[no squash] Revamp dump

builtin/common/misc_helpers.lua

@@ -7,7 +7,15 @@ local math = math
 local function basic_dump(o)
 	local tp = type(o)
 	if tp == "number" then
-		return tostring(o)
+		local s = tostring(o)
+		if tonumber(s) == o then
+			return s
+		end
+		-- Prefer an exact representation over a compact representation.
+		-- e.g. basic_dump(0.3) == "0.3",
+		-- but basic_dump(0.1 + 0.2) == "0.30000000000000004"
+		-- so the user can see that 0.1 + 0.2 ~= 0.3
+		return string.format("%.17g", o)
 	elseif tp == "string" then
 		return string.format("%q", o)
 	elseif tp == "boolean" then
@@ -100,65 +108,141 @@ function dump2(o, name, dumped)
 	return string.format("%s = {}\n%s", name, table.concat(t))
 end
 
---------------------------------------------------------------------------------
--- This dumps values in a one-statement format.
+
+-- This dumps values in a human-readable expression format.
+-- If possible, the resulting string should evaluate to an equivalent value if loaded and executed.
 -- For example, {test = {"Testing..."}} becomes:
 -- [[{
 -- 	test = {
 -- 		"Testing..."
 -- 	}
 -- }]]
--- This supports tables as keys, but not circular references.
--- It performs poorly with multiple references as it writes out the full
--- table each time.
--- The indent field specifies a indentation string, it defaults to a tab.
--- Use the empty string to disable indentation.
--- The dumped and level arguments are internal-only.
-
-function dump(o, indent, nested, level)
-	local t = type(o)
-	if not level and t == "userdata" then
-		-- when userdata (e.g. player) is passed directly, print its metatable:
-		return "userdata metatable: " .. dump(getmetatable(o))
-	end
-	if t ~= "table" then
-		return basic_dump(o)
+function dump(value, indent)
+	indent = indent or "\t"
+	local newline = indent == "" and "" or "\n"
+
+	local rope = {}
+	local write
+	do
+		-- Keeping the length of the table as a local variable is *much*
+		-- faster than invoking the length operator.
+		-- See https://gitspartv.github.io/LuaJIT-Benchmarks/#test12.
+		local i = 0
+		function write(str)
+			i = i + 1
+			rope[i] = str
+		end
 	end
 
-	-- Contains table -> true/nil of currently nested tables
-	nested = nested or {}
-	if nested[o] then
-		return "<circular reference>"
+	local n_refs = {}
+	local function count_refs(val)
+		if type(val) ~= "table" then
+			return
+		end
+		local tbl = val
+		if n_refs[tbl] then
+			n_refs[tbl] = n_refs[tbl] + 1
+			return
+		end
+		n_refs[tbl] = 1
+		for k, v in pairs(tbl) do
+			count_refs(k)
+			count_refs(v)
+		end
 	end
-	nested[o] = true
-	indent = indent or "\t"
-	level = level or 1
+	count_refs(value)
 
