@dforge-core/dforge-mcp 0.2.44 → 0.2.46-next

package/skills/dforge-1c-import/scripts/lib/inventory.py

@@ -0,0 +1,535 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+"""
+Every register in the infobase, with the facts a declaration needs.
+
+    python tools/1c.py inventory                 # the client's registers, as a table
+    python tools/1c.py inventory --all           # the vendor's too
+    python tools/1c.py inventory --json          # out/registers.json, for other tools
+    python tools/1c.py inventory --name _AccumRg17011     # one register, in full
+    python tools/1c.py inventory --name мсбТараВЦехах     # the same, by 1C name
+
+Written after the same survey was done three times in throwaway scripts, each with the
+register names typed in by hand. That is the wrong shape twice over: the names belong to
+the configuration rather than to a script, and a script that names them cannot notice a
+register nobody has thought of yet.
+
+**What is derivable and what is not.** Everything here is a FACT read from the extracted
+metadata: which registers exist, their tables and fields, how many rows, whether 1C keeps
+a record kind or a recorder, which ported entity holds them and under what column names.
+None of it is a DECISION - that a register is in scope, what KIND it becomes in dForge,
+which document binds to it with which direction. Those live in `tools/project/registers.py`
+and should key off the names printed here rather than restate them.
+
+The two are easy to confuse because 1C's own family looks like a decision and is not:
+`мсбФактическаяРаботаЛиний` is an information register that the client writes signed values
+into and sums, so the faithful port is `turnover`. The family is evidence, not an answer,
+which is exactly why this tool reports evidence and stops.
+
+**Writers are reported from the handler text, and that source is weak in two ways.** 1C
+names a movement set two ways - `Движения.<имя>` and `Движения["<имя>"]` - and a scan that
+knows one of them reads as a clean result while missing whole registers; eight of these are
+written through the bracket form alone, so both are counted here. Worse, a register is also
+written from COMMON modules, which no scan of document handlers can reach at all:
+`мсбПродажиНалоговыйУчет` holds 478,260 rows from eight document types and its own handlers
+name it nowhere.
+
+So the writer column is evidence, never a census, and an empty one reads `ask data` rather
+than `none`. The authoritative answer is in the rows: `python tools/1c.py recorders <table>`
+reads the recorder type ids and cannot miss a syntax or a common module.
+"""
+from __future__ import print_function
+
+import codecs
+import collections
+import io
+import json
+import os
+import re
+import sys
+
+import binding
+import paths
+
+OUT = paths.OUT
+
+# The two ways a handler names a movement set, and the marker of a handler that consumes
+# from the register's own balance rather than writing what its lines say.
+DOT = u"Движения."
+BRACKET = u"Движения["
+CONSUME = u"ВыполнитьРасход"
+CLIENT = binding.client_prefixes()
+
+# THE SECOND WAY A HANDLER CONSUMES, and the one that hid a 31,166-row writer.
+# `ВыполнитьРасход` is one spelling: ask the register to take from its balance. The other is
+# to JOIN the register's own Остатки inside the posting handler and write movements built
+# from what came back - `мсбВыпускБлюд` takes Рецептура and Цена off the balance that way,
+# then writes an expense with `ВыполнитьДвижения`. Same algorithm, no marker, and the sweep
+# called that register bindable.
+#
+# WHERE the read sits decides, which is why this is not a plain substring test: a fill
+# procedure reading the balance is a user pressing "Заполнить" and says nothing about
+# posting. `мсбПланИзготовленияБлюд` reads the same balance in `ЗаполнитьОстатки()` and
+# posts only what its lines carry, so it is NOT algorithmic - and a test that ignored the
+# procedure would have said it was.
+POSTING = u"ОбработкаПроведения"
+OWN_READ = (u"РегистрНакопления\\.%s\\.(?:ОстаткиИОбороты|Остатки|Обороты)")
+PROCEDURE = re.compile(u"(?m)^[ 	]*(?:Процедура|Функция)[ 	]+([А-Яа-яA-Za-z0-9_]+)")
+
+FAMILIES = ("AccumRg", "InfoRg")
+
+# How a query asks a register for an answer. 1C spells the virtual tables by suffixing the
+# register's name, so counting them says what the configuration actually WANTS from it -
+# which is the evidence that decides whether a port models it as a balance or a turnover.
+# Longest first: ОстаткиИОбороты contains both of the others.
+READS = [
+	(u"ОстаткиИОбороты", "both"),
+	(u"Остатки", "remains"),
+	(u"Обороты", "turnover"),
+	(u"СрезПоследних", "slice"),
+]
+
+# A read that names no virtual table at all. "РегистрНакопления.<name>" with nothing appended is
+# the register's own table, which is how a turnover register kept as a fact table gets read -
+# and the virtual-table count says nothing at all about that. "Движения.<name>" is a WRITE,
+# which is why the prefix is part of the needle rather than the bare name.
+DIRECT = [u"РегистрНакопления.", u"РегистрСведений."]
+
+
+def _mentions(text, name):
+	"""Does the text name this register, as opposed to one whose name starts the same?
+
+	`мсбПродажиНалоговыйУчет` is a prefix of `мсбПродажиНалоговыйУчетВременный`, so a plain
+	substring search hands the shorter register the longer one's writers - which it did:
+	`КорректировкаДолга` was reported as writing the first when it writes the second. 1C
+	identifiers run on letters, digits and underscore, so the character after the name has
+	to be none of those. Cyrillic counts as alphanumeric here, which is exactly what makes
+	the check work.
+	"""
+	at = text.find(name)
+	while at >= 0:
+		after = text[at + len(name):at + len(name) + 1]
+		if not (after.isalnum() or after == "_"):
+			return True
+		at = text.find(name, at + 1)
+	return False
+
+
+def _kind(family, record_kind):
+	"""What the infobase's own shape says this register is.
+
+	`_RecordKind` tells a balance register from a turnover one **among accumulation
+	registers**. An information register never has one, so reading its absence the same way
+	labelled all 56 of the client's information registers "turnover" and offered that as
+	modelling guidance. The family comes first.
+	"""
+	if family != "AccumRg":
+		return "info"
+	return "bal" if record_kind else "turn"
+
+
+def _every_module():
+	"""Object -> the text of every module in the configuration, not only documents.
+
+	A register is READ from anywhere - reports, common modules, other documents - so the
+	question "what does the configuration want from this register" cannot be asked of
+	document handlers alone.
+	"""
+	index_path = os.path.join(OUT, "code", "index.json")
+	if not os.path.isfile(index_path):
+		return {}
+	texts = collections.OrderedDict()
+	with io.open(index_path, encoding="utf-8") as f:
+		index = json.load(f)
+	for entry in index:
+		path = os.path.join(OUT, entry["path"])
+		if not os.path.isfile(path):
+			continue
+		with io.open(path, encoding="utf-8", errors="ignore") as f:
+			texts.setdefault(entry["object"], []).append(f.read())
+	return dict((name, u"".join(parts)) for name, parts in texts.items())
+
+
+def _reads(modules, name):
+	"""How the configuration asks this register for an answer, and who asks.
+
+	"ОстаткиИОбороты" CONTAINS both "Остатки" and "Обороты", so its occurrences are
+	subtracted from theirs rather than stopping the scan. Stopping at the first suffix that
+	matched threw away every other suffix the same module used, and a module that asks for
+	the balance in one query and the turnover in another is the ordinary case: it is the
+	difference between "this register is read as a balance" and "read as both".
