ljkiwi - Free LuaJIT FFI kiwi (Cassowary derived) constraint solver.

Introduction

Kiwi is a reasonably efficient C++ implementation of the Cassowary constraint solving algorithm. It is an implementation of the algorithm as described in the paper "The Cassowary Linear Arithmetic Constraint Solving Algorithm" by Greg J. Badros and Alan Borning. The Kiwi implementation is not based on the original C++ implementation, but is a ground-up reimplementation with performance 10x to 500x faster in typical use. Cassowary constraint solving is a technique that is particularly well suited to user interface layout. It is the algorithm Apple uses for iOS and OS X Auto Layout.

There are a few Lua implementations or attempts. The SILE typesetting system has a pure Lua implementation of the original Cassowary code, which appears to be correct but is quite slow. There are two extant Lua ports of Kiwi, one that is based on a C rewrite of Kiwi. However testing of these was not encouraging with either segfaults or incorrect results. Since the C++ Kiwi library is well tested and widely used it was simpler to provide a LuaJIT FFI wrapper and use that. This package has no dependencies other than a C++14 toolchain to compile the included Kiwi library and a small C wrapper.

The Lua API has a pure Lua expression builder. There is of course some overhead to this, however in most cases expression building is infrequent and the underlying structures can be reused.

The wrapper is quite close to the Kiwi C++/Python port with a few naming changes.

Example

local kiwi = require("kiwi")
local Var = kiwi.Var

local Button = setmetatable({}, {
   __call = function(_, identifier)
      return setmetatable({
         left = Var(identifier .. " left"),
         width = Var(identifier .. " width"),
      }, {
         __tostring = function(self)
            return "Button(" .. self.left:value() .. ", " .. self.width:value() .. ")"
         end,
      })
   end,
})

local b1 = Button("b1")
local b2 = Button("b2")

local left_edge = Var("left")
local right_edge = Var("width")

local STRONG = kiwi.Strength.STRONG

-- stylua: ignore start
local constraints = {
   left_edge    :eq(0.0),
   -- two buttons are the same width
   b1.width     :eq(b2.width),
   -- button1 starts 50 from the left margin
   b1.left      :eq(left_edge + 50),
   -- button2 ends 50 from the right margin
   right_edge   :eq(b2.left + b2.width + 50),
   -- button2 starts at least 100 from the end of button1. This is the "elastic" constraint
   b2.left      :ge(b1.left + b1.width + 100),
   -- button1 has a minimum width of 87
   b1.width     :ge(87),
   -- button1 has a preferred width of 87
   b1.width     :eq(87, STRONG),
   -- button2 has minimum width of 113
   b2.width     :ge(113),
   -- button2 has a preferred width of 113
   b2.width     :eq(113, STRONG),
}
-- stylua: ignore end

local solver = kiwi.Solver()

for _, c in ipairs(constraints) do
   solver:add_constraint(c)
end

solver:update_vars()

print(b1) -- Button(50, 113)
print(b2) -- Button(263, 113)
print(left_edge:value()) -- 0
print(right_edge:value()) -- 426

solver:add_edit_var(right_edge, STRONG)
solver:suggest_value(right_edge, 500)
solver:update_vars()
print(b1) -- Button(50, 113)
print(b2) -- Button(337, 113)
print(right_edge:value()) -- 500

In addition to the expression builder there is a convenience constraints submodule with: pair_ratio, pair, and single to allow efficient construction of the most common simple expression types for GUI layout.

Documentation

WIP - However the API is fully annotated and will work with lua-language-server. Documentation can also be generated with lua-language-server.