Constrained dynamically extensible layout events: a patch
David, Spencer et al, are there any problems with restricting layoutMsg arguments to being members of the (open) Message class? I'd (strongly) prefer this over just Dynamic, since it restores most of the static checking we'd need. The only change for something like Mosaic would be to add the mosaic events to Message, with instance Message MyType, and to switch layoutMsg to sendMessage (its our 'throw' for dynamically extensible messages, in fact). The idea of constrained extensible messages is from SimonM's 06 HW paper on extensible exceptions. Comments? -- Don New patches: [Constrain layout messages to be members of a Message class Don Stewart <dons@cse.unsw.edu.au>**20070504075233 Using Typeables as the only constraint on layout messages is a bit scary, as a user can send arbitrary values to layoutMsg, whether they make sense or not: there's basically no type feedback on the values you supply to layoutMsg. Folloing Simon Marlow's dynamically extensible exceptions paper, we use an existential type, and a Message type class, to constrain valid arguments to layoutMsg to be valid members of Message. That is, a user writes some data type for messages their layout algorithm accepts: data MyLayoutEvent = Zoom | Explode | Flaming3DGlassEffect deriving (Typeable) and they then add this to the set of valid message types: instance Message MyLayoutEvent Done. We also reimplement the dynamic type check while we're here, to just directly use 'cast', rather than expose a raw fromDynamic/toDyn. With this, I'm much happier about out dynamically extensible layout event subsystem. ] { hunk ./Config.hs 157 - , ((modMask, xK_h ), layoutMsg Shrink) - , ((modMask, xK_l ), layoutMsg Expand) + , ((modMask, xK_h ), sendMessage Shrink) + , ((modMask, xK_l ), sendMessage Expand) hunk ./Operations.hs 19 -import Data.Dynamic ( Typeable, toDyn, fromDynamic ) hunk ./Operations.hs 77 +-- | Throw an (extensible) message value to the current Layout scheme, +-- possibly modifying how we layout the windows, then refresh. hunk ./Operations.hs 80 --- TODO, using Typeable for extensible stuff is a bit gunky. Check -- --- 'extensible exceptions' paper for other ideas. +-- TODO, this will refresh on Nothing. hunk ./Operations.hs 82 --- Basically this thing specifies the basic operations that vary between --- layouts. --- -data ShrinkOrExpand = Shrink | Expand deriving (Typeable, Eq) - -layoutMsg :: Typeable a => a -> X () -- FIXME: The below shouldn't refresh on Nothing -layoutMsg a = layout $ \x@(l:ls) -> maybe x (:ls) (modifyLayout l (toDyn a)) +sendMessage :: Message a => a -> X () +sendMessage m = layout $ \xs -> case xs of + [] -> [] + (l:ls) -> maybe xs (:ls) $ modifyLayout l (SomeMessage m) hunk ./Operations.hs 87 +------------------------------------------------------------------------ hunk ./Operations.hs 89 --- Standard layout algorithms: +-- Builtin layout algorithms: hunk ./Operations.hs 94 +-- +-- The latter algorithms support the following operations: hunk ./Operations.hs 97 -full :: Layout -tall, wide :: Rational -> Rational -> Layout +-- Shrink +-- Expand +-- + +data Resize = Shrink | Expand deriving (Typeable, Show) +instance Message Resize hunk ./Operations.hs 104 -full = Layout { doLayout = \sc ws -> [ (w,sc) | w <- ws ] - , modifyLayout = const Nothing } +full :: Layout +full = Layout { doLayout = \sc ws -> [ (w,sc) | w <- ws ] + , modifyLayout = const Nothing } -- no changes hunk ./Operations.hs 108 +tall, wide :: Rational -> Rational -> Layout hunk ./Operations.hs 112 - , modifyLayout = fmap f . fromDynamic } + , modifyLayout = fmap handler . fromMessage } hunk ./Operations.hs 114 - where f s = tall delta ((op s) frac delta) - op Shrink = (-) ; op Expand = (+) + where handler s = tall delta $ (case s of + Shrink -> (-) + Expand -> (+)) frac delta hunk ./XMonad.hs 20 + Typeable, Message, SomeMessage(..), fromMessage, hunk ./XMonad.hs 32 -import Data.Dynamic ( Dynamic ) +import Data.Typeable hunk ./XMonad.hs 40 - , layouts :: !(M.Map WorkspaceId [Layout]) -- ^ mapping of workspaces + , layouts :: !(M.Map WorkspaceId [Layout]) } -- ^ mapping of workspaces hunk ./XMonad.hs 42 - } hunk ./XMonad.hs 54 - , focusedBorder :: !Color -- ^ border color of the focused window - } + , focusedBorder :: !Color } -- ^ border color of the focused window hunk ./XMonad.hs 96 --- 'doLayout', a pure function to layout a Window set --- 'modifyLayout', +-- 'doLayout', a pure function to layout a Window set 'modifyLayout', +-- 'modifyLayout' can be considered a branch of an exception handler. +-- hunk ./XMonad.hs 100 - , modifyLayout :: Dynamic -> Maybe Layout } + , modifyLayout :: SomeMessage -> Maybe Layout } + +-- Based on ideas in /An Extensible Dynamically-Typed Hierarchy of Exceptions/, +-- Simon Marlow, 2006. Use extensible messages to the modifyLayout handler. +-- +-- User-extensible messages must be a member of this class: +-- +class (Typeable a, Show a) => Message a + +-- +-- A wrapped value of some type in the Message class. +-- +data SomeMessage = forall a. Message a => SomeMessage a + +-- +-- And now, unwrap a given, unknown Message type, performing a (dynamic) +-- type check on the result. +-- +fromMessage :: Message m => SomeMessage -> Maybe m +fromMessage (SomeMessage m) = cast m } Context: [refactoring, style, comments on new layout code Don Stewart <dons@cse.unsw.edu.au>**20070504023618] [use anyKey constant instead of magic number Jason Creighton <jcreigh@gmail.com>**20070504015043] [added mirrorLayout to mirror arbitrary layouts Jason Creighton <jcreigh@gmail.com>**20070504014653] [Fix layout switching order Spencer Janssen <sjanssen@cse.unl.edu>**20070503235632] [More Config.hs bugs Spencer Janssen <sjanssen@cse.unl.edu>**20070503234607] [Revert accidental change to Config.hs Spencer Janssen <sjanssen@cse.unl.edu>**20070503233148] [Add -fglasgow-exts for pattern guards. Properties.hs doesn't complain anymore Spencer Janssen <sjanssen@cse.unl.edu>**20070503214221] [Avoid the unsafe pattern match, in case Config.hs has no layouts Spencer Janssen <sjanssen@cse.unl.edu>**20070503214007] [add support for extensible layouts. David Roundy <droundy@darcs.net>**20070503144750] [comments. and stop tracing events to stderr Don Stewart <dons@cse.unsw.edu.au>**20070503075821] [-Wall police Don Stewart <dons@cse.unsw.edu.au>**20070503074937] [elaborate documentation in Config.hs Don Stewart <dons@cse.unsw.edu.au>**20070503074843] [Use updated refreshKeyboardMapping. Requires latest X11-extras Spencer Janssen <sjanssen@cse.unl.edu>**20070503032040] [run QC tests in addition to LOC test Jason Creighton <jcreigh@gmail.com>**20070503003202] [Add 'mod-n': refreshes current layout Spencer Janssen <sjanssen@cse.unl.edu>**20070503002252] [Fix tests after StackSet changes Spencer Janssen <sjanssen@cse.unl.edu>**20070502201622] [First steps to adding floating layer Spencer Janssen <sjanssen@cse.unl.edu>**20070502195917] [update motivational text using xmonad.org Don Stewart <dons@cse.unsw.edu.au>**20070502061859] [Sort dependencies in installation order Spencer Janssen <sjanssen@cse.unl.edu>**20070501204249] [Recommend X11-extras 0.1 Spencer Janssen <sjanssen@cse.unl.edu>**20070501204121] [elaborate description in .cabal Don Stewart <dons@cse.unsw.edu.au>**20070501035414] [use -fasm by default. Much faster Don Stewart <dons@cse.unsw.edu.au>**20070501031220] [Make border width configurable Spencer Janssen <sjanssen@cse.unl.edu>**20070430163515] [Add Config.hs-boot, remove defaultLayoutDesc from XConf Spencer Janssen <sjanssen@cse.unl.edu>**20070430162647] [Comment only Spencer Janssen <sjanssen@cse.unl.edu>**20070430161635] [Comment only Spencer Janssen <sjanssen@cse.unl.edu>**20070430161511] [check we never generate invalid stack sets Don Stewart <dons@cse.unsw.edu.au>**20070430065946] [view n . shift n . view i . shift i) x == x --> shift + view is invertible Don Stewart <dons@cse.unsw.edu.au>**20070430062901] [add rotate all and view idempotency tests Don Stewart <dons@cse.unsw.edu.au>**20070430055751] [Add XConf for values that don't change. Spencer Janssen <sjanssen@cse.unl.edu>**20070430054715] [Control.Arrow is suspicious, add an explicit import Spencer Janssen <sjanssen@cse.unl.edu>**20070430053623] [push is idempotent Don Stewart <dons@cse.unsw.edu.au>**20070430054345] [add two properties relating to empty window managers Don Stewart <dons@cse.unsw.edu.au>**20070430051016] [configurable border colors Jason Creighton <jcreigh@gmail.com>**20070430043859 This also fixes a bug where xmonad was assuming a 24-bit display, and just using, eg, 0xff0000 as an index into a colormap without querying the X server to determine the proper pixel value for "red". ] [new QC property: opening a window only affects the current screen Don Stewart <dons@cse.unsw.edu.au>**20070430050133] [a bit more precise about building non-empty stacksets for one test Don Stewart <dons@cse.unsw.edu.au>**20070430035729] [remove redundant call to 'delete' in 'shift' Don Stewart <dons@cse.unsw.edu.au>**20070430031151] [clean 'delete' a little Don Stewart <dons@cse.unsw.edu.au>**20070430025319] [shrink 'swap' Don Stewart <dons@cse.unsw.edu.au>**20070430024813] [shrink 'rotate' a little Don Stewart <dons@cse.unsw.edu.au>**20070430024525] [move size into Properties.hs Don Stewart <dons@cse.unsw.edu.au>**20070430021758] [don't need 'size' operation on StackSet Don Stewart <dons@cse.unsw.edu.au>**20070430015927] [avoid grabbing all keys when a keysym is undefined Jason Creighton <jcreigh@gmail.com>**20070428180046 XKeysymToKeycode() returns zero if the keysym is undefined. Zero also happens to be the value of AnyKey. ] [add homepage: field to .cabal file Don Stewart <dons@cse.unsw.edu.au>**20070429041011] [add fromList to Properties.hs Don Stewart <dons@cse.unsw.edu.au>**20070429035823] [move fromList into Properties.hs, -17 loc Don Stewart <dons@cse.unsw.edu.au>**20070429035804] [Further refactoring Spencer Janssen <sjanssen@cse.unl.edu>**20070426212257] [Refactor in Config.hs (no real changes) Spencer Janssen <sjanssen@cse.unl.edu>**20070426211407] [Add the manpage to extra-source-files Spencer Janssen <sjanssen@cse.unl.edu>**20070426014105] [add xmonad manpage David Lazar <davidlazar@styso.com>**20070426010812] [Remove toList Spencer Janssen <sjanssen@cse.unl.edu>**20070426005713] [Ignore numlock and capslock in keybindings Jason Creighton <jcreigh@gmail.com>**20070424013357] [Clear numlock bit Spencer Janssen <sjanssen@cse.unl.edu>**20070424010352] [force window border to 1px Jason Creighton <jcreigh@gmail.com>**20070423050824] [s/creigh// Don Stewart <dons@cse.unsw.edu.au>**20070423024026] [some other things to do Don Stewart <dons@cse.unsw.edu.au>**20070423023151] [Start TODOs for 0.2 Spencer Janssen <sjanssen@cse.unl.edu>**20070423021526] [update readme Don Stewart <dons@cse.unsw.edu.au>**20070422090507] [TAG 0.1 Spencer Janssen <sjanssen@cse.unl.edu>**20070422083033] Patch bundle hash: 58f3c7c9b1d98e93c2a0aa388fd91f7ee5567446
dons:
David, Spencer et al, are there any problems with restricting layoutMsg arguments to being members of the (open) Message class? I'd (strongly) prefer this over just Dynamic, since it restores most of the static checking we'd need.
