3 Bugs can be reported on the help mailing list
4 sbcl-help@lists.sourceforge.net
5 or on the development mailing list
6 sbcl-devel@lists.sourceforge.net
8 Please include enough information in a bug report that someone reading
9 it can reproduce the problem, i.e. don't write
10 Subject: apparent bug in PRINT-OBJECT (or *PRINT-LENGTH*?)
11 PRINT-OBJECT doesn't seem to work with *PRINT-LENGTH*. Is this a bug?
13 Subject: apparent bug in PRINT-OBJECT (or *PRINT-LENGTH*?)
14 In sbcl-1.2.3 running under OpenBSD 4.5 on my Alpha box, when
15 I compile and load the file
16 (DEFSTRUCT (FOO (:PRINT-OBJECT (LAMBDA (X Y)
17 (LET ((*PRINT-LENGTH* 4))
20 then at the command line type
22 the program loops endlessly instead of printing the object.
27 There is also some information on bugs in the manual page and
28 in the TODO file. Eventually more such information may move here.
30 The gaps in the number sequence belong to old bug descriptions which
31 have gone away (typically because they were fixed, but sometimes for
32 other reasons, e.g. because they were moved elsewhere).
36 DEFSTRUCT almost certainly should overwrite the old LAYOUT information
37 instead of just punting when a contradictory structure definition
38 is loaded. As it is, if you redefine DEFSTRUCTs in a way which
39 changes their layout, you probably have to rebuild your entire
40 program, even if you know or guess enough about the internals of
41 SBCL to wager that this (undefined in ANSI) operation would be safe.
43 3: "type checking of structure slots"
45 ANSI specifies that a type mismatch in a structure slot
46 initialization value should not cause a warning.
48 This one might not be fixed for a while because while we're big
49 believers in ANSI compatibility and all, (1) there's no obvious
50 simple way to do it (short of disabling all warnings for type
51 mismatches everywhere), and (2) there's a good portable
52 workaround, and (3) by their own reasoning, it looks as though
53 ANSI may have gotten it wrong. ANSI justifies this specification
55 The restriction against issuing a warning for type mismatches
56 between a slot-initform and the corresponding slot's :TYPE
57 option is necessary because a slot-initform must be specified
58 in order to specify slot options; in some cases, no suitable
60 However, in SBCL (as in CMU CL or, for that matter, any compiler
61 which really understands Common Lisp types) a suitable default
62 does exist, in all cases, because the compiler understands the
63 concept of functions which never return (i.e. has return type NIL).
64 Thus, as a portable workaround, you can use a call to some
65 known-never-to-return function as the default. E.g.
67 (BAR (ERROR "missing :BAR argument")
68 :TYPE SOME-TYPE-TOO-HAIRY-TO-CONSTRUCT-AN-INSTANCE-OF))
70 (DECLAIM (FTYPE (FUNCTION () NIL) MISSING-ARG))
71 (DEFUN REQUIRED-ARG () ; workaround for SBCL non-ANSI slot init typing
72 (ERROR "missing required argument"))
74 (BAR (REQUIRED-ARG) :TYPE TRICKY-TYPE-OF-SOME-SORT)
75 (BLETCH (REQUIRED-ARG) :TYPE TRICKY-TYPE-OF-SOME-SORT)
76 (N-REFS-SO-FAR 0 :TYPE (INTEGER 0)))
77 Such code should compile without complaint and work correctly either
78 on SBCL or on any other completely compliant Common Lisp system.
81 And as long as we're wishing, it would be awfully nice if INSPECT could
82 also report on closures, telling about the values of the bound variables.
84 Currently INSPECT and DESCRIBE do show the values, but showing the
85 names of the bindings would be even nicer.
88 The compiler assumes that any time a function of declared FTYPE
89 doesn't signal an error, its arguments were of the declared type.
90 E.g. compiling and loading
91 (DECLAIM (OPTIMIZE (SAFETY 3)))
92 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
94 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
96 (COND ((> (FACTORIAL X) 1.0E6)
97 (FORMAT T "too big~%"))
99 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
101 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
104 will cause the INTEGERP case to be selected, giving bogus output a la
106 This violates the "declarations are assertions" principle.
107 According to the ANSI spec, in the section "System Class FUNCTION",
108 this is a case of "lying to the compiler", but the lying is done
109 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
110 not by the unexpectedly general definition of FACTORIAL. In any case,
111 "declarations are assertions" means that lying to the compiler should
112 cause an error to be signalled, and should not cause a bogus
113 result to be returned. Thus, the compiler should not assume
114 that arbitrary functions check their argument types. (It might
115 make sense to add another flag (CHECKED?) to DEFKNOWN to
116 identify functions which *do* check their argument types.)
117 (Also, verify that the compiler handles declared function
118 return types as assertions.)
121 The definitions of SIGCONTEXT-FLOAT-REGISTER and
122 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
123 supported on FreeBSD because the floating point state is not saved,
124 but at least as of FreeBSD 4.0, the floating point state *is* saved,
125 so they could be supported after all. Very likely
126 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
129 Compiling and loading
130 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
132 then requesting a BACKTRACE at the debugger prompt gives no information
133 about where in the user program the problem occurred.
135 (this is apparently mostly fixed on the SPARC, PPC, and x86 architectures:
136 while giving the backtrace the non-x86 systems complains about "unknown
137 source location: using block start", but apart from that the
138 backtrace seems reasonable. On x86 this is masked by bug 353. See
139 tests/debug.impure.lisp for a test case)
142 Using the pretty-printer from the command prompt gives funny
143 results, apparently because the pretty-printer doesn't know
144 about user's command input, including the user's carriage return
145 that the user, and therefore the pretty-printer thinks that
146 the new output block should start indented 2 or more characters
147 rightward of the correct location.
150 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
151 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
152 crashes SBCL. In general tracing anything which is used in the
153 implementation of TRACE is likely to have the same problem.
156 ANSI says in one place that type declarations can be abbreviated even
157 when the type name is not a symbol, e.g.
158 (DECLAIM ((VECTOR T) *FOOVECTOR*))
159 SBCL doesn't support this. But ANSI says in another place that this
160 isn't allowed. So it's not clear this is a bug after all. (See the
161 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
164 (Actually this has changed changed since, and types as above are
165 now supported. This may be a bug.)
168 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
169 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
170 the mean of the distribution can be systematically O(0.1%) wrong.
171 Just increasing R-I-E-B is probably not a good solution, since
172 it would decrease efficiency more than is probably necessary. Perhaps
173 using some sort of accept/reject method would be better.
176 Internally the compiler sometimes evaluates
177 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
178 (I stumbled across this when I added an
179 (assert (not (eq type1 *wild-type*)))
180 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
181 in a type context should probably be translated to T, and so it's
182 probably wrong to ask whether it's equal to the T type and then (using
183 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
184 (I haven't tried to investigate this bug enough to guess whether
185 there might be any user-level symptoms.)
187 In fact, the type system is likely to depend on this inequality not
188 holding... * is not equivalent to T in many cases, such as
189 (VECTOR *) /= (VECTOR T).
192 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
193 CL), out-of-line structure slot setters are horribly inefficient
194 whenever the type of the slot is declared, because out-of-line
195 structure slot setters are implemented as closures to save space,
196 so the compiler doesn't compile the type test into code, but
197 instead just saves the type in a lexical closure and interprets it
199 To exercise the problem, compile and load
200 (cl:in-package :cl-user)
202 (bar (error "missing") :type bar))
205 (loop (setf (foo-bar *foo*) x)))
207 (defvar *bar* (make-bar))
208 (defvar *foo* (make-foo :bar *bar*))
209 (defvar *setf-foo-bar* #'(setf foo-bar))
211 (loop (funcall *setf-foo-bar* x *foo*)))
212 then run (WASTREL1 *BAR*) or (WASTREL2 *BAR*), hit Ctrl-C, and
213 use BACKTRACE, to see it's spending all essentially all its time
214 in %TYPEP and VALUES-SPECIFIER-TYPE and so forth.
215 One possible solution would be simply to give up on
216 representing structure slot accessors as functions, and represent
217 them as macroexpansions instead. This can be inconvenient for users,
218 but it's not clear that it's worse than trying to help by expanding
219 into a horribly inefficient implementation.
220 As a workaround for the problem, #'(SETF FOO) expressions
221 can be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
222 (defmacro efficient-setf-function (place-function-name)
223 (or #+sbcl (and (sb-int:info :function :accessor-for place-function-name)
224 ;; a workaround for the problem, encouraging