Thanks much for these thoughtful questions and implementation perspectives.<br><br>Internally, Phooey works very like an imperative GUI program, pushing computations when inputs change. So I'd expect it to have pretty much the same performance properties. As you mentioned, lazy evaluation can play a role in making evaluation subtler. The laziness probably doesn't work out as nicely as we'd like, however. On an input change, a new output value is constructed lazily, but then a call is to the underlying low-level gui toolkit (eg wxWidgets), which will cause the output value to be fully evaluated. I'm assuming that wxHaskell doesn't do clever postponing & collapsing of output calls to wxWidgets. But perhaps it could do so, or Phooey could do so on top of wxHaskell. Probably a good idea, and it certainly hadn't occurred to me.
<br><br>The paper I'm working is partly about Phooey the interface, but more about systematically refactoring the imperative programming style into the Phooey interface and its implementation. This derivation is why I believe that Phooey is operationally very like the imperative approach (while presenting a radically different interface). It could be that Phooey internally operates like a
<span style="font-style: italic;">naive</span> imperative GUI program, of the sort I tend to write.<br><br>Thanks also for the suggested example and discussion of requirements of the underlying imperative toolkit.<br><br>
Best regards, - Conal<br><br><div><span class="gmail_quote">On 12/12/06, <b class="gmail_sendername">Brian Hulley</b> <<a href="mailto:email@example.com">firstname.lastname@example.org</a>> wrote:</span><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;">
Brian Hulley wrote:<br>> Conal Elliott wrote:<br>><br>>> GUIs are usually programmed in an "unnatural" style, in that ...<br>><br>> Also, suppose you have a gui consisting of an edit widget such that
<br>> when the user types something it gets lexed, parsed, and fontified as<br>> a Haskell program, ie from the user's point of view the widget<br>> maintains a syntax highlighted view of the code which is<br>> "continuously" updated. Would your system do the lexing/parsing for
<br>> every key that was pressed or only once, when the widget needs to be<br>> re-rendered (a typical gui for Windows would only propagate changes<br>> and re-draw the gui (ie lex/parse/fontify) when there are no event
<br>> messages pending)?<br><br>Thinking about this more, a better example would be a gui with 2 widgets: an<br>edit widget to type in code and a list widget which displays a list of top<br>level functions defined in that code. (The code could be Haskell or some
<br>simple toy language.)<br>Then the question becomes: how to set things up such that the code in the<br>edit widget is only parsed when the list widget (and/or edit widget) is<br>asked to render it's contents rather than every time the user presses a key.
<br>I think perhaps the problem would be solved by the edit widget passing a<br>lazy parse tree (ie an expression evaluating to the parse tree) to the list<br>widget.<br><br>But an interesting thing is if you compare this to imperative guis, in the
<br>functional gui each widget passes lazy expressions eagerly whereas in<br>imperative guis each widget passes eagerly evaluated expressions lazily ie<br>functional is "push messages" whereas imperative is "pull changes when I'm
<br>asked to pick or render something etc and I'm dirty". (Have I got this<br>wrong?)<br><br>I imagine that the overhead of eager passing in the functional gui is a<br>small price to pay for the ease of using a functional gui, since the
<br>important thing is that in both guis, the hard work of parsing the code in<br>the edit buffer would only be done on demand, unless I'm much mistaken.<br><br>If you're still thinking of examples for your paper it would also be really
<br>great to see how you'd create a widget that displays the result of another<br>process (eg a window that displays the output as ghc compiles a program) or<br>some other example of how to use the IO monad inside a widget for those
<br>unfamiliar with combining arrows with monads.<br><br>It would also be useful to know more about the relationship between the<br>functional gui and WxWidgets to see what characteristics an imperative gui<br>toolkit must have in order to be usable with it. (Like the extremely useful
<br>discussion of low level details in the Fudgets thesis.)<br><br>Best regards, Brian.<br>--<br><a href="http://www.metamilk.com">http://www.metamilk.com</a><br><br></blockquote></div><br>