<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8">
<meta name="Generator" content="Microsoft Word 15 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0cm;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri",sans-serif;}
span.EmailStyle19
{mso-style-type:personal-reply;
font-family:"Calibri",sans-serif;
color:windowtext;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri",sans-serif;
mso-fareast-language:EN-US;}
.MsoPapDefault
{mso-style-type:export-only;
margin-top:6.0pt;
margin-right:0cm;
margin-bottom:6.0pt;
margin-left:0cm;}
@page WordSection1
{size:612.0pt 792.0pt;
margin:72.0pt 72.0pt 72.0pt 72.0pt;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]-->
</head>
<body lang="EN-GB" link="#0563C1" vlink="#954F72">
<div class="WordSection1">
<p class="MsoNormal">David<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Yes, quite right. Do make a ticket. But it’s not easy to see a truly robust way to fix this.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Selector thunks look like<o:p></o:p></p>
<p class="MsoNormal"> a = case x of (p,q) -> p<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">You example has thunks of the form<o:p></o:p></p>
<p class="MsoNormal"> a = case x of (_,b) -> case b of (c,_) -> c<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">We want this thunk to vanish entirely if x is bound to a pair whose second component is a pair. But, if x is bound to a pair, whose second component is not yet evaluated, we want the thunk to partially vanish, becoming<o:p></o:p></p>
<p class="MsoNormal"> a = case b of (c,_) -> c<o:p></o:p></p>
<p class="MsoNormal">where b is the second component of x. <o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">This seems hard in general.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">You could imagine translating the example into<o:p></o:p></p>
<p class="MsoNormal"> b = case x of (_,b) -> b<o:p></o:p></p>
<p class="MsoNormal"> a = case b of (c,_) -> c<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Here I have built two thunks rather than one, but each is (independently) a selector thunk. We get just the right thing happening if x is evaluated but its second component is note. Hooray. What’s not nice is that execution is slower
in the case where the selector-thunk mechanism doesn’t fire. <o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Another possibility. Make selector thunks carry a kind of “path” indicating which fields to pick out as they select a nested component of the data structure. A kind of selector-thunk chain.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Another possibility: for suitable selector-like thunks, compile special code that is run by the garbage collector.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">This deserves a ticket and a wiki page. Feel free to plunder the above.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Simon<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<div style="border:none;border-left:solid blue 1.5pt;padding:0cm 0cm 0cm 4.0pt">
<div>
<div style="border:none;border-top:solid #E1E1E1 1.0pt;padding:3.0pt 0cm 0cm 0cm">
<p class="MsoNormal"><b><span lang="EN-US">From:</span></b><span lang="EN-US"> ghc-devs <ghc-devs-bounces@haskell.org>
<b>On Behalf Of </b>David Feuer<br>
<b>Sent:</b> 09 March 2020 17:28<br>
<b>To:</b> ghc-devs <ghc-devs@haskell.org><br>
<b>Subject:</b> Selector thunks again<o:p></o:p></span></p>
</div>
</div>
<p class="MsoNormal"><o:p> </o:p></p>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
The fragility of this feature remains frustrating. A few days ago, I wrote this code for building a complete binary tree from its breadth-first traversal. (This is an improvement of a version by Will Ness.)<o:p></o:p></p>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
data Tree a<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
= Empty<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
| Node a (Tree a) (Tree a)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
deriving Show<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
-- An infinite list<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
data IL a = a :< IL a<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
infixr 5 :<<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
bft :: [a] -> Tree a<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
bft xs = tree<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
where<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
tree :< subtrees = go xs subtrees<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
go :: [a] -> IL (Tree a) -> IL (Tree a)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
go (a : as) ~(b1 :< ~(b2 :< bs)) = Node a b1 b2 :< go as bs<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
go [] _ = fix (Empty :<)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
When GHC compiles the lazy patterns, we get something essentially like this:<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
go (a : as) ys =<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
Node a<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
(case ys of b1 :< _ -> b1)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
(case ys of _ :< b2 :< _ -> b2)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
:<<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
go as (case ys of _ :< _ :< bs -> bs)<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
Now `<span style="font-family:"Arial",sans-serif">case ys of b1 :< _ -> b1` is a selector thunk, which is cool. The GC can reduce it as soon as either of the other thunks is forced. But neither of the other two case expressions is a selector thunk, so neither
will ever be reduced by the GC. If I consume the result tree using an inorder traversal, for example, then all the elements in the left subtree of the root will remain live until I start to consume the right subtree of the root.</span><o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
I can instead write this:<o:p></o:p></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<o:p> </o:p></p>
</div>
<div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"> go (a : as) ys = Node a b1 b2 :< go as bs<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"> where<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"> {-# NOINLINE b2bs #-}<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"> b1 :< b2bs = ys<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"> b2 :< bs = b2bs<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"><o:p> </o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif">Now all the suspended selections are selector thunks, so things should clean up nicely. There are still three problems, though. The first is that this is harder to read. The second is that now we have four suspended
selections instead of three. Finally, if b1 is not the first one forced, we'll need to force two thunks instead of one.<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif"><o:p> </o:p></span></p>
</div>
<div>
<p class="MsoNormal" style="mso-margin-top-alt:6.0pt;margin-right:0cm;margin-bottom:6.0pt;margin-left:0cm">
<span style="font-family:"Arial",sans-serif">Can't we do any better?<o:p></o:p></span></p>
</div>
</div>
</div>
</div>
</div>
</div>
</div>
</body>
</html>