#2840: Top level string literals -------------------------------------+------------------------------------- Reporter: simonpj | Owner: (none) Type: bug | Status: closed Priority: lowest | Milestone: 7.6.2 Component: Compiler | Version: 6.10.1 Resolution: duplicate | Keywords: strings Operating System: Unknown/Multiple | Architecture: Type of failure: Runtime | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * keywords: => strings Old description:

At the moment GHC's internal language does not allow any top-level definitions of unlifted type, and for the most part rightly so. But consider this: {{{ f :: Int -> String f n = let a::Addr# = "foo" in let g y = ...a...g... in g n }}} Here we'd like to float the definitions out thus: {{{ a::Addr# = "foo" g y = ...a...g... f n = g n }}} This is much better. Usually this happens, but not here, because we don't allow a top-level binding for an `Addr#`. But really perhaps we should allow an exception for ''literals'', which can safely be bound at top level.

For literals other than strings, this doesn't make any difference, because we inline them freely. But for literal strings we don't want to make lots of copies of them; on the contrary we'd like to CSE identical strings. So it'd help to be able to bind them at top level.

Simon

New description: At the moment GHC's internal language does not allow any top-level definitions of unlifted type, and for the most part rightly so. But consider this: {{{#!hs f :: Int -> String f n = let a::Addr# = "foo" in let g y = ...a...g... in g n }}} Here we'd like to float the definitions out thus: {{{#!hs a::Addr# = "foo" g y = ...a...g... f n = g n }}} This is much better. Usually this happens, but not here, because we don't allow a top-level binding for an `Addr#`. But really perhaps we should allow an exception for ''literals'', which can safely be bound at top level. For literals other than strings, this doesn't make any difference, because we inline them freely. But for literal strings we don't want to make lots of copies of them; on the contrary we'd like to CSE identical strings. So it'd help to be able to bind them at top level. Simon -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/2840#comment:10 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler