#8095: TypeFamilies painfully slow -------------------------------------+------------------------------------- Reporter: MikeIzbicki | Owner: goldfire Type: bug | Status: new Priority: high | Milestone: 8.8.1 Component: Compiler (Type | Version: 7.6.3 checker) | Resolution: | Keywords: TypeFamilies Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: #5321, #11598, | Differential Rev(s): Phab:D3752, #12506, #13386 | Phab:D4766 Wiki Page: | -------------------------------------+------------------------------------- Comment (by hussein.aitlahcen): Hello everyone, My team experience the same issue when using Squeal extensively (https://github.com/morphismtech/squeal). I found two things: - GHC is unfolding the exported types (a lot of type families involved) in the `.hi` file. - Stack is asking GHC to dump the `.hi` to know which module it has to rebuild (https://github.com/commercialhaskell/stack/blob/master/src/Stack/Build/Execu...). This exact thing is making GHC crazy, forcing him to dump a pretty printed 3.5gb .txt (head of the file https://gist.githubusercontent.com/haitlahcen/3f1e83cdc78d48ed2100d51a0ea058...). Because of that, GHC was taking **12gb+** to compile some files. I found that omitting the interfaces pragmas with `-fomit-interface-pragmas` was helping a lot. In fact, no more unfolded type in the interface file, making the dumped .txt file going from **3.5gb** to **700kb** with a maximum allocated memory of **2.5gb**. I already patched Stack to stop asking for a dump and directly read the binary interface file (the startup time is greatly improved) and will probably push a PR. Finally, It looks like unfolding options for GHC does not apply to types (or I missed something?). I'm asking because of https://github.com/ghc/ghc/blob/886ddb27bfbbb52c41690cd29e2ab3ed80bf5450/com.... Since we are now able to compute stuff at type level, it would be handy to be able to control the unfolding depending on the `typeLevelExprSize`. Thanks for reading -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8095#comment:58 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler