Skip to content
Commits on Oct 2, 2012
  1. @csrhodes

    slightly more reproducible builds

    in vop templates, (:constant <type>) eventually boiled down to a TYPEP
    test.  For (:constant fixnum) in the cross-compiler, this is not right,
    because the typep test runs on the host.  Use SB!XC:TYPEP instead, which
    means making sure that types are available to the cross-compiler (using
    DEF!TYPE instead of DEFTYPE).  Tested on x86-64; one issue on ppc fixed
    blindly.
    csrhodes committed Sep 20, 2012
  2. @csrhodes

    allow builds with :trace-file flags even with non-sbcl host compilers

    :allow-other-keys t is a marvellous thing
    csrhodes committed Sep 20, 2012
  3. @csrhodes

    mostly make the build deterministic

    From clisp, about 10 files still differ in xc fasls after these changes.
    At least one remaining issue is obvious (floats, where our float constants
    aren't representable on clisp) but there are other mysteries.
    csrhodes committed with csrhodes Sep 18, 2012
Commits on Oct 1, 2012
  1. @csrhodes
Commits on Sep 30, 2012
  1. @lichtblau

    Try to make test :deferrables-unblocked-by-lock more robust

      - Use a join thread, not a kill thread, so that we do not kill the
        background thread while it is currently performing the actual
        test.
    
      - Give the thread more time to start up, to make it more likely
        that its GRAB-MUTEX call needs to wait -- otherwise deferrables would
        not be unblocked.
    lichtblau committed Sep 30, 2012
  2. @lichtblau

    Do not use RETURN-FROM in MAKE-THREAD

    Fix for lp#1058799, regression in 1.0.56.55-f0da2f6.
    lichtblau committed Sep 30, 2012
Commits on Sep 27, 2012
  1. @lichtblau

    Mention safepoints in NEWS

    lichtblau committed Sep 18, 2012
  2. @lichtblau

    Fix the cheneygc build

    Thanks to Christoph Egger for the report.
    lichtblau committed Sep 27, 2012
  3. @lichtblau

    Disable ALLOCATION-INFORMATION.4 on gencgc for this release

    This test currently exposes a bug in the contrib, yet users might prefer
    to install the contrib in spite of this issue.  For this reason I am
    disabling the test for the moment.
    lichtblau committed Sep 27, 2012
Commits on Sep 23, 2012
  1. @nikodemus

    replace DEFINE-MORE-FUN with compiler smarts

      Now that the compiler knows how to implement (NTH I REST-ARG) efficiently,
      we don't need to play tricks with &MORE: the compiler does that for us.
    nikodemus committed Sep 23, 2012
  2. @nikodemus

    more funky &REST smartness

      Extend the earlier VALUES-LIST optimization for &REST arguments
      into other operations as well. For starters:
    
        CAR
        ELT
        ENDP
        FIRST
        IF
        LENGTH
        LIST-LENGTH
        NTH
    
      All of these can now access the hidden &MORE context when given a &REST
      argument that is not used by other operations, making it possible for the
      compiler to elide the entire rest-list allocation in those cases.
    nikodemus committed Sep 23, 2012
Commits on Sep 22, 2012
  1. @nikodemus

    fix "unable to read" compiler-error reporting during SBCL build

      By moving stream-error-position-info into early-extensions.lisp,
      so it's available early enough.
    nikodemus committed Sep 22, 2012
  2. @nikodemus

    add DEFINE-MORE-FUN, use it for vararg arithmetic functions

      More efficient than consing a rest-list -- even a stack-allocated one, and
      doesn't add extra DX cleanup frames to backtraces.
    
      Done this way instead of just using &MORE directly in lambda-lists in order
      to mangle the lambda-list into &REST shape for user-consumption.
    nikodemus committed Sep 22, 2012
  3. @nikodemus

    proclaim *CURRENT-THREAD* as always-bound

       Eliminates quite a few "check if it's bound" cases from the system, and
       protects against MAKUNBOUND.
    nikodemus committed Sep 19, 2012
  4. @jre

    Fix nonthreaded build.

    jre committed Sep 22, 2012
  5. @jre
Commits on Sep 19, 2012
  1. @lichtblau

    Add a safepoint-based mechanism to avoid SIGALRM for the TIMER facility

      - Retrofits the signal-free timer thread for Windows (thanks to
        Anton Kovalenko) to POSIXy platforms.
    
      - Provide os_* functions in the C runtime which simulate the win32
        API for waitable timers.
    
    Currently supported on Linux (timerfd), FreeBSD (kqueue), and SunOS
    (completion ports).  A tentative (untested) implementation is
    provided for Darwin's kqueue.
    lichtblau committed Sep 13, 2012
  2. @lichtblau

    New predicate THREAD-EPHEMERAL-P for certain system threads

    Adds an `ephemeral' flag to the thread structure and exports it.
    When true, we can assume that the thread will be started and stopped
    by the runtime automatically as needed.
    
    The flag is currently of a purely informational nature; it does not
    change the behaviour of thread-related functions.
    
    Thanks to Anton Kovalenko.
    lichtblau committed Sep 18, 2012
  3. @froydnj

    add truly-dynamic-extent declarations for various &rest number functions

    This change eliminates some spurious heap consing when using, e.g.
    (REDUCE #'+ ...).  We ought to be able to do a better job of optimizing
    REDUCE, but this is a helpful first step in any event.
    froydnj committed Sep 18, 2012
Commits on Sep 18, 2012
  1. @pkhuong @lichtblau

    Kill leftover threads after each test

    Otherwise, slightly broken tests manifest as hard failures in later
    tests.
    
    Thanks to Paul Khuong.
    pkhuong committed with lichtblau Aug 17, 2012
  2. @pkhuong @lichtblau

    Fix "exit from normal thread" test

     The test did not wait for the child to exit, resulting in spurious
     failures on loaded or single-core machines.
    pkhuong committed with lichtblau Aug 17, 2012
  3. @lichtblau

    Reduce the numbers of threads in test ATOMIC-UPDATE on 32bit platforms

    This test creates 1000 threads in a row -- too many for some 32bit
    platforms due to lack of memory or fragmentation.
    lichtblau committed Sep 18, 2012
  4. @lichtblau

    sunos-os.c: Do not lose on mmap failure

    Brings this platform in line with other POSIX backends.  It is desirable
    that we do not `lose' just because too many threads have been created,
    for reasons of user-friendliness -- and also because individual
    ill-behaved tests should not cause the entire run-tests process to fail.
    lichtblau committed Sep 18, 2012
  5. @lichtblau

    Minor repair work around odxprint

    Fixes the build on SunOS.
    lichtblau committed Sep 18, 2012
Commits on Sep 17, 2012
  1. @nikodemus
Commits on Sep 16, 2012
  1. @nikodemus

    fix unthreaded build

    nikodemus committed Sep 16, 2012
Commits on Sep 15, 2012
  1. @csrhodes

    Fix clisp cross-compile

    Problems:
    
    * least-positive-long-float is unrepresentable in clisp, ending up
      with being zero
    * clisp's decode-float returns a 0 exponent for a 0 float (rather
      than something large and negative)
    * everything got constant-folded in the cross-compiler, thus using
      the host's runtime.
    
    Solution: compute the value using sb!vm:double-float-digits/bias
    csrhodes committed Sep 15, 2012
  2. @nikodemus

    timeouts for WITH-MUTEX and WITH-RECURSIVE-LOCK

      * Also support :WAIT-P in WITH-RECUSIVE-LOCK.
    
      * Deprecate GET-MUTEX properly (been deprecated since early 2010, but didn't signal
        a compile-time warning, and we used it internally.)
    
      * Make WITH-MUTEX signal a runtime error when :VALUE is used and is other
        than current thread or NIL. Releasing it isn't going to work right if
        someone else holds it.
    nikodemus committed Sep 15, 2012
  3. @nikodemus
  4. @nikodemus
Commits on Sep 14, 2012
  1. @lichtblau

    Port safepoints to FreeBSD/x86

    lichtblau committed Sep 14, 2012
  2. @lichtblau

    x86-bsd-os.c: Include stdio

    Fixes the build on FreeBSD 9.1/x86.
    lichtblau committed Sep 14, 2012
  3. @lichtblau

    Port tests/expect.sh, foreign.test.sh to FreeBSD's shell

    Fixes clos.test.sh, foreign.test.sh on FreeBSD 9.1.
    lichtblau committed Sep 14, 2012
Commits on Sep 11, 2012
  1. @lichtblau

    Tweak slam.sh to be slightly more configurable regarding the host binary

    Developers building with an existing SBCL as a host, which is not
    installed as "sbcl" in the path, or which needs a --core option, can
    now set XC_LISP and XC_CORE prior to invoking slam.sh.  Previously,
    such a host could only be used with slam.sh after custom
    modifications to the script.
    
    The default behaviour is unchanged.
    lichtblau committed Jul 12, 2011
  2. @jsnell
Something went wrong with that request. Please try again.