Commits on Jun 17, 2011
Commits on Apr 17, 2011
  1. Backward ported GHC7 patches by A Serrano & JP Moresmau

    based commits e6480ab
    and 9e3b6ed
    committed Jan 30, 2011
Commits on Oct 4, 2010
  1. Don't use goto-line.

    bos committed Oct 4, 2010
  2. Add bug and source URLs

    bos committed Oct 4, 2010
Commits on Oct 2, 2010
  1. Improve haskell-insert-language.

    Now checks to see if pragma is already in use before inserting.
    bos committed Oct 2, 2010
  2. Cache supported languages.

    bos committed Oct 2, 2010
  3. Cache scion-supported-pragmas.

    bos committed Oct 2, 2010
  4. Ignore more Mercurial junk

    bos committed Oct 2, 2010
  5. Ignore junk files under Mercurial

    bos committed Oct 2, 2010
Commits on Sep 16, 2010
  1. Add haddock target to Makefile.

    nominolo committed Sep 16, 2010
Commits on Sep 15, 2010
  1. Merge branch 'devel' of into devel

    Removes all vim stuff.  Vim-stuff should now live in a separate repo.
    nominolo committed Sep 15, 2010
  2. Docs and spelling.

    nominolo committed Sep 15, 2010
  3. Makefile tweaks.

    nominolo committed Sep 15, 2010
Commits on Aug 23, 2010
Commits on Jun 22, 2010
  1. Fix warning.

    nominolo committed Jun 22, 2010
Commits on Jun 18, 2010
  1. Move over a few more commands.

    nominolo committed Jun 18, 2010
Commits on Jun 17, 2010
  1. MEGA-COMMIT: Start support for two-tier architecture.

    To avoid various problems with the GHC API, we now use a
    multi-process architecture in Scion.  The Scion client still
    connects to only one Scion server, but this server now acts as a
    proxy and forwards each request to a number of Scion workers.
    The proxy and the server communicate via stdin/sdtout using a
    binary protocol, so this should be quite efficient.
    This avoids the following problems:
      - Memory bloat: GHC currently never shrinks a heap, so memory
        intensive processes (like indexing the package database)
        would cause high memory usage of the current session.  GHC
        API caches also tend to never get cleared, causing a similar
      - Static flags: Some flags (e.g., -prof) can only be set once
        per *process*.  It is now simpler to just start a new process
        with the new flags.  Similarly, the package DB can only
        safely be read once (at the beginning of the session).  If a
        dependency was missing, the user can now install it and it's
        easy to restart the Scion worker.
      - All these above problems could have been solved by requiring
        the Scion client to use and manage multiple Scion servers.
        By doing it in Scion, we hopefully avoid unnecessary code
      - During development of Scion we now mostly only need to
        recompile and restart the worker, not the proxy.  This could
        make it easier to work on Scion itself.
    I'm not quite sure yet how the final client interface should look
    like.  For now there's only one worker, but the path to using
    multiple workers is now much clearer.
    nominolo committed Jun 17, 2010
Commits on Apr 23, 2010
  1. Remove redundant import.

    nominolo committed Apr 23, 2010
  2. Fix warning.

    nominolo committed Apr 23, 2010
Commits on Apr 13, 2010