub(open_paren, open_paren) ~= "(" then return nil, fallback or open_paren + 1 end
local inner, after_paren = duffle.read_parens(source, open_paren)
local inner, after_paren = duffle.read_parens(source, open_paren)
return inner, after_paren, open_paren
end
@@ -189,7 +189,7 @@ end
local function find_body_braces(source, after_paren, fallback)
local brace = duffle.scan_to_char(source, "{", after_paren)
if not brace then return nil, fallback or (after_paren + 1) end
local body, after_brace = duffle.read_braces(source, brace)
local body, after_brace = duffle.read_braces(source, brace)
return body, after_brace, brace + 1
end
@@ -249,7 +249,7 @@ local function preceding_comment_walk_backward(source, start_pos)
line_start = line_start - 1
end
local line = source:sub(line_start, non_ws)
if line:sub(1, 2) ~= "//" then break end
if line:sub(1, 2) ~= "//" then break end
table.insert(pieces, 1, line)
scan_pos = line_start - 1
end
@@ -303,8 +303,8 @@ local function register_atom(out, kind, declaration_line, name, body, body_off,
-- Capture the pending marker BEFORE attaching so the walker can anchor the backward comment walk on the marker's marker_pos
-- (which is the correct anchor even when an `FI_ MipsAtom ac_X(args)` proc-prelude separates the marker from the declaration).
local pending_marker = nil
local markers = out.debug_skip_markers
local m = markers[#markers]
local markers = out.debug_skip_markers
local m = markers[#markers]
if m and m.pending then pending_marker = m end
local positive = attach_debug_skip_marker(out, kind)
@@ -314,7 +314,7 @@ local function register_atom(out, kind, declaration_line, name, body, body_off,
-- The walker does not need to detect marker shape.
-- A pending_marker record (or the declaration ident_pos fallback) supplies the anchor position.
local start_pos = comment_walk_start(pending_marker, pos)
comment = preceding_comment_walk_backward(source, start_pos)
comment = preceding_comment_walk_backward(source, start_pos)
end
out.atoms[#out.atoms + 1] = {
line = declaration_line,
@@ -335,7 +335,7 @@ end
local function register_raw_atom(out, declaration_line, name, body, body_off, raw_name, pos)
out.raw_atoms[#out.raw_atoms + 1] = {
line = declaration_line, name = name, body = body, body_off = body_off,
kind = "raw_atom", raw_name = raw_name,
kind = "raw_atom", raw_name = raw_name,
}
end
@@ -344,7 +344,7 @@ end
local function parse_type_chain(text, pos)
if pos > #text then return nil end
-- Skip leading whitespace before the type ident.
local start = duffle.skip_ws_and_cmt(text, pos)
local start = duffle.skip_ws_and_cmt(text, pos)
local ident, after = duffle.read_ident(text, start)
if not ident then return nil end
local depth = 0
@@ -367,6 +367,9 @@ local BUILTIN_BYTE_SIZES = {
["S1"] = 1,
["S2"] = 2,
["S4"] = 4,
["B1"] = 1,
["B2"] = 2,
["B4"] = 4,
-- GCC __UINT*/__INT*_TYPE__ family (used by the duffle TSet_ convention in dsl.h).
-- MIPS32 has no 64-bit types; __UINT64_TYPE__/__INT64_TYPE__ are excluded.
["__UINT8_TYPE__"] = 1,
@@ -422,7 +425,7 @@ end
-- Returns the raw fields array with `{name, type_name, pointer_depth}` only (NO offset / byte_size).
-- The propagation pass `resolve_struct_field_sizes` walks each struct's fields AFTER type resolution and populates offset + byte_size in place.
local function parse_struct_body_fields(body)
local fields = {}
local fields = {}
local body_pos = 1
local body_len = #body
while body_pos <= body_len do
@@ -563,6 +566,16 @@ local function propagate_type_sizes(out)
for _ = 1, TYPE_CHAIN_MAX_DEPTH do
local any_change = false
for _, entry in pairs(reg) do
if entry.kind == "array" and entry.byte_size == nil and entry.counts then
local elem_size = BUILTIN_BYTE_SIZES[entry.elem]
or (reg[entry.elem] and reg[entry.elem].byte_size)
if elem_size then
local n = 1
for _, c in ipairs(entry.counts) do n = n * c end
entry.byte_size = elem_size * n
any_change = true
end
end
if entry.kind == "struct" and entry.fields then
local byte_off = 0
local gap_seen = false
@@ -710,7 +723,7 @@ local function scan_atom_info_subcalls(info_inner, info_line)
-- reads/writes arrays contain ONLY register idents;
-- `atom_type(...)` sub-entry (when present and well-formed) is recorded as a per-atom reg_type_override.
local entries = duffle.split_top_level_commas(sub_inner)
local regs = {}
local regs = {}
for _, entry in ipairs(entries) do
local reg_name, override, malformed = parse_atom_info_reg_entry(entry)
if reg_name then
@@ -742,9 +755,9 @@ local function scan_atom_info_subcalls(info_inner, info_line)
end
end
local function reg_types_handler(sub_inner, info_line)
local args = duffle.split_top_level_commas(sub_inner)
local args = duffle.split_top_level_commas(sub_inner)
if not args[1] then return end
reg_overrides = reg_overrides or {}
reg_overrides = reg_overrides or {}
local reg_name = duffle.trim(args[1])
local type_name, depth = nil, 0
if args[2] then
@@ -763,13 +776,13 @@ local function scan_atom_info_subcalls(info_inner, info_line)
}
end
local SUBCALL_HANDLERS = {
atom_bind = function(sub_inner) binds = duffle.trim(sub_inner) end, -- scan: atom_bind(<Binds_X>)
atom_reads = function(sub_inner, info_line) rw_handler(sub_inner, info_line, "atom_reads") end, -- scan: atom_reads(<R_X [atom_type(<T>)], ...>)
atom_writes = function(sub_inner, info_line) rw_handler(sub_inner, info_line, "atom_writes") end, -- scan: atom_writes(<R_X [atom_type(<T>)], ...>)
atom_view = function(sub_inner) view_binds = duffle.trim(sub_inner) end, -- scan: atom_view(<Binds_X>)
atom_reg_types = reg_types_handler, -- scan: atom_reg_types(<R_X>, <T>)
atom_ctx = function(sub_inner, info_line) ident_handler(sub_inner, info_line, "ctx_atom_name") end, -- scan: atom_ctx(<atom_name>)
atom_phase = function(sub_inner, info_line) ident_handler(sub_inner, info_line, "phase_label") end, -- scan: atom_phase(<label>)
atom_bind = function(sub_inner) binds = duffle.trim(sub_inner) end, -- scan: atom_bind(<Binds_X>)
atom_reads = function(sub_inner, info_line) rw_handler(sub_inner, info_line, "atom_reads") end, -- scan: atom_reads(<R_X [atom_type(<T>)], ...>)
atom_writes = function(sub_inner, info_line) rw_handler(sub_inner, info_line, "atom_writes") end, -- scan: atom_writes(<R_X [atom_type(<T>)], ...>)
atom_view = function(sub_inner) view_binds = duffle.trim(sub_inner) end, -- scan: atom_view(<Binds_X>)
atom_reg_types = reg_types_handler, -- scan: atom_reg_types(<R_X>, <T>)
atom_ctx = function(sub_inner, info_line) ident_handler(sub_inner, info_line, "ctx_atom_name") end, -- scan: atom_ctx(<atom_name>)
atom_phase = function(sub_inner, info_line) ident_handler(sub_inner, info_line, "phase_label") end, -- scan: atom_phase(<label>)
}
local sub_pos = 1
@@ -783,7 +796,7 @@ local function scan_atom_info_subcalls(info_inner, info_line)
local sub_open = duffle.skip_ws_and_cmt(info_inner, sub_end)
if info_inner:sub(sub_open, sub_open) == "(" then
local sub_inner, sub_after2 = duffle.read_parens(info_inner, sub_open)
local handler = SUBCALL_HANDLERS[sub_ident]
local handler = SUBCALL_HANDLERS[sub_ident]
if handler then handler(sub_inner, info_line) end
sub_pos = sub_after2
else
@@ -798,7 +811,7 @@ end
local function scan_skip_qualifiers(source, pos)
while true do
pos = duffle.skip_ws_and_cmt(source, pos)
local ident, after = duffle.read_ident(source, pos)
local ident, after = duffle.read_ident(source, pos)
if not ident then return pos end
if QUALIFIER_KEYWORDS[ident] then pos = after else return pos end
end
@@ -870,11 +883,11 @@ local function read_trailing_cmt_after(body, pos)
local body_len = #body
while pos <= body_len do
local b = body:byte(pos)
if b == BYTE_SPACE or b == BYTE_TAB or b == BYTE_NEWLINE or b == BYTE_CR then
if b == BYTE_SPACE or b == BYTE_TAB or b == BYTE_NEWLINE or b == BYTE_CR then
pos = pos + 1
elseif b == BYTE_SLASH then
local b2 = body:byte(pos + 1)
if b2 == BYTE_STAR then
if b2 == BYTE_STAR then
-- Block comment /* ... */
local i = pos + 2
while i < body_len do
@@ -909,7 +922,7 @@ end
parse_enum_int_literal = function(text, start)
local pos = start
local len = #text
if pos > len then return nil, start end
if pos > len then return nil, start end
local sign = 1
if text:byte(pos) == BYTE_DASH then
@@ -927,7 +940,7 @@ parse_enum_int_literal = function(text, start)
local value = 0
local has_digit = false
while pos <= len do
local d = hex_digit_value(text:byte(pos))
local d = hex_digit_value(text:byte(pos))
if not d then break end
value = value * 16 + d
has_digit = true
@@ -1024,17 +1037,17 @@ end
--- Always saves the raw RHS text into `code_macro_bodies` (for cross-source fallback during chain resolution),
--- then (if resolvable) stores the resolved integer code into `code_macros` keyed by the macro name.
--- `directive_start` points at the `#` byte. The function is silent on non-matching directives, the caller skips the line in any case.
--- @param source string
--- @param directive_start integer -- byte position of `#`
--- @param code_macros table -- out._code_macros / ctx.shared._code_macros
--- @param code_macro_bodies table -- out._code_macro_bodies / ctx.shared._code_macro_bodies
--- @param source string
--- @param directive_start integer -- byte position of `#`
--- @param code_macros table -- out._code_macros / ctx.shared._code_macros
--- @param code_macro_bodies table -- out._code_macro_bodies / ctx.shared._code_macro_bodies
local function try_extract_code_macro(source, directive_start, code_macros, code_macro_bodies)
local rest = duffle.skip_ws_and_cmt(source, directive_start + 1)
local kw, kw_end = duffle.read_ident(source, rest)
if kw ~= "define" then return end
local after_kw = duffle.skip_ws_and_cmt(source, kw_end)
local macro_name, macro_end = duffle.read_ident(source, after_kw)
local after_kw = duffle.skip_ws_and_cmt(source, kw_end)
local macro_name, macro_end = duffle.read_ident(source, after_kw)
if not macro_name then return end
if not is_r_code_macro(macro_name) then return end
@@ -1043,7 +1056,7 @@ local function try_extract_code_macro(source, directive_start, code_macros, code
local rhs_pos = duffle.skip_ws_and_cmt(source, macro_end)
local rhs_end = duffle.find_byte(source, BYTE_NEWLINE, rhs_pos) or (#source + 1)
local rhs_text = duffle.trim(source:sub(rhs_pos, rhs_end - 1))
if rhs_text ~= "" then
if rhs_text ~= "" then
code_macro_bodies[macro_name] = rhs_text
end
@@ -1081,8 +1094,7 @@ local function scan_source_pre_pass(source, code_macros, code_macro_bodies)
end
-- Check whether `atom_reg` appears as a BARE token at byte position `pos`.
-- `pos` should be at the first non-whitespace byte after the value position
-- (caller is responsible for skipping whitespace before calling).
-- `pos` should be at the first non-whitespace byte after the value position (caller is responsible for skipping whitespace before calling).
-- Returns (true, end_pos) iff the identifier at `pos` is exactly `atom_reg` AND the byte immediately before `pos` is a non-word char
-- (whitespace, `,`, `=`, `{`, `}`, `(`, `)`, `[`, `]`, etc.) AND the byte immediately after the ident is a non-word char or end-of-input.
-- Returns (false, pos) otherwise.
@@ -1090,7 +1102,7 @@ local function check_bare_atom_reg(body, pos)
if pos > #body then return false, pos end
local ident, ident_end = duffle.read_ident(body, pos)
if ident ~= "atom_reg" then return false, pos end
if ident ~= "atom_reg" then return false, pos end
-- Word-boundary check on the LEFT side: the byte at `pos - 1` must NOT be an alphanumeric/underscore byte (otherwise `atom_reg` is a suffix of `_not_atom_reg` or similar).
if pos > 1 then
@@ -1277,22 +1289,22 @@ end
local function parse_atom_info_after_decl(source, after_paren, raw_name, line_of, out, dest)
local lookahead = duffle.skip_ws_and_cmt(source, after_paren)
local look_ident, look_end = duffle.read_ident(source, lookahead)
if look_ident ~= "atom_info" then return after_paren end
if look_ident ~= "atom_info" then return after_paren end
local info_open = duffle.skip_ws_and_cmt(source, look_end)
if source:sub(info_open, info_open) ~= "(" then return after_paren end
local info_inner, info_after = duffle.read_parens(source, info_open)
local info_inner, info_after = duffle.read_parens(source, info_open)
if not info_inner then return after_paren end
local info_line = line_of(info_open)
local ai_binds, ai_reads, ai_writes, ai_view, ai_overrides, ai_ctx, ai_phase = scan_atom_info_subcalls(info_inner, info_line)
dest = dest or out.atom_infos
dest[#dest + 1] = {
atom_name = raw_name or "?", binds = ai_binds,
reads = ai_reads or {}, writes = ai_writes or {},
view = ai_view,
atom_name = raw_name or "?", binds = ai_binds,
reads = ai_reads or {}, writes = ai_writes or {},
view = ai_view,
reg_type_overrides = ai_overrides,
ctx_atom = ai_ctx,
phase = ai_phase,
info_line = line_of(lookahead),
ctx_atom = ai_ctx,
phase = ai_phase,
info_line = line_of(lookahead),
}
if ai_view and raw_name then
out.atom_views[raw_name] = {
@@ -1321,24 +1333,41 @@ end
local DECL_FORMS = {
MipsAtom_ = {
kind = "atom", name = "paren_ident", body = "braces_after",
info_dest = "atom_infos", strip = false,
kind = "atom",
name = "paren_ident",
body = "braces_after",
info_dest = "atom_infos",
strip = false,
},
MipsAtom_Proc_ = {
kind = "atom_proc", name = "backward_atom_proc", body = "last_brace_in_args",
info_dest = "atom_infos", strip = false, after = "reguse_hook",
kind = "atom_proc",
name = "backward_atom_proc",
body = "last_brace_in_args",
info_dest = "atom_infos",
strip = false,
after = "reguse_hook",
},
MipsAtomComp_ = {
kind = "comp_bare", name = "paren_ident", body = "braces_after",
info_dest = "component_atom_infos", strip = "ac_",
kind = "comp_bare",
name = "paren_ident",
body = "braces_after",
info_dest = "component_atom_infos",
strip = "ac_",
},
MipsAtomComp_Proc_ = {
kind = "comp_proc", name = "backward_fi", body = "last_brace_in_args",
info_dest = nil, strip = "ac_",
kind = "comp_proc",
name = "backward_fi",
body = "last_brace_in_args",
info_dest = nil,
strip = "ac_",
},
MipsAtomComp_ProcMap_ = {
kind = "comp_proc", name = "backward_fi", body = "comma_arg_2",
info_dest = nil, strip = "ac_", after = "map_command_hook",
kind = "comp_proc",
name = "backward_fi",
body = "comma_arg_2",
info_dest = nil,
strip = "ac_",
after = "map_command_hook",
},
}
@@ -1357,7 +1386,7 @@ local function last_brace_body(inner, open_paren)
end
local function reguse_hook(source, pos, line_of, out, extras)
local entry = out.atoms[#out.atoms]
local entry = out.atoms[#out.atoms]
if not entry then return end
local reg_use_schema_name, reg_use_param_name
if extras.args_inner then
@@ -1396,10 +1425,10 @@ end
local function parse_decl_form(source, pos, ident_end, line_of, out)
local ident = duffle.read_ident(source, pos)
local form = ident and DECL_FORMS[ident]
local form = ident and DECL_FORMS[ident]
if not form then return ident_end end
local inner, after_paren, open_paren = read_parens_after(source, ident_end)
local inner, after_paren, open_paren = read_parens_after(source, ident_end)
if not inner then return after_paren end
local extras = {}
@@ -1428,8 +1457,7 @@ local function parse_decl_form(source, pos, ident_end, line_of, out)
if form.body == "braces_after" then
local brace_search = after_paren
if info_dest then
brace_search = parse_atom_info_after_decl(
source, after_paren, name, line_of, out, info_dest)
brace_search = parse_atom_info_after_decl(source, after_paren, name, line_of, out, info_dest)
end
local after_brace
body, after_brace, body_off = find_body_braces(source, brace_search, open_paren + 1)
@@ -1440,11 +1468,8 @@ local function parse_decl_form(source, pos, ident_end, line_of, out)
if not body then return after_paren end
resume = after_paren
if form.info_dest then
if extras.after_func_paren then
parse_atom_info_after_decl(
source, extras.after_func_paren, name, line_of, out, out.atom_infos)
else
parse_atom_info_after_decl(source, pos, name, line_of, out, out.atom_infos)
if extras.after_func_paren then parse_atom_info_after_decl(source, extras.after_func_paren, name, line_of, out, out.atom_infos)
else parse_atom_info_after_decl(source, pos, name, line_of, out, out.atom_infos)
end
end
elseif form.body == "comma_arg_2" then
@@ -1489,8 +1514,8 @@ local function parse_mips_code(source, pos, ident_end, line_of, out)
return ident_end
end
local atom_name = next_ident:sub(6)
local body, after_brace, body_off = find_body_braces(source, next_after, ident_end)
local atom_name = next_ident:sub(6)
local body, after_brace, body_off = find_body_braces(source, next_after, ident_end)
if not body then return after_brace end
register_raw_atom(out, line_of(pos), atom_name, body, body_off, atom_name, pos)
@@ -1579,11 +1604,24 @@ local function register_typedef_alias(underlying, name, pos, line_of, out)
}
end
local function register_array_type(name, elem, counts, pos, line_of, out)
out.type_name_registry[name] = {
name = name,
kind = "array",
elem = elem,
counts = counts,
byte_size = nil,
source_line = line_of(pos),
source_file = out._source_file,
pointer_depth = 0,
}
end
local parse_reg_use_schema_body
local function fields_for_reg_type(type_name, type_registry)
local reg_name = "Reg_" .. type_name
local entry = type_registry and type_registry[reg_name]
local entry = type_registry and type_registry[reg_name]
if entry and entry.fields and #entry.fields > 0 then
local names = {}
for _, field in ipairs(entry.fields) do
@@ -1607,11 +1645,11 @@ end
parse_reg_use_schema_body = function(body, type_registry, opts)
opts = opts or {}
local require_types = opts.require_types == true
local pending = false
local slots = {}
local pending = false
local slots = {}
local alias_to_slot = {}
local slot_names = {}
local errors = {}
local slot_names = {}
local errors = {}
local function add_alias(path, slot)
if alias_to_slot[path] then
@@ -1637,7 +1675,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
local names = {}
while pos <= #text do
pos = duffle.skip_ws_and_cmt(text, pos)
local name, name_end = duffle.read_ident(text, pos)
local name, name_end = duffle.read_ident(text, pos)
if not name then return nil, pos end
names[#names + 1] = name
pos = duffle.skip_ws_and_cmt(text, name_end)
@@ -1674,9 +1712,9 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
end
local views = {}
local views = {}
local union_readonly = nil
local inner_pos = 1
local inner_pos = 1
local function note_readonly(flag)
if union_readonly == nil then
@@ -1761,7 +1799,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
while s_pos <= #struct_inner do
s_pos = duffle.skip_ws_and_cmt(struct_inner, s_pos)
if s_pos > #struct_inner then break end
local s_ty, s_ty_end = duffle.read_ident(struct_inner, s_pos)
local s_ty, s_ty_end = duffle.read_ident(struct_inner, s_pos)
if not s_ty then
s_pos = s_pos + 1
goto continue_struct
@@ -1772,14 +1810,14 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
end
local type_inner, after_paren = duffle.read_parens(struct_inner, after_ty)
local type_inner, after_paren = duffle.read_parens(struct_inner, after_ty)
if not type_inner then
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
end
local typed_fields = fields_for_reg_type(duffle.trim(type_inner), type_registry)
after_paren = duffle.skip_ws_and_cmt(struct_inner, after_paren)
local inst_names, new_s = parse_reg_names(struct_inner, after_paren)
local inst_names, new_s = parse_reg_names(struct_inner, after_paren)
if not inst_names or #inst_names == 0 then
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
@@ -1800,7 +1838,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
end
s_pos = new_s
elseif s_ty == "Reg" then
local s_after = duffle.skip_ws_and_cmt(struct_inner, s_ty_end)
local s_after = duffle.skip_ws_and_cmt(struct_inner, s_ty_end)