+	"""
+	counts = collections.OrderedDict((label, 0) for _, label in READS)
+	counts["direct"] = 0
+	askers = {}
+	for obj, text in modules.items():
+		big = READS[0][0]
+		both = text.count(name + big)
+		for suffix, label in READS:
+			found = text.count(name + suffix)
+			if label in ("remains", "turnover"):
+				found -= both
+			if found <= 0:
+				continue
+			counts[label] += found
+			askers.setdefault(label, set()).add(obj)
+		bare = sum(text.count(prefix + name) for prefix in DIRECT)
+		qualified = sum(text.count(prefix + name + suffix)
+		                for prefix in DIRECT for suffix, _ in READS)
+		# The long suffix was counted three times in `qualified`: once as itself and once
+		# inside each of the two shorter suffixes it contains.
+		qualified -= 2 * sum(text.count(prefix + name + big) for prefix in DIRECT)
+		if bare - qualified > 0:
+			counts["direct"] += bare - qualified
+			askers.setdefault("direct", set()).add(obj)
+	return counts, dict((k, sorted(v)) for k, v in askers.items())
+
+
+def _handlers():
+	"""Document -> the text of every module it has, from the extracted code."""
+	index_path = os.path.join(OUT, "code", "index.json")
+	if not os.path.isfile(index_path):
+		return {}
+	with io.open(index_path, encoding="utf-8") as f:
+		index = json.load(f)
+	texts = collections.OrderedDict()
+	for entry in index:
+		if entry.get("family") != "Document":
+			continue
+		path = os.path.join(OUT, entry["path"])
+		if not os.path.isfile(path):
+			continue
+		with io.open(path, encoding="utf-8", errors="ignore") as f:
+			texts.setdefault(entry["object"], []).append(f.read())
+	return dict((name, u"".join(parts)) for name, parts in texts.items())
+
+
+def _ported():
+	"""Physical table -> (module.entity, {column: source}), from the column map."""
+	import columnmap
+
+	generated, mapped = columnmap.build()
+	out = {}
+	for module in mapped:
+		for entity in mapped[module]:
+			found = mapped[module][entity]
+			if not isinstance(found, dict):
+				continue
+			table = found.get("table")
+			if table:
+				out[table] = ("%s.%s" % (module, entity), found.get("columns") or {})
+	return out
+
+
+def _consumes_in_posting(text, name):
+	"""The procedure where this document's POSTING path reads the register's own balance.
+
+	The procedures the posting handler can reach are itself plus everything it names - one
+	hop, which is as far as this needs to go: a handler that builds its movements out of the
+	balance does the reading where it writes, or in a procedure it calls with the movement
+	table in hand.
+
+	Returns that procedure's name, or None. Naming it matters: "this is an algorithm" is
+	worth no more than the line someone can go and look at.
+	"""
+	bounds = [(m.group(1), m.start()) for m in PROCEDURE.finditer(text)]
+	if not bounds:
+		return None
+	ends = [start for _, start in bounds[1:]] + [len(text)]
+	bodies = [(proc, text[start:end]) for (proc, start), end in zip(bounds, ends)]
+
+	reach = set()
+	for proc, body in bodies:
+		if proc == POSTING:
+			reach.add(proc)
+			reach.update(other for other, _ in bodies if other != proc and other in body)
+
+	asked = re.compile(OWN_READ % re.escape(name))
+	for proc, body in bodies:
+		if proc in reach and asked.search(body):
+			return proc
+	return None
+
+
+def collect(everything=False):
+	"""One record per register: what it is, where it landed, and who writes it."""
+	import attrs
+	import autospec
+
+	generated, model = autospec.build()
+	ported = _ported()
+	handlers = _handlers()
+	modules = _every_module()
+
+	rows_by_table = {}
+	report_path = os.path.join(OUT, "report.json")
+	if os.path.isfile(report_path):
+		with codecs.open(report_path, "r", "utf-8") as f:
+			for row in json.load(f):
+				if row.get("physical"):
+					rows_by_table[row["physical"]] = row.get("rows")
+
+	found = []
+	for obj in model.objects:
+		family = obj.get("family")
+		if family not in FAMILIES:
+			continue
+		name = obj.get("name") or ""
+		mine = name.startswith(CLIENT)
+		if not (everything or mine):
+			continue
+		table, _ = model.physical(name, family)
+		columns = model.table_columns(table) if table else set()
+
+		fields = []
+		for field in model.fields(name, family):
+			physical = model.fld.get(field.get("guid"))
+			fields.append({
+				"name": field.get("name"),
+				"type": field.get("type"),
+				"column": physical,
+				# Which of the three a field is, as 1C declared it. This is the one thing
+				# the declaration cannot be guessed from: a register's KEY is its
+				# dimensions and what it SUMS is its resources, and reading the two off
+				# the column types gets it wrong as soon as a resource is a quantity and
+				# a dimension is a number. 1C says so outright, in the collection the
+				# field sits in, and the inventory was throwing that away.
+				"role": attrs.ROLE.get(field.get("collection"), "?"),
+				# 1C splits a reference that may point at several catalogues into
+				# _RRRef/_RTRef/_TYPE. dForge references are single-target, so this is
+				# a modelling problem and worth carrying as a fact.
+				"composite": bool(physical and (physical + "_TYPE") in columns),
+			})
+
+		writers = []
+		for document, text in handlers.items():
+			# The bracket form carries its own boundary - the closing quote - so only the
+			# dotted one needs the identifier check.
+			dot = _mentions(text, DOT + name)
+			bracket = (BRACKET + '"' + name + '"') in text
+			if not (dot or bracket):
+				continue
+			near = []
+			lines = text.splitlines()
+			for i, line in enumerate(lines):
+				if _mentions(line, name):
+					near.extend(lines[max(0, i - 25):i + 25])
+			writers.append({
+				"document": document,
+				"client": document.startswith(CLIENT),
+				"how": ("dot" if dot else "") + ("+bracket" if dot and bracket
+				                                 else "bracket" if bracket else ""),
+				# A handler that reads the balance and takes from it is an algorithm, and
+				# no declarative binding expresses one. Two spellings, both counted -
+				# see the note on POSTING for why the second one needs the procedure.
+				"consumes": (CONSUME in u"\n".join(near)
+				             or bool(_consumes_in_posting(text, name))),
+				"consumesIn": _consumes_in_posting(text, name),
+			})
+
+		read_counts, read_by = _reads(modules, name)
+		entity, mapped_columns = ported.get(table, (None, {}))
+		found.append({
+			"reads": read_counts,
+			"readBy": read_by,
+			"name": name,
+			"family": family,
+			"table": table,
+			"rows": rows_by_table.get(table),
+			"client": mine,
+			"recordKind": "_RecordKind" in columns,
+			"recorder": "_RecorderRRef" in columns,
+			"period": "_Period" in columns,
+			"entity": entity,
+			"columns": mapped_columns,
+			"fields": fields,
+			"writers": sorted(writers, key=lambda w: w["document"]),
+		})
+	return sorted(found, key=lambda r: -(r["rows"] or 0))
+
+
+def declared():
+	"""Every register the port has declared: the entity it holds -> its register code.
+
+	Read from the GENERATED packages (`modules/<code>/registers.json`) rather than from a
+	project's overlay. A first attempt went through `binding.registers()` and reported two
+	declared registers as missing, because that overlay only carries the registers of
+	GENERATED modules - an authored module declares its own inline, and `shipping` has two.
+	The package is where both end up, and it is also what actually gets installed, so it is
+	the one place that cannot disagree with the tenant.
+	"""
+	out = {}
+	if not os.path.isdir(paths.MODULES):
+		return out
+	for module in sorted(os.listdir(paths.MODULES)):
+		path = os.path.join(paths.MODULES, module, "registers.json")
+		if not os.path.isfile(path):
+			continue
+		with io.open(path, encoding="utf-8") as f:
+			try:
+				declarations = json.load(f)
+			except ValueError:
+				continue
+		for code, definition in (declarations or {}).items():
+			if not isinstance(definition, dict):
+				continue
+			held = definition.get("journal") or definition.get("entity")
+			if held:
+				out["%s.%s" % (module, held)] = "%s.%s" % (module, code)
+	return out
+
+
+def undeclared(records):
+	"""Registers a document writes that the port has not declared, worst first.
+
+	This is the reason the inventory exists. A register nobody has thought of cannot be
+	remembered, and a list typed into a script cannot grow. The configuration knows; ask it.
+
+	**A register with no ported entity belongs here too.** The first version skipped them,
+	on the reasoning that a declaration needs an entity to be declared over - which is true
+	and is not a reason to hide them: not being ported at all is a bigger gap than not being
+	declared, not a smaller one. Four registers and 460,813 rows were missing from the
+	report because of it. They are listed with the entity shown as `unmapped`.
+
+	**Nor does a register qualify by having a writer we can SEE.** The writer list comes
+	from document handler text, and 1C also writes registers from common modules, which no
+	scan of documents can reach: `мсбПродажиНалоговыйУчет` holds 478,260 rows written by
+	eight document types - the data says so through their recorder ids - and its own
+	handlers name it nowhere. Gating on the text would have dropped it from the report
+	entirely. So what qualifies is holding ROWS: something wrote them, whether or not we
+	can point at it.
+	"""
+	held = declared()
+	out = []
+	for record in records:
+		if not (record["writers"] or record["rows"]):
+			continue
+		if record["entity"] and record["entity"] in held:
+			continue
+		out.append(record)
+	return out
+
+
+def _table(records):
+	print("%-46s %-7s %9s %6s %7s %7s  %s"
+	      % ("register", "family", "rows", "kind", "writers", "consume", "ported entity"))
+	print("-" * 118)
+	for record in records:
+		writers = record["writers"]
+		print("%-46s %-7s %9s %6s %7s %7s  %s"
+		      % (record["name"][:46], record["family"],
+		         record["rows"] if record["rows"] is not None else "?",
+		         _kind(record["family"], record["recordKind"]),
+		         "%d/%d" % (sum(1 for w in writers if w["client"]), len(writers)),
+		         sum(1 for w in writers if w["consumes"]) or "-",
+		         record["entity"] or "-"))
+	written = [r for r in records if r["writers"]]
+	algorithmic = [r for r in written if any(w["consumes"] for w in r["writers"])]
+	print("-" * 118)
+	print("%d registers, %d written by a document, %d of those algorithmic; %s rows"
+	      % (len(records), len(written), len(algorithmic),
+	         sum(r["rows"] or 0 for r in records)))
+
+
+def _detail(record):
+	print("%s   %s   %s" % (record["name"], record["family"], record["table"]))
+	shape = {
+		"bal": "a record kind, so a balance register",
+		"turn": "no record kind, so a turnover register",
+		"info": "an information register - a record kind would not apply",
+	}[_kind(record["family"], record["recordKind"])]
+	print("   rows %s, %s, %s, %s"
+	      % (record["rows"], shape,
+	         "written by a recorder" if record["recorder"] else "maintained directly",
+	         "periodic" if record["period"] else "not periodic"))
+	print("   ported to %s" % (record["entity"] or "nothing - not ported at all"))
+	print("\n   fields (dimensions key the register, resources are what it sums):")
+	order = {"dimension": 0, "resource": 1, "attr": 2}
+	for field in sorted(record["fields"], key=lambda f: order.get(f.get("role"), 3)):
+		ported = [c for c, source in (record["columns"] or {}).items()
+		          if field["column"] and source and field["column"] in source]
+		print("      %-10s %-26s %-34s %-12s %s%s"
+		      % (field.get("role") or "?", field["name"], str(field["type"])[:34],
+		         field["column"] or "", ported[0] if ported else "",
+		         "   COMPOSITE" if field["composite"] else ""))
+	asked = [(label, n) for label, n in record["reads"].items() if n]
+	print("\n   what the configuration asks it for:")
+	if asked:
+		for label, n in asked:
+			who = record["readBy"].get(label) or []
+			print("      %-10s %4d query(ies)   %s%s"
+			      % (label, n, ", ".join(who[:4]), " ..." if len(who) > 4 else ""))
+	else:
+		print("      nothing - no query names its virtual tables")
+	print("\n   writers:")
+	for writer in record["writers"]:
+		print("      %-46s %-8s %-8s %s"
+		      % (writer["document"][:46], "client" if writer["client"] else "vendor",
+		         writer["how"], "CONSUMES from the balance" if writer["consumes"] else ""))
+	if not record["writers"]:
+		print("      (no document writes it)")
+	print("\n   the data has the last word on writers:")
+	print("      python tools/1c.py recorders %s" % record["table"])
+
+
+def main():
+	args = sys.argv[1:]
+	everything = "--all" in args
+	as_json = "--json" in args
+	name = None
+	if "--name" in args:
+		name = args[args.index("--name") + 1].decode("utf-8") \
+			if sys.version_info[0] == 2 else args[args.index("--name") + 1]
+
+	records = collect(everything=everything or bool(name))
+	if name:
+		# By 1C name or by physical table. The table is the ASCII handle: it is what an
+		# error, a log line or `recorders` gives you, and it survives a shell that mangles
+		# Cyrillic on the command line.
+		for record in records:
+			if name in (record["name"], record["table"]):
+				_detail(record)
+				return 0
+		print("no register called %s - try --all, or give its _AccumRg/_InfoRg table" % name)
+		return 1
+
+	if "--undeclared" in args:
+		left = undeclared(records)
+		visible = [r for r in left if r["writers"]]
+		print("%d register(s) the port has not declared: they hold rows, or a handler names "
+		      "them, or both.\n%d have a writer we can see; the rest say `ask data`, which "
+		      "means\n`recorders <table>` is the one that knows.\n"
+		      % (len(left), len(visible)))
+		print("%-46s %9s %5s %8s %10s  %s"
+		      % ("register", "rows", "kind", "writers", "consuming", "ported entity"))
+		print("-" * 106)
+		for record in left:
+			writers = record["writers"]
+			consuming = sum(1 for w in writers if w["consumes"])
+			print("%-46s %9s %5s %8s %10s  %s"
+			      % (record["name"][:46], record["rows"],
+			         _kind(record["family"], record["recordKind"]),
+			         "%d/%d" % (sum(1 for w in writers if w["client"]), len(writers))
+			         if writers else "ask data",
+			         "%d ALGO" % consuming if consuming else "-",
+			         record["entity"] or "unmapped - not ported at all"))
+		print("-" * 106)
+		print("%s rows, of which %s behind a handler that consumes from the balance"
+		      % (sum(r["rows"] or 0 for r in left),
+		         sum(r["rows"] or 0 for r in left
+		             if any(w["consumes"] for w in r["writers"]))))
+		return 0
+
+	if as_json:
+		path = paths.derived("registers.json")
+		with io.open(path, "w", encoding="utf-8") as f:
+			f.write(json.dumps(records, ensure_ascii=False, indent="\t"))
+		print("%s: %d register(s)" % (path, len(records)))
+		return 0
+
+	_table(records)
+	return 0
+
+
+if __name__ == "__main__":
+	sys.exit(main())

package/skills/dforge-1c-import/scripts/lib/layers.py

@@ -0,0 +1,106 @@
+#!/usr/bin/env python
+# -*- coding: utf-8 -*-
+u"""Does a lower layer of modules name anything in the layer installed on top of it?
+
+A port of a configuration built on a vendor's comes out as two layers: the vendor's,
+generated from the client's own configuration, and the client's, installed on top and
+depending on it - never the other way round. That is what lets the vendor layer install
+for the next client without this one's, and what lets this client's changes be seen and
+priced on their own (docs/task-vendor-logic.md section 2).
+
+A convention does not keep that true; this does. It reads every module of the lower layer
+and reports each place it names a module of the upper one: an entity link, a foreign key,
+a register binding, a declared dependency. Exit status 1 when there is any.
+
+    python tools/1c.py layers <lower dir> <upper dir>     # e.g. modules-utp modules
+
+The directories are arguments rather than names in here, so the check carries to the next
+project unchanged.
+"""
+from __future__ import print_function
+
+import codecs
+import collections
+import io
+import json
+import os
+import re
+import sys
+
+import paths
+
+QUALIFIED = re.compile(r"^([a-z][a-z0-9_]*)\.([a-z0-9_]+)$")
+
+
+def modules_in(directory):
+	"""Module code -> its directory, for every module (a folder holding a manifest)."""
+	found = {}
+	for name in sorted(os.listdir(directory)):
+		manifest = os.path.join(directory, name, "manifest.json")
+		if os.path.isfile(manifest):
+			with io.open(manifest, encoding="utf-8-sig") as f:
+				found[json.load(f).get("code", name)] = os.path.join(directory, name)
+	return found
+
+
+def crossings(lower, upper):
+	"""(lower module, file, json path, named upper module.object) for every crossing."""
+	upper_codes = set(modules_in(upper))
+	found = []
+	for code, folder in sorted(modules_in(lower).items()):
+		with io.open(os.path.join(folder, "manifest.json"), encoding="utf-8-sig") as f:
+			for dependency in json.load(f).get("dependencies", {}):
+				if dependency in upper_codes:
+					found.append((code, "manifest.json", "dependencies", dependency))
+		for root, _, files in os.walk(folder):
+			for name in sorted(files):
+				if not name.endswith(".json") or name == "manifest.json":
+					continue
+				path = os.path.join(root, name)
+				with io.open(path, encoding="utf-8-sig") as f:
+					try:
+						tree = json.load(f)
+					except ValueError:
+						continue
+				relative = os.path.relpath(path, folder).replace(os.sep, "/")
+				for where, value in _strings(tree, ()):
+					match = QUALIFIED.match(value)
+					if match and match.group(1) in upper_codes:
+						found.append((code, relative, "/".join(where), value))
+	return found
+
+
+def _strings(node, where):
+	if isinstance(node, dict):
+		for key, value in node.items():
+			for item in _strings(value, where + (key,)):
+				yield item
+	elif isinstance(node, list):
+		for index, value in enumerate(node):
+			for item in _strings(value, where + (str(index),)):
+				yield item
+	elif isinstance(node, type(u"")):
+		yield where, node
+
+
+def main():
+	args = [a for a in sys.argv[1:] if not a.startswith("-")]
+	if len(args) != 2:
+		raise SystemExit("usage: layers <lower dir> <upper dir>")
+	lower, upper = (os.path.join(paths.ROOT, a) if not os.path.isabs(a) else a for a in args)
+	found = crossings(lower, upper)
+	out = codecs.getwriter("utf-8")(sys.stdout.buffer)
+	by_target = collections.Counter(c[3] for c in found)
+	places = collections.defaultdict(set)
+	for module, path, where, target in found:
+		places[target].add(u"%s/%s" % (module, path))
+	for target, count in by_target.most_common():
+		out.write(u"%6d  %-34s from %d file(s), e.g. %s\n"
+		          % (count, target, len(places[target]), sorted(places[target])[0]))
+	out.write(u"%d reference(s) from %s into %s, %d distinct object(s)\n"
+	          % (len(found), args[0], args[1], len(by_target)))
+	return 1 if found else 0
+
+
+if __name__ == "__main__":
+	sys.exit(main() or 0)