Version 45 (modified by quchen, 4 years ago) (diff)


Release plans for GHC 7.8


The planned timeline for 7.8 is to have a feature freeze approximately at the time of ICFP, followed by a period of frantic bugfixing. A release candidate (and the 7.8 branch) will be formed after this, likely sometime in October.


We would like to fix all of the high and highest priority tickets in the 7.8.1 milestone, but there are currently a lot of them so this seems optimistic. Please feel free to take a ticket and help us!

Completed new features

The features already completed are documented in the release notes: docs/users_guide/7.8.1-notes.xml

Pending new features

The following new features are planned for 7.8. They are at varying degrees of completeness, and may not all make it in.

  • Patrick Palka has been working on making --make mode parallel with a -j flag. He has a stable set of improvements he'd like to see in 7.8: see the latest updates in #910.
  • Nicolas Frisby is adding a few Core-to-Core optimisations. They usually slightly improve performance, but can sometimes make a big difference, both good and bad. They are off by default.
    • (will merge by 6 Sept)-fdicts-strict will make all dictionary arguments strict if they are certainly not part of a knot.
    • (merged, cf af12cf66d1a416a135cb98b86717aba2cd247e1a and section in LateDmd) -ffun-to-thunk flag to revert a new (default) behavior that prevents GHC from creating sharing
    • (merged, cf LateDmd) -flate-dmd-anal will run demand analysis near the end of the pipeline. cf #7782.
      • TODO communally determine if -O2 should imply it

  • Pedro and Richard were working on tidying up the poly-kinded Typable, and propositional equality (gcast and friends) story. See #8132.
  • Geoff Mainland is working generalising the support for SSE-like instructions. ETA: September 13.
  • Austin Seipp would like to do some official ARMv7 binary releases with a working stage2 compiler and GHCi, but there are probably some bugs waiting here.
  • Dynamic GHCi (#3658). This is working in HEAD, and enabled if DYNAMIC_GHC_PROGRAMS=YES. Currently it's enabled by default if dynamic libraries are supported, except for FreeBSD and Windows. On FreeBSD the reason it's disabled is due to a bug in FreeBSD's rtld. This has been fixed, but we're waiting for the fix to make it into releases. This might be in time for 7.8, but certainly will be for 7.10. See #7819. On Windows, there are a couple of build time annoyances: -dynamic-too doesn't work on Windows, and linking takes a very long time when dynamic linking is used. There's no technical reason why it couldn't be enabled, though. The plan is/was to use dynamic GHCi on as many platforms as possible in 7.8, and to remove support for non-dynamic-ghci in HEAD soon after. See discussion in #8039, however.
  • Ryan Newton has added a larger set of atomic memory primops than were previously available. The code was developed on the atomics branch, which is now merged. The set of PrimOps may continue to expand slightly, but for now includes casArray#, casIntArray#, and fetchAddIntArray#. In future work, Carter Schonwald will provide optimized (inline) versions of these PrimOps, at least for the LLVM backend.
  • The Applicative-Monad warnings would preferably be integrated, see #8004. David Luposchainsky is driving this. (Current status: Warnings implemented, GHC compiles AMP-warning-free. What's left to do is getting the testsuite up to date with this. 13th is a very generous estimate.)

Features that will definitely not make it