Oh, I forgot an example. This is why Dynamic-only is scary: , ((modMask, xK_h ), layoutMsg "hey Joe!") , ((modMask, xK_l ), layoutMsg Expand) is well typed. Under the new system, it becomes a static error: Config.hs:157:43: No instance for (Message [Char]) arising from use of `sendMessage' at Config.hs:157:43-65 -- Don
dons:
dons:
David, Spencer et al, are there any problems with restricting layoutMsg arguments to being members of the (open) Message class? I'd (strongly) prefer this over just Dynamic, since it restores most of the static checking we'd need.
Oh, I forgot an example. This is why Dynamic-only is scary:
, ((modMask, xK_h ), layoutMsg "hey Joe!") , ((modMask, xK_l ), layoutMsg Expand)
is well typed. Under the new system, it becomes a static error:
Config.hs:157:43: No instance for (Message [Char]) arising from use of `sendMessage' at Config.hs:157:43-65
-- Don
Ah, and a version that actually applies, given Spencer's last changes. -- Don
On Fri, May 04, 2007 at 06:07:52PM +1000, Donald Bruce Stewart wrote:
dons:
David, Spencer et al, are there any problems with restricting layoutMsg arguments to being members of the (open) Message class? I'd (strongly) prefer this over just Dynamic, since it restores most of the static checking we'd need.
Oh, I forgot an example. This is why Dynamic-only is scary:
, ((modMask, xK_h ), layoutMsg "hey Joe!") , ((modMask, xK_l ), layoutMsg Expand)
is well typed. Under the new system, it becomes a static error:
Config.hs:157:43: No instance for (Message [Char]) arising from use of `sendMessage' at Config.hs:157:43-65
I personally like the ability to send unaccepted messages. It can't pose a problem on the receiving end, since every layout has to be able to handle messages it knows nothing about. So the only danger is foolish users who send messages that have no effect, but that really seems harmless to me. I guess getting better error checking is nice, but it seems like you're talking about error checking for something that really isn't likely to be a problem. This is different from the NonEmpty list issue, which still is working on avoiding the dangers of foolish users--who define an empty layout list--but in that case those foolish users could actually crash xmonad, rather than simply defining a key binding that has no effect. -- David Roundy http://www.darcs.net
droundy:
On Fri, May 04, 2007 at 06:07:52PM +1000, Donald Bruce Stewart wrote:
dons:
David, Spencer et al, are there any problems with restricting layoutMsg arguments to being members of the (open) Message class? I'd (strongly) prefer this over just Dynamic, since it restores most of the static checking we'd need.
Oh, I forgot an example. This is why Dynamic-only is scary:
, ((modMask, xK_h ), layoutMsg "hey Joe!") , ((modMask, xK_l ), layoutMsg Expand)
is well typed. Under the new system, it becomes a static error:
Config.hs:157:43: No instance for (Message [Char]) arising from use of `sendMessage' at Config.hs:157:43-65
I personally like the ability to send unaccepted messages. It can't pose a
You can still send unaccepted messages, but you can't accidentally send a non-message type anymore.
problem on the receiving end, since every layout has to be able to handle messages it knows nothing about. So the only danger is foolish users who
Right.
send messages that have no effect, but that really seems harmless to me. I guess getting better error checking is nice, but it seems like you're talking about error checking for something that really isn't likely to be a problem.
Well, its pretty light, and more static checking is always good, in my view. This guides the user a bit more towards valid code. Is there any downside?
This is different from the NonEmpty list issue, which still is working on avoiding the dangers of foolish users--who define an empty layout list--but in that case those foolish users could actually crash xmonad, rather than simply defining a key binding that has no effect.
And that issue is fixed today, too, safeLayouts = case defaultLayouts of [] -> (full, []); (x:xs) -> (x, xs) I think the combination of a Message class, and the non-empty layout handled via another mechanism, covers the holes we opened up. -- Don
On Sat, May 05, 2007 at 12:20:07AM +1000, Donald Bruce Stewart wrote:
Well, its pretty light, and more static checking is always good, in my view. This guides the user a bit more towards valid code. Is there any downside?
It makes the code more complicated without IMO sufficient justification. Stefan
participants (3)
-
David Roundy -
dons＠cse.unsw.edu.au -
Stefan O'Rear