local s_readonly = false
local maybe_const, maybe_end = duffle.read_ident(struct_inner, s_after)
if maybe_const == "const" then
@@ -1808,7 +1846,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
s_after = duffle.skip_ws_and_cmt(struct_inner, maybe_end)
end
if not note_readonly(s_readonly) then return nil, errors end
local names, new_s = parse_reg_names(struct_inner, s_after)
local names, new_s = parse_reg_names(struct_inner, s_after)
if not names or #names == 0 then
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
@@ -1832,12 +1870,12 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
if inner:sub(inner_pos, inner_pos) == ";" then inner_pos = inner_pos + 1 end
elseif m_type == "Reg" then
local m_after = duffle.skip_ws_and_cmt(inner, m_type_end)
local m_after = duffle.skip_ws_and_cmt(inner, m_type_end)
local m_readonly = false
local maybe_const, maybe_end = duffle.read_ident(inner, m_after)
if maybe_const == "const" then
m_readonly = true
m_after = duffle.skip_ws_and_cmt(inner, maybe_end)
m_after = duffle.skip_ws_and_cmt(inner, maybe_end)
end
if not note_readonly(m_readonly) then return nil, errors end
local names, new_inner = parse_reg_names(inner, m_after)
@@ -1854,7 +1892,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
::continue_inner::
end
local after_close = duffle.skip_ws_and_cmt(body, after_braces)
local after_close = duffle.skip_ws_and_cmt(body, after_braces)
local inst_name, inst_end = duffle.read_ident(body, after_close)
if #views == 0 then
@@ -1932,7 +1970,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
end
local type_inner, after_paren = duffle.read_parens(body, after)
local type_inner, after_paren = duffle.read_parens(body, after)
if not type_inner then
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
@@ -1954,7 +1992,7 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
readonly = true
after = duffle.skip_ws_and_cmt(body, maybe_end)
end
local names, new_pos = parse_reg_names(body, after)
local names, new_pos = parse_reg_names(body, after)
if not names or #names == 0 then
errors[#errors + 1] = { kind = "reguse_malformed" }
return nil, errors
@@ -1992,6 +2030,86 @@ parse_reg_use_schema_body = function(body, type_registry, opts)
return { slots = slots, alias_to_slot = alias_to_slot, pending = pending }, errors
end
-- ── Shape 1: `typedef Struct_(<name>) { <body> } <alias>;` ────────────
local function parse_typedef_struct(source, pos, id2_end, line_of, out, after_typedef)
local inner, after_paren, open_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
local name = duffle.trim(inner)
local body, after_brace = find_body_braces(source, after_paren, open_paren + 1)
if not body then return after_brace end
register_struct_type(body, name, pos, line_of, out)
if name:sub(1, 7) == "RegUse_" then
local schema, schema_errors = parse_reg_use_schema_body(body, out.type_name_registry)
if schema then
schema.name = name
schema.source_file = out._source_file
schema.source_line = line_of(pos)
out.reg_use_schemas[name] = schema
end
for _, err in ipairs(schema_errors or {}) do
err.schema_name = name
err.source_file = out._source_file
err.source_line = line_of(pos)
out.reg_use_errors[#out.reg_use_errors + 1] = err
end
end
attach_debug_skip_marker(out, "unrelated")
return after_brace
end
-- ── Shape 2: `typedef Enum_(<underlying>, <name>) { <body> } <alias>;`
local function parse_typedef_enum (source, pos, id2_end, line_of, out, after_typedef)
local inner, after_paren, open_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
-- Split `inner` on the first top-level comma into (<underlying>, <name>).
local args = duffle.split_top_level_commas(inner)
if #args < 2 then return after_paren end
local underlying = duffle.trim(args[1])
local name = duffle.trim(args[2])
local body, after_brace = find_body_braces(source, after_paren, open_paren + 1)
if not body then return after_brace end
register_enum_type(underlying, name, body, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
return after_brace
end
-- Shape 4 (TSet_ at id2 position): no preceding underlying span.
local function parse_typedef_tset (source, pos, id2_end, line_of, out, after_typedef)
local inner, after_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
local tset_name = duffle.trim(inner)
-- Empty underlying span is acceptable; the TSet_ wrapper itself encodes the alias identity (per the duffle TSet_ convention).
register_typedef_alias("", tset_name, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
return after_paren
end
local function parse_typedef_array(source, pos, id2_end, line_of, out, after_typedef)
local inner, after_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
local args = duffle.split_top_level_commas(inner)
if #args < 2 then return after_paren end
local elem = duffle.trim(args[1])
local len = tonumber(duffle.trim(args[2]), 10)
if type(elem) ~= "string" or elem == "" or not len or len < 1 or len ~= math.floor(len) then
return after_paren
end
local name = "A" .. tostring(len) .. "_" .. elem
register_array_type(name, elem, { len }, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
local semi = duffle.find_byte(source, BYTE_SEMI, after_paren)
return semi and (semi + 1) or after_paren
end
local TYPE_FORMS = {
Struct_ = parse_typedef_struct,
Enum_ = parse_typedef_enum,
TSet_ = parse_typedef_tset,
Array_ = parse_typedef_array,
}
--- Parse: `typedef` declarations.
---
--- Recognizes four shapes:
@@ -2015,49 +2133,9 @@ local function parse_typedef_binds(source, pos, ident_end, line_of, out)
local after_typedef = duffle.skip_ws_and_cmt(source, ident_end)
local id2, id2_end = duffle.read_ident(source, after_typedef)
if not id2 then return ident_end end
-- ── Shape 1: `typedef Struct_(<name>) { <body> } <alias>;` ────────────
if id2 == "Struct_" then
local inner, after_paren, open_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
local name = duffle.trim(inner)
local body, after_brace = find_body_braces(source, after_paren, open_paren + 1)
if not body then return after_brace end
register_struct_type(body, name, pos, line_of, out)
if name:sub(1, 7) == "RegUse_" then
local schema, schema_errors = parse_reg_use_schema_body(body, out.type_name_registry)
if schema then
schema.name = name
schema.source_file = out._source_file
schema.source_line = line_of(pos)
out.reg_use_schemas[name] = schema
end
for _, err in ipairs(schema_errors or {}) do
err.schema_name = name
err.source_file = out._source_file
err.source_line = line_of(pos)
out.reg_use_errors[#out.reg_use_errors + 1] = err
end
end
attach_debug_skip_marker(out, "unrelated")
return after_brace
-- ── Shape 2: `typedef Enum_(<underlying>, <name>) { <body> } <alias>;`
elseif id2 == "Enum_" then
local inner, after_paren, open_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
-- Split `inner` on the first top-level comma into (<underlying>, <name>).
local args = duffle.split_top_level_commas(inner)
if #args < 2 then return after_paren end
local underlying = duffle.trim(args[1])
local name = duffle.trim(args[2])
local body, after_brace = find_body_braces(source, after_paren, open_paren + 1)
if not body then return after_brace end
register_enum_type(underlying, name, body, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
return after_brace
local form = TYPE_FORMS[id2]
if form then
return form(source, pos, id2_end, line_of, out, after_typedef)
end
-- ── Shapes 3 + 4: `typedef <span> <alias>;` or
@@ -2084,17 +2162,6 @@ local function parse_typedef_binds(source, pos, ident_end, line_of, out)
local semi_pos = duffle.find_byte(source, BYTE_SEMI, id2_end)
if not semi_pos then return id2_end end
-- Shape 4 (TSet_ at id2 position): no preceding underlying span.
if id2 == "TSet_" then
local inner, after_paren = read_parens_after(source, id2_end, id2_end)
if not inner then return id2_end end
local tset_name = duffle.trim(inner)
-- Empty underlying span is acceptable; the TSet_ wrapper itself encodes the alias identity (per the duffle TSet_ convention).
register_typedef_alias("", tset_name, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
return after_paren
end
-- Walk idents forward to find the alias ident (last ident before `;`), or the TSet_(<arg>) form (capture the arg, use it as the alias).
local last_ident = nil
local last_ident_pos = nil
@@ -2129,6 +2196,33 @@ local function parse_typedef_binds(source, pos, ident_end, line_of, out)
end
end
-- C-array suffix: `typedef S2 A3x3_S2[3][3];` → kind=array, not a typedef alias.
-- Malformed `[` / non-decimal dims fall through to the typedef-alias path.
if last_ident and not tset_arg then
local dims = {}
local dim_scan = duffle.skip_ws_and_cmt(source, last_ident_end)
while dim_scan < semi_pos and source:sub(dim_scan, dim_scan) == "[" do
local close = source:find("]", dim_scan + 1, true)
if not close or close >= semi_pos then
dims = nil
break
end
local n = tonumber(duffle.trim(source:sub(dim_scan + 1, close - 1)), 10)
if not n or n < 1 or n ~= math.floor(n) then
dims = nil
break
end
dims[#dims + 1] = n
dim_scan = duffle.skip_ws_and_cmt(source, close + 1)
end
if dims and #dims > 0 then
local elem = duffle.trim(source:sub(after_typedef, last_ident_pos - 1))
register_array_type(last_ident, elem, dims, pos, line_of, out)
attach_debug_skip_marker(out, "unrelated")
return semi_pos + 1
end
end
if tset_arg then
-- Shape 4: alias is the TSet_ argument; the underlying span is the trimmed text from the start of id2 up to (but not including) the TSet_ ident.
local underlying_span = source:sub(after_typedef, tset_pos - 1)
@@ -2361,12 +2455,12 @@ local function parse_addrs_assign(source, pos, ident_end, line_of, out)
local after = duffle.skip_ws_and_cmt(source, ident_end)
if source:sub(after, after) ~= "[" then return ident_end end
local inner, after_br = duffle.read_brackets(source, after)
local idx = inner and tonumber(duffle.trim(inner))
local idx = inner and tonumber(duffle.trim(inner))
after_br = duffle.skip_ws_and_cmt(source, after_br or after)
if not (idx and source:sub(after_br, after_br) == "=") then
return after_br or (after + 1)
end
local rhs = duffle.skip_ws_and_cmt(source, after_br + 1)
local rhs = duffle.skip_ws_and_cmt(source, after_br + 1)
local rhs_ident = duffle.read_ident(source, rhs)
if rhs_ident then out._addrs[idx] = rhs_ident end
return rhs
@@ -2376,7 +2470,7 @@ local function parse_tb_emit_(source, pos, ident_end, line_of, out)
local after = duffle.skip_ws_and_cmt(source, ident_end)
if source:sub(after, after) ~= "(" then return ident_end end
local inner, after_p = duffle.read_parens(source, after)
local name = duffle.trim(inner or ""):match("^([%w_]+)")
local name = duffle.trim(inner or ""):match("^([%w_]+)")
if name then
out._chain = out._chain or {}
out._chain[#out._chain + 1] = name
@@ -2388,14 +2482,12 @@ local function parse_tb_emit(source, pos, ident_end, line_of, out)
local after = duffle.skip_ws_and_cmt(source, ident_end)
if source:sub(after, after) ~= "(" then return ident_end end
local inner, after_p = duffle.read_parens(source, after)
local args = duffle.split_top_level_commas(inner or "")
local last = duffle.trim(args[#args] or "")
local idx = last:match("^addrs%s*%[%s*(%d+)%s*%]$")
local args = duffle.split_top_level_commas(inner or "")
local last = duffle.trim(args[#args] or "")
local idx = last:match("^addrs%s*%[%s*(%d+)%s*%]$")
local name
if idx then
name = out._addrs[tonumber(idx)]
else
name = last:match("([%w_]+)$")
if idx then name = out._addrs[tonumber(idx)]
else name = last:match("([%w_]+)$")
end
if name then
out._chain = out._chain or {}
@@ -2419,18 +2511,17 @@ local C_STMT_PARSERS = {
-- Adding a new construct = 1 row here + 1 parser function above.
local DECL_PARSERS = {
MipsAtom_ = parse_decl_form,
MipsAtom_Proc_ = parse_decl_form,
MipsAtomComp_ = parse_decl_form,
MipsAtomComp_Proc_ = parse_decl_form,
MipsAtom_ = parse_decl_form,
MipsAtom_Proc_ = parse_decl_form,
MipsAtomComp_ = parse_decl_form,
MipsAtomComp_Proc_ = parse_decl_form,
MipsAtomComp_ProcMap_ = parse_decl_form,
-- `atom_dbg_skip` is the only debug-skip parser entry. Every other
-- identifier follows the ordinary unrelated-token path; there is no alias.
-- `atom_dbg_skip` is the only debug-skip parser entry.
-- Every other identifier follows the ordinary unrelated-token path; there is no alias.
atom_dbg_skip = parse_dbg_skip_marker,
atom_dbg_reg_default = parse_atom_dbg_reg_default,
-- `atom_auto_reg(atom, R_<Sym>)` and `phase_auto_reg(phase, R_<Sym>)` populate per-source
-- `out.atom_auto_regs` / `out.phase_auto_regs`; the cross-source merge lands in
-- `corpus.atom_auto_regs` / `corpus.phase_auto_regs` (first-wins).
-- `atom_auto_reg(atom, R_<Sym>)` and `phase_auto_reg(phase, R_<Sym>)` populate per-source `out.atom_auto_regs` / `out.phase_auto_regs`;
-- The cross-source merge lands in `corpus.atom_auto_regs` / `corpus.phase_auto_regs` (first-wins).
atom_auto_reg = parse_auto_reg_marker,
phase_auto_reg = parse_auto_reg_marker,
MipsCode = parse_mips_code,
@@ -2457,50 +2548,48 @@ local DECL_PARSERS = {
local function scan_source(source, source_file, code_macros, code_macro_bodies)
local line_of = duffle.LineIndex(source)
local out = {
atoms = {},
raw_atoms = {},
binds = {},
atom_infos = {},
atoms = {},
raw_atoms = {},
binds = {},
atom_infos = {},
component_atom_infos = {},
macros = {},
-- Raw marker evidence for annotation validation. The `debug_skip` boolean
-- is stamped on the declaration record itself; the projection lives on AtomEntry.debug_skip.
debug_skip_markers = {},
types = {},
atom_views = {},
macros = {},
-- Raw marker evidence for annotation validation. The `debug_skip` boolean is stamped on the declaration record itself; the projection lives on AtomEntry.debug_skip.
debug_skip_markers = {},
types = {},
atom_views = {},
-- Per-source projection for `atom_auto_reg(<atom>, R_<Sym>)` markers.
-- Each entry is keyed by atom_name; the inner table maps `R_<Sym>` -> `R_<Sym>` (raw LHS sym).
-- Merged cross-source into `corpus.atom_auto_regs` (first-wins).
atom_auto_regs = {},
atom_auto_regs = {},
-- Per-source projection for `phase_auto_reg(<phase>, R_<Sym>)` markers.
-- Each entry is keyed by phase_label; the inner table maps `R_<Sym>` -> `R_<Sym>` (raw LHS sym).
-- Merged cross-source into `corpus.phase_auto_regs` (first-wins).
phase_auto_regs = {},
line_of = line_of,
phase_auto_regs = {},
line_of = line_of,
-- Source-derived register-alias registry (atom_reg opt-in entries).
-- Keys are full R_* idents (never stripped); see parse_enum / parse_enum_body.
register_alias_registry = {},
-- Source-derived type-name registry.
-- Populated from `typedef Struct_(...)`, `typedef Enum_(...)`, `typedef <type> <alias>`, and `typedef <type> TSet_(<name>)` declarations.
-- The propagation pass at the end of `scan_source()` resolves byte_size via the builtin map,
-- typedef chain walking (cycle-guarded, depth <= 8), and struct field sums.
-- See `propagate_type_sizes()` below.
type_name_registry = {},
reg_use_schemas = {},
tape_chains = {},
_addrs = {},
_chain = nil,
_brace_depth = 0,
reg_use_errors = {},
-- typedef chain walking (cycle-guarded, depth <= 8), and struct field sums. See `propagate_type_sizes()` below.
type_name_registry = {},
reg_use_schemas = {},
tape_chains = {},
_addrs = {},
_chain = nil,
_brace_depth = 0,
reg_use_errors = {},
-- Shared `R_*_Code -> integer code` registry
-- (passed in from M.run pass 1; same reference so preprocessor intercept writes are visible to the enum-value resolver).
-- Stripped from `src.scan` before return.
_code_macros = code_macros or {},
_code_macros = code_macros or {},
-- Shared raw RHS body table (passed in from M.run pass 1a;
-- same reference so preprocessor intercept writes are visible to the cross-source chain walker in resolve_code_macro_value).
-- Stripped from `src.scan` before return.
_code_macro_bodies = code_macro_bodies or {},
_source_file = source_file,
_code_macro_bodies = code_macro_bodies or {},
_source_file = source_file,
}
local pos = 1
local src_len = #source
@@ -2526,9 +2615,9 @@ local function scan_source(source, source_file, code_macros, code_macro_bodies)
if parser then
pos = parser(source, pos, ident_end, line_of, out)
else
-- Unsupported identifiers follow the unrelated-token path. If a
-- pending marker is still open, consume it so it cannot drift to a
-- later declaration. Unsupported identifiers never create marker records.
-- Unsupported identifiers follow the unrelated-token path.
-- If a pending marker is still open, consume it so it cannot drift to a later declaration.
-- Unsupported identifiers never create marker records.
local markers = out.debug_skip_markers
local marker = markers[#markers]
if marker and marker.pending then
@@ -2543,7 +2632,7 @@ local function scan_source(source, source_file, code_macros, code_macro_bodies)
else
local markers = out.debug_skip_markers
local marker = markers[#markers]
local c = source:sub(pos, pos)
local c = source:sub(pos, pos)
if marker and marker.pending and marker.proc_prelude then
if c == "{" or c == ";" then
attach_debug_skip_marker(out, "unrelated")
@@ -2572,8 +2661,8 @@ local function scan_source(source, source_file, code_macros, code_macro_bodies)
if out._chain and #out._chain > 0 then
out.tape_chains[#out.tape_chains + 1] = out._chain
end
out._addrs = nil
out._chain = nil
out._addrs = nil
out._chain = nil
out._brace_depth = nil
return out
@@ -2698,7 +2787,7 @@ end
-- * conflicting shape: Keep first entry, append ONE typed collision record with shape diff.
local function merge_named_with_sites(registry, name, new_entry, site, collisions, kind, shape_fn)
if registry[name] == nil then
registry[name] = new_entry
registry[name] = new_entry
registry[name].sites = { site }
return
end
@@ -2715,8 +2804,7 @@ local function merge_named_with_sites(registry, name, new_entry, site, collision
collisions[#collisions + 1] = {
kind = kind,
name = name,
first_site = existing.sites and existing.sites[1]
or build_site(existing.source_file, existing.source_line),
first_site = existing.sites and existing.sites[1] or build_site(existing.source_file, existing.source_line),
conflicting_site = site,
first_shape = old_shape,
conflicting_shape = new_shape,
@@ -2748,9 +2836,18 @@ local function merge_corpus_registries(corpus)
-- Replace the existing corpus collections with empty tables so a re-run on the same corpus produces identical state (deterministic merge).
-- This is safe because M.run is the only writer to these tables within a single orchestrator invocation.
for _, key in ipairs({
"register_alias_registry", "type_name_registry", "binds_by_name",
"atoms_by_name", "atom_views", "atom_ctxs", "atom_phases",
"atom_infos", "component_atom_infos", "collisions", "reg_use_schemas", "reg_use_errors",
"register_alias_registry",
"type_name_registry",
"binds_by_name",
"atoms_by_name",
"atom_views",
"atom_ctxs",
"atom_phases",
"atom_infos",
"component_atom_infos",
"collisions",
"reg_use_schemas",
"reg_use_errors",
"tape_chains",
}) do
corpus[key] = {}
scripts/passes/static_analysis.lua
+834 -527
View File
File diff suppressed because it is too large Load Diff
scripts/ps1_meta.lua
+2 -2
View File
@@ -531,7 +531,7 @@ local function build_ctx(args)
end
local text = file:read("*a")
file:close()
local source = {
path = path,
text = text,
@@ -727,7 +727,7 @@ local function main(argv)
local ok, err = pcall(function()
local args = parse_args(argv)
local ctx = build_ctx(args)
local requested = args.requested_set
local closed = topo_sort(PASSES, requested)