-	local ret = {}
-	local dumped_indexes = {}
-	for i, v in ipairs(o) do
-		ret[#ret + 1] = dump(v, indent, nested, level + 1)
-		dumped_indexes[i] = true
-	end
-	for k, v in pairs(o) do
-		if not dumped_indexes[k] then
-			if type(k) ~= "string" or not is_valid_identifier(k) then
-				k = "["..dump(k, indent, nested, level + 1).."]"
+	local refs = {}
+	local cur_ref = 1
+	local function write_value(val, level)
+		if type(val) ~= "table" then
+			write(basic_dump(val))
+			return
+		end
+
+		local tbl = val
+		if refs[tbl] then
+			write(refs[tbl])
+			return
+		end
+
+		if n_refs[val] > 1 then
+			refs[val] = ("getref(%d)"):format(cur_ref)
+			write(("setref(%d)"):format(cur_ref))
+			cur_ref = cur_ref + 1
+		end
+		write("{")
+		if next(tbl) == nil then
+			write("}")
+			return
+		end
+		write(newline)
+
+		local function write_entry(k, v)
+			write(indent:rep(level))
+			write("[")
+			write_value(k, level + 1)
+			write("] = ")
+			write_value(v, level + 1)
+			write(",")
+			write(newline)
+		end
+
+		local keys = {string = {}, number = {}}
+		for k in pairs(tbl) do
+			local t = type(k)
+			if keys[t] then
+				table.insert(keys[t], k)
 			end
-			v = dump(v, indent, nested, level + 1)
-			ret[#ret + 1] = k.." = "..v
 		end
+
+		-- Write string-keyed entries
+		table.sort(keys.string)
+		for _, k in ipairs(keys.string) do
+			local v = val[k]
+			if is_valid_identifier(k) then
+				write(indent:rep(level))
+				write(k)
+				write(" = ")
+				write_value(v, level + 1)
+				write(",")
+				write(newline)
+			else
+				write_entry(k, v)
+			end
+		end
+
+		-- Write number-keyed entries
+		local len = 0
+		for i in ipairs(tbl) do
+			len = i
+		end
+		if #keys.number == len then -- table is a list
+			for _, v in ipairs(tbl) do
+				write(indent:rep(level))
+				write_value(v, level + 1)
+				write(",")
+				write(newline)
+			end
+		else -- table harbors arbitrary number keys
+			table.sort(keys.number)
+			for _, k in ipairs(keys.number) do
+				write_entry(k, tbl[k])
+			end
+		end
+
+		-- Write all remaining entries
+		for k, v in pairs(val) do
+			if not keys[type(k)] then
+				write_entry(k, v)
+			end
+		end
+
+		write(indent:rep(level - 1))
+		write("}")
 	end
-	nested[o] = nil
-	if indent ~= "" then
-		local indent_str = "\n"..string.rep(indent, level)
-		local end_indent_str = "\n"..string.rep(indent, level - 1)
-		return string.format("{%s%s%s}",
-				indent_str,
-				table.concat(ret, ","..indent_str),
-				end_indent_str)
-	end
-	return "{"..table.concat(ret, ", ").."}"
+	write_value(value, 1)
+	return table.concat(rope)
 end
 
 --------------------------------------------------------------------------------

builtin/common/serialize.lua

@@ -218,9 +218,13 @@ function core.serialize(value)
 		core.log("deprecated", "Support for dumping functions in `core.serialize` is deprecated.")
 	end
 	local rope = {}
+	-- Keeping the length of the table as a local variable is *much*
+	-- faster than invoking the length operator.
+	-- See https://gitspartv.github.io/LuaJIT-Benchmarks/#test12.
+	local i = 0
 	serialize(value, function(text)
-		 -- Faster than table.insert(rope, text) on PUC Lua 5.1
-		rope[#rope + 1] = text
+		i = i + 1
+		rope[i] = text
 	end)
 	return table_concat(rope)
 end

builtin/common/tests/misc_helpers_spec.lua

@@ -230,3 +230,124 @@ describe("math", function()
 		assert.equal(0, math.round(-0.49999999999999994))
 	end)
 end)
+
+describe("dump", function()
+	local function test_expression(expr)
+		local chunk = assert(loadstring("return " .. expr))
+		local refs = {}
+		setfenv(chunk, {
+			setref = function(id)
+				refs[id] = {}
+				return function(fields)
+					for k, v in pairs(fields) do
+						refs[id][k] = v
+					end
+					return refs[id]
+				end
+			end,
+			getref = function(id)
+				return assert(refs[id])
+			end,
+		})
+		assert.equal(expr, dump(chunk()))
+	end
+
+	it("nil", function()
+		test_expression("nil")
+	end)
+
+	it("booleans", function()
+		test_expression("false")
+		test_expression("true")
+	end)
+
+	describe("numbers", function()
+		it("formats integers nicely", function()
+			test_expression("42")
+		end)
+		it("avoids misleading rounding", function()
+			test_expression("0.3")
+			assert.equal("0.30000000000000004", dump(0.1 + 0.2))
+		end)
+	end)
+
+	it("strings", function()
+		test_expression('"hello world"')
+		test_expression([["hello \"world\""]])
+	end)
+
+	describe("tables", function()
+		it("empty", function()
+			test_expression("{}")
+		end)
+
+		it("lists", function()
+			test_expression([[
+{
+	false,
+	true,
+	"foo",
+	1,
+	2,
+}]])
+		end)
+
+		it("number keys", function()
+test_expression([[
+{
+	[0.5] = false,
+	[1.5] = true,
+	[2.5] = "foo",
+}]])
+		end)
+
+		it("dicts", function()
+			test_expression([[{
+	a = 1,
+	b = 2,
+	c = 3,
+}]])
+		end)
+
+		it("mixed", function()
+			test_expression([[{
+	a = 1,
+	b = 2,
+	c = 3,
+	["d e"] = true,
+	"foo",
+	"bar",
+}]])
+		end)
+
+		it("nested", function()
+test_expression([[{
+	a = {
+		1,
+		{},
+	},
+	b = "foo",
+	c = {
+		[0.5] = 0.1,
+		[1.5] = 0.2,
+	},
+}]])
+		end)
+
+		it("circular references", function()
+test_expression([[setref(1){
+	child = {
+		parent = getref(1),
+	},
+	other_child = {
+		parent = getref(1),
+	},
+}]])
+		end)
+
+		it("supports variable indent", function()
+			assert.equal('{1,2,3,{foo = "bar",},}', dump({1, 2, 3, {foo = "bar"}}, ""))
+			assert.equal('{\n  "x",\n  "y",\n}', dump({"x", "y"}, "  "))
+		end)
+	end)
+end)

doc/lua_api.md

@@ -4162,9 +4162,11 @@ Helper functions
     * `obj`: arbitrary variable
     * `name`: string, default: `"_"`
     * `dumped`: table, default: `{}`
-* `dump(obj, dumped)`: returns a string which makes `obj` human-readable
-    * `obj`: arbitrary variable
-    * `dumped`: table, default: `{}`
+* `dump(value, indent)`: returns a string which makes `value` human-readable
+    * `value`: arbitrary value
+      * Circular references are supported. Every table is dumped only once.
+    * `indent`: string to use for indentation, default: `"\t"`
+      * `""` disables indentation & line breaks (compact output)
 * `math.hypot(x, y)`
     * Get the hypotenuse of a triangle with legs x and y.
       Useful for distance calculation.
@@ -7609,9 +7611,10 @@ Misc.
     * Example: `write_json({10, {a = false}})`,
       returns `'[10, {"a": false}]'`
 * `core.serialize(table)`: returns a string
-    * Convert a table containing tables, strings, numbers, booleans and `nil`s
-      into string form readable by `core.deserialize`
+    * Convert a value into string form readable by `core.deserialize`.
+    * Supports tables, strings, numbers, booleans and `nil`.
     * Support for dumping function bytecode is **deprecated**.
+    * Note: To obtain a human-readable representation of a value, use `dump` instead.
     * Example: `serialize({foo="bar"})`, returns `'return { ["foo"] = "bar" }'`
 * `core.deserialize(string[, safe])`: returns a table
     * Convert a string returned by `core.serialize` into a table

src/script/lua_api/l_areastore.cpp

@@ -330,7 +330,7 @@ void LuaAreaStore::Register(lua_State *L)
 		{"__gc", gc_object},
 		{0, 0}
 	};
-	registerClass(L, className, methods, metamethods);
+	registerClass<LuaAreaStore>(L, methods, metamethods);
 
 	// Can be created from Lua (AreaStore())
 	lua_register(L, className, create_object);