#13344: Core string literal patch regresses compiler performance considerably -------------------------------------+------------------------------------- Reporter: bgamari | Owner: bgamari Type: bug | Status: new Priority: high | Milestone: 8.2.1 Component: Compiler | Version: 8.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by rwbarton):
Consequently, we came up with a few ideas for instead reducing the cost of the now-larger interface files. Happily, this effort may also have helpful effects beyond string literals. The gist is that we want make interface file deserialisation more lazy. Reid is working on a patch.
This patch is Phab:D3269.
Back to comment:16. Let's assume that this string-CSE thing makes no difference. Then all my questions in comment:16 remain. I just want to be sure that we are fixing the right problem!
I think it will make some substantial difference, but I suspect there is still something else going on. When your CSE fix is ready (I assume it is a simple change), I will try comparing * the commit before the top-level string patch, with * the top-level string patch plus your fix to CSE by doing a `validate --build-only` with detailed tracking of where time is being spent. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13344#comment:19 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler