#14519: Exponential runtime performance regression in GHC 8.2 + Data.Text.Lazy + Text.RE.TDFA -------------------------------------+------------------------------------- Reporter: ntc2 | Owner: tdammers Type: bug | Status: new Priority: normal | Milestone: 8.6.1 Component: Compiler | Version: 8.2.2 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: Runtime | Test Case: performance bug | https://github.com/ntc2/ghc-8.2.1 | -regex-lazy-text- | bug/tree/07b7bb32c6e90e8f2d2eada4b59943f37e632d53 Blocked By: | Blocking: Related Tickets: #13745, #14564 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by tdammers): Evidence so far: - We already suspected that whatever we're doing in `uncons` for lazy `Text` might break things. - The `next` function, which uses `uncons` a lot, is where things blow up in ticky profiles - Problem disappears when disabling rewrite rules. - Problem also disappears in profiling builds; this may however be due to rewrites and other optimizations not being fully applied in the presence of explicit cost centres. - Some `RULES` wrt `Text` fusion appear in the dump. So, updated hypothesis: This may not actually be a bug in GHC itself, but in the `text` library, particularly the fusion-related rules. Things like [https://github.com/haskell/text/pull/200 this PR] support the idea that `text`'s fusion behavior may not be perfect yet. If this is the case, then disabling the relevant RULES in `text` should also make the problem disappear. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14519#comment:38 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler