Repository : ssh://darcs.haskell.org//srv/darcs/ghc On branch :
http://hackage.haskell.org/trac/ghc/changeset/25a4e4b8d443974c3bbb754f1482df7bf09227aa >--------------------------------------------------------------- commit 25a4e4b8d443974c3bbb754f1482df7bf09227aa Author: Max Bolingbroke <batterseapo...@hotmail.com> Date: Fri Apr 20 18:03:43 2012 +0100 Further musing about the common-heap-vars hack for MSG >--------------------------------------------------------------- compiler/supercompile/Supercompile/Drive/MSG.hs | 24 +++++++++++++++++++++++ 1 files changed, 24 insertions(+), 0 deletions(-) diff --git a/compiler/supercompile/Supercompile/Drive/MSG.hs b/compiler/supercompile/Supercompile/Drive/MSG.hs index aa1bd64..c8510ef 100644 --- a/compiler/supercompile/Supercompile/Drive/MSG.hs +++ b/compiler/supercompile/Supercompile/Drive/MSG.hs @@ -313,6 +313,30 @@ type MSGResult = ((Deeds, Heap, Renaming, Stack), (Heap, Stack, Anned QA), (Deed -- 1. For every variable bound in the outgoing heap/stack of the {left,right} and common outgoing states.. -- 2. ..IF it is in the {left,right} incoming InScopeSet.. -- 3. ..then the thing it is bound in the {left,right} incoming state MUST have the same meaning as the new binding. +-- +-- Things which are definitely safe in my current approach: +-- 1. New left/right binders introduced for floated terms: they are chosen to be distinct from the existing left/right InScopeSets +-- 2. Elements of the left/right heap copied into each individual heap: clearly these are already in scope +-- 3. Suffixes of the left/right stack copied into each individual stack: clearly these are already in scope +-- +-- Things which are problematic: +-- 1. Common heap variables chosen for PendingVars/PendingTerms +-- 2. Common stack variables resulting from coupling update frames +-- +-- In both of the problematic cases, it is sufficient to ensure that either: +-- 1. The common variable gets a name not present in *either* input InScopeSet, OR +-- 2. If the common name *is* present in the left/right InScopeSet, then it is *identical* to the +-- left/right occurrence from which it was derived (because this is sufficient for us to be satisified +-- that the meanings of what they will get bound to will be the same) +-- +-- So if I have {f, g, x} InScope in the left input and {f, g, a} on the right, and: +-- a) I (msgFlexiVar f f) to f, then everything is kosher +-- b) I (msgFlexiVar f g) to f/g, then I have to choose a brand new name instead (e.g. b) +-- c) I (msgFlexiVar x a) to x/a, then things are OK (because each of x and a are only InScope on one side) +-- d) msgFlexiVar returns anything outside the set {f, g, x, a}, then everything is OK +-- +-- Basically we have to watch out for the case where the returned name is present in *both* input +-- InScopeSets *and* the left/right + common vars are not all identical. msg :: MSGMode -- ^ How to match -> State -- ^ Tieback semantics _______________________________________________ Cvs-ghc mailing list Cvs-ghc@haskell.org http://www.haskell.org/mailman/listinfo/cvs-ghc
