comfortable merging the patches in the CVS version of SBCL.
108:
- (TIME (ROOM T)) reports more than 200 Mbytes consed even for
- a clean, just-started SBCL system. And it seems to be right:
- (ROOM T) can bring a small computer to its knees for a *long*
- time trying to GC afterwards. Surely there's some more economical
- way to implement (ROOM T).
+ ROOM issues:
- Daniel Barlow doesn't know what fixed this, but observes that it
- doesn't seem to be the case in 0.8.7.3 any more. Instead, (ROOM T)
- in a fresh SBCL causes
+ a) ROOM works by walking over the heap linearly, instead of
+ following the object graph. Hence, it report garbage objects that
+ are unreachable. (Maybe this is a feature and not a bug?)
- debugger invoked on a SB-INT:BUG in thread 5911:
- failed AVER: "(SAP= CURRENT END)"
-
- unless a GC has happened beforehand.
+ b) ROOM uses MAP-ALLOCATED-OBJECTS to walk the heap, which doesn't
+ check all pointers as well as it should, and can hence become
+ confused, leading to aver failures. As of 1.0.13.21 these (the
+ SAP= aver in particular) should be mostly under control, but push
+ ROOM hard enough and it still might croak.
117:
When the compiler inline expands functions, it may be that different
(print (incf start 22))
(print (incf start 26))))))
+ [ Update: 1.0.14.36 improved this quite a bit (20-25%) by
+ eliminating useless work from PROPAGATE-FROM-SETS -- but as alluded
+ below, maybe we should be smarter about when to decide a derived
+ type is "good enough". ]
+
This example could be solved with clever enough constraint
propagation or with SSA, but consider
c. The cross-compiler cannot inline functions defined in a non-null
lexical environment.
-206: ":SB-FLUID feature broken"
- (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
- Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
- the build.
-
207: "poorly distributed SXHASH results for compound data"
SBCL's SXHASH could probably try a little harder. ANSI: "the
intent is that an implementation should make a good-faith
For some more details see comments for (define-alien-type-method
(c-string :deport-gen) ...) in host-c-call.lisp.
-402: "DECLAIM DECLARATION does not inform the PCL code-walker"
- reported by Vincent Arkesteijn:
-
- (declaim (declaration foo))
- (defgeneric bar (x))
- (defmethod bar (x)
- (declare (foo x))
- x)
-
- ==> WARNING: The declaration FOO is not understood by
- SB-PCL::SPLIT-DECLARATIONS.
- Please put FOO on one of the lists SB-PCL::*NON-VAR-DECLARATIONS*,
- SB-PCL::*VAR-DECLARATIONS-WITH-ARG*, or
- SB-PCL::*VAR-DECLARATIONS-WITHOUT-ARG*.
- (Assuming it is a variable declaration without argument).
-
403: FORMAT/PPRINT-LOGICAL-BLOCK of CONDITIONs ignoring *PRINT-CIRCLE*
In sbcl-0.9.13.34,
(defparameter *c*
3: (SB-C::BOUND-FUNC ...)
4: (SB-C::%SINGLE-FLOAT-DERIVE-TYPE-AUX ...)
+ These are now fixed, but (COERCE HUGE 'SINGLE-FLOAT) still signals a
+ type-error at runtime. The question is, should it instead signal a
+ floating-point overflow, or return an infinity?
+
408: SUBTYPEP confusion re. OR of SATISFIES of not-yet-defined predicate
As reported by Levente M\'{e}sz\'{a}ros sbcl-devel 2006-02-20,
(aver (equal (multiple-value-list
implementation of read circularity, using a symbol as a marker for
the previously-referenced object.
-411: NAN issues on x86-64
- Test :NAN-COMPARISONS in float.pure.lisp fails on x86-64, and has been
- disabled on those platforms. Since x86 does not exhibit any problems
- the problem is probably with the new FP implementation.
+415: Issues creating large arrays on x86-64/Linux and x86/Darwin
+
+ (make-array (1- array-dimension-limit))
+
+ causes a GC invariant violation on x86-64/Linux, and
+ an unhandled SIGILL on x86/Darwin.
+
+416: backtrace confusion
+
+ (defun foo (x)
+ (let ((v "foo"))
+ (flet ((bar (z)
+ (oops v z)
+ (oops z v)))
+ (bar x)
+ (bar v))))
+ (foo 13)
+
+ gives the correct error, but the backtrace shows
+ 1: (SB-KERNEL:FDEFINITION-OBJECT 13 NIL)
+ as the second frame.
+
+418: SUBSEQ on lists doesn't support bignum indexes
-413: type-errors in ROOM
+ LIST-SUBSEQ* now has all the works necessary to support bignum indexes,
+ but it needs to be verified that changing the DEFKNOWN doesn't kill
+ performance elsewhere.
+
+ Other generic sequence functions have this problem as well.
+
+419: stack-allocated indirect closure variables are not popped
+
+ (locally (declare (optimize sb-c::stack-allocate-dynamic-extent
+ sb-c::stack-allocate-value-cells))
+ (defun bug419 (x)
+ (multiple-value-call #'list
+ (eval '(values 1 2 3))
+ (let ((x x))
+ (declare (dynamic-extent x))
+ (flet ((mget (y)
+ (+ x y))
+ (mset (z)
+ (incf x z)))
+ (declare (dynamic-extent #'mget #'mset))
+ ((lambda (f g) (eval `(progn ,f ,g (values 4 5 6)))) #'mget #'mset))))))
+
+ (ASSERT (EQUAL (BUG419 42) '(1 2 3 4 5 6))) => failure
+
+ Note: as of SBCL 1.0.26.29 this bug no longer affects user code, as
+ SB-C::STACK-ALLOCATE-VALUE-CELLS needs to be explicitly turned on for
+ that to happen. Proper fix for this bug requires (Nikodemus thinks)
+ storing the relevant LAMBDA-VARs in a :DYNAMIC-EXTENT cleanup, and
+ teaching stack analysis how to deal with them.
+
+420: The MISC.556 test from gcl/ansi-tests/misc.lsp fails hard.
+
+In sbcl-1.0.13 on Linux/x86, executing
+ (FUNCALL
+ (COMPILE NIL
+ '(LAMBDA (P1 P2)
+ (DECLARE
+ (OPTIMIZE (SPEED 1) (SAFETY 0) (DEBUG 0) (SPACE 0))
+ (TYPE (MEMBER 8174.8604) P1) (TYPE (MEMBER -95195347) P2))
+ (FLOOR P1 P2)))
+ 8174.8604 -95195347)
+interactively causes
+ SB-SYS:MEMORY-FAULT-ERROR: Unhandled memory fault at #x8.
+The gcl/ansi-tests/doit.lisp program terminates prematurely shortly after
+MISC.556 by falling into gdb with
+ fatal error encountered in SBCL pid 2827: Unhandled SIGILL
+unless the MISC.556 test is commented out.
+
+Analysis: + and a number of other arithmetic functions exhibit the
+same behaviour. Here's the underlying problem: On x86 we perform
+single-float + integer normally using double-precision, and then
+coerce the result back to single-float. (The FILD instruction always
+gives us a double-float, and unless we do MOVE-FROM-SINGLE it remains
+one. Or so it seems to me, and that would also explain the observed
+behaviour below.)
+
+During IR1 we derive the types for both
+
+ (+ <single> <integer>) ; uses double-precision
+ (+ <single> (FLOAT <integer> <single>)) ; uses single-precision
+
+and get a mismatch for a number of unlucky arguments. This leads to
+derived result type NIL, and ends up flushing the whole whole
+operation -- and finally we generate code without a return sequence,
+and fall through to whatever.
+
+The use of double-precision in the first case appears to be an
+(un)happy accident -- interval arithmetic gives us the
+double-precision result because that's what the backend does.
+
+ (+ 8172.0 (coerce -95195347 'single-float)) ; => -9.518717e7
+ (+ 8172.0 -95195347) ; => -9.5187176e7
+ (coerce (+ 8172.0 (coerce -95195347 'double-float)) 'single-float)
+ ; => -9.5187176e7
+
+Which should be fixed, the IR1, or the backend?
+
+421: READ-CHAR-NO-HANG misbehaviour on Windows Console:
+
+ It seems that on Windows READ-CHAR-NO-HANG hangs if the user
+ has pressed a key, but not yet enter (ie. SYSREAD-MAY-BLOCK-P
+ seems to lie if the OS is buffering input for us on Console.)
+
+ reported by Elliot Slaughter on sbcl-devel 2008/1/10.
+
+422: out-of-extent return not checked in safe code
+
+ (declaim (optimize safety))
+ (funcall (catch 't (block nil (throw 't (lambda () (return))))))
+
+behaves ...erratically. Reported by Kevin Reid on sbcl-devel
+2007-07-06. (We don't _have_ to check things like this, but we
+generally try to check returns in safe code, so we should here too.)
+
+424: toplevel closures and *CHECK-CONSISTENCY*
+
+ The following breaks under COMPILE-FILE if *CHECK-CONSISTENCY* is true.
+
+ (let ((exported-symbols-alist
+ (loop for symbol being the external-symbols of :cl
+ collect (cons symbol
+ (concatenate 'string
+ "#"
+ (string-downcase symbol))))))
+ (defun hyperdoc-lookup (symbol)
+ (cdr (assoc symbol exported-symbols-alist))))
+
+ (Test-case adapted from CL-PPCRE.)
+
+425: reading from closed streams
+
+ Reported by Damien Cassou on sbcl-devel. REPL transcript follows:
+
+ * (open ".bashrc" :direction :input)
+ #<SB-SYS:FD-STREAM for "file /home/cassou/.bashrc" {A6ADFC9}>
+ * (defparameter *s* *)
+ *S*
+ * (read-line *s*)
+ "# -*- Mode: Sh -*-"
+ * (read-line *s*)
+ "# Files you make look like rw-r--r--"
+ * (open-stream-p *s*)
+ T
+ * (close *s*)
+ T
+ * (open-stream-p *s*)
+ NIL
+ * (read-line *s*)
+ "umask 022"
+
+ The problem is with the fast path using ansi-stream-cin-buffer not hitting
+ closed-flame.
+
+426: inlining failure involving multiple nested calls
+
+ (declaim (inline foo))
+ (defun foo (x y)
+ (cons x y))
+ (defun bar (x)
+ (foo (foo x x) (foo x x)))
+ ;; shows a full call to FOO
+ (disassemble 'bar)
+ ;; simple way to test this programmatically
+ (let ((code (sb-c::fun-code-header #'bar))
+ (foo (sb-impl::fdefinition-object 'foo nil)))
+ (loop for i from sb-vm:code-constants-offset below (sb-kernel:get-header-data code)
+ do (assert (not (eq foo (sb-kernel:code-header-ref code i))))))
- (defvar *a* (make-array (expt 2 27)))
- (room)
+ This appears to be an ancient bug, inherited from CMUCL: reportedly
+ 18c does the same thing. RECOGNIZE-KNOWN-CALL correctly picks up only
+ one of the calls, but local call analysis fails to inline the call
+ for the second time. Nikodemus thinks (but is not 100% sure based on
+ very brief investigation) that the call that is not inlined is the
+ second nested one. A trivial fix is to call CHANGE-REF-LEAF in known
+ call for functions already inline converted there, but he is not sure
+ if this has adverse effects elsewhere.
- Causes a type-error on 32bit SBCL, as various byte-counts in ROOM
- implementation overrun fixnums.
+427: ANY-REG not good for primitive type T
+
+ ...which is true, of course, but the following should not complain
+ about it (on x86 and x86-64):
+
+ (sb-alien:with-alien ((buf (array (sb-alien:signed 8) 16))))
- This was fixed in 1.0.4.89, but the patch was reverted as it caused
- ROOM to cons sufficiently to make running it in a loop deadly on
- GENCGC: newly allocated objects survived to generation 1, where next
- call to ROOM would see them, and allocate even more...
+ reported by Stelian Ionescu on sbcl-devel.
- Reported by Faré Rideau on sbcl-devel.