-- main entrypoint for LÖVE -- -- Most apps can just use the default, but we need to override it to -- install a test harness. -- -- A test harness needs to check what the 'real' code did. -- To do this it needs to hook into primitive operations performed by code. -- Our hooks all go through the `App` global. When running tests they operate -- on fake screen, keyboard and so on. Once all tests pass, the App global -- will hook into the real screen, keyboard and so on. -- -- Scroll below this function for more details. function love.run() -- Tests always run at the start. App.run_tests() --? print('==') App.disable_tests() if App.initialize_globals then App.initialize_globals() end if App.initialize then App.initialize(love.arg.parseGameArguments(arg), arg) end if love.timer then love.timer.step() end local dt = 0 return function() if love.event then love.event.pump() for name, a,b,c,d,e,f in love.event.poll() do if name == "quit" then if not love.quit or not love.quit() then return a or 0 end end love.handlers[name](a,b,c,d,e,f) end end if love.timer then dt = love.timer.step() end if App.update then App.update(dt) end -- will pass 0 if love.timer is disabled if love.graphics and love.graphics.isActive() then love.graphics.origin() love.graphics.clear(love.graphics.getBackgroundColor()) if App.draw then App.draw() end love.graphics.present() end if love.timer then love.timer.sleep(0.001) end end end -- I've been building LÖVE apps for a couple of months now, and often feel -- stupid. I seem to have a smaller short-term memory than most people, and -- LÖVE apps quickly grow to a point where I need longer and longer chunks of -- focused time to make changes to them. The reason: I don't have a way to -- write tests yet. So before I can change any piece of an app, I have to -- bring into my head all the ways it can break. This isn't the case on other -- platforms, where I can be productive in 5- or 10-minute increments. Because -- I have tests. -- -- Most test harnesses punt on testing I/O, and conventional wisdom is to test -- business logic, not I/O. However, any non-trivial app does non-trivial I/O -- that benefits from tests. And tests aren't very useful if it isn't obvious -- after reading them what the intent is. Including the I/O allows us to write -- tests that mimic how people use our program. -- -- There's a major open research problem in testing I/O: how to write tests -- for graphics. Pixel-by-pixel assertions get too verbose, and they're often -- brittle because you don't care about the precise state of every last pixel. -- Except when you do. Pixels are usually -- but not always -- the trees -- rather than the forest. -- -- I'm not in the business of doing research, so I'm going to shave off a -- small subset of the problem for myself here: how to write tests about text -- (ignoring font, color, etc.) on a graphic screen. -- -- For example, here's how you may write a test of a simple text paginator -- like `less`: -- function test_paginator() -- -- initialize environment -- App.filesystem['/tmp/foo'] = filename([[ -- >abc -- >def -- >ghi -- >jkl -- ]]) -- App.args = {'/tmp/foo'} -- App.screen.init{ -- width=100 -- height=30 -- } -- App.font{ -- height=15 -- } -- App.run_after_keychord('pagedown') -- App.check_screen_contents{ -- y0='ghi' -- y15='' -- } -- end -- -- All functions starting with 'test_' (no modules) will run before the app -- runs "for real". Each such test is a fake run of our entire program. It can -- set as much of the environment as it wants, then run the app. Here we've -- got a 30px screen and a 15px font, so the screen has room for 2 lines. The -- file we're viewing has 4 lines. We assert that hitting the 'pagedown' key -- shows the third and fourth lines. -- -- Programs can still perform graphics, and all graphics will work in the real -- program. We can't yet write tests for graphics, though. Those pixels are -- basically always blank in tests. Really, there isn't even any -- representation for them. All our fake screens know about is lines of text, -- and what (x,y) coordinates they start at. There's some rudimentary support -- for concatenating all blobs of text that start at the same 'y' coordinate, -- but beware: text at y=100 is separate and non-overlapping with text at -- y=101. You have to use the test harness within these limitations for your -- tests to faithfully model the real world. -- -- One drawback of this approach: the y coordinate used depends on font size, -- which feels brittle. -- -- In the fullness of time App will support all side-effecting primitives -- exposed by LÖVE, but so far it supports just a rudimentary set of things I -- happen to have needed so far. App = {screen={}} function App.initialize_for_test() App.screen.init({width=100, height=50}) App.screen.contents = {} -- clear screen App.filesystem = {} if App.initialize_globals then App.initialize_globals() end end function App.screen.init(dims) App.screen.width = dims.width App.screen.height = dims.height end function App.screen.print(msg, x,y) local screen_row = 'y'..tostring(y) --? print('drawing "'..msg..'" at y '..tostring(y)) local screen = App.screen if screen.contents[screen_row] == nil then screen.contents[screen_row] = {} for i=0,screen.width-1 do screen.contents[screen_row][i] = '' end end if x < screen.width then screen.contents[screen_row][x] = msg end end -- LÖVE's Text primitive retains no trace of the string it was created from, -- so we'll wrap it for our tests. -- -- This implies that we need to hook any operations we need on Text objects. function App.newText(font, s) return {type='text', data=s, text=love.graphics.newText(font, s)} end function App.screen.draw(obj, x,y) if type(obj) == 'userdata' then -- ignore most things as graphics the test harness can't handle elseif obj.type == 'text' then App.screen.print(obj.data, x,y) else print(obj.type) assert(false) end end function App.run_after_textinput(t) App.textinput(t) App.screen.contents = {} App.draw() end function App.run_after_keychord(key) App.keychord_pressed(key) App.screen.contents = {} App.draw() end function App.run_after_mousepress(x,y, button) App.mousepressed(x,y, button) App.screen.contents = {} App.draw() end function App.width(text) return text.text:getWidth() end function App.screen.check(y, expected_contents, msg) --? print('checking for "'..expected_contents..'" at y '..tostring(y)) local screen_row = 'y'..tostring(y) local contents = '' if App.screen.contents[screen_row] == nil then error('no text at y '..tostring(y)) end for i,s in ipairs(App.screen.contents[screen_row]) do contents = contents..s end check_eq(contents, expected_contents, msg) end -- fake files function App.open_for_writing(filename) App.filesystem[filename] = '' return { write = function(self, s) App.filesystem[filename] = App.filesystem[filename]..s end, close = function(self) end } end function App.run_tests() local sorted_names = {} for name,binding in pairs(_G) do if name:find('test_') == 1 then table.insert(sorted_names, name) end end table.sort(sorted_names) for _,name in ipairs(sorted_names) do if name:find('test_') == 1 then App.initialize_for_test() _G[name]() end end print() end -- call this once all tests are run -- can't run any tests after this function App.disable_tests() -- have LÖVE delegate all handlers to App if they exist for name in pairs(love.handlers) do if App[name] then love.handlers[name] = App[name] end end -- test methods are disallowed outside tests App.screen.init = nil App.filesystem = nil App.run_after_textinput = nil App.run_after_keychord = nil -- other methods dispatch to real hardware App.screen.print = love.graphics.print App.newText = love.graphics.newText App.screen.draw = love.graphics.draw App.width = function(text) return text:getWidth() end App.open_for_writing = function(filename) return io.open(filename, 'w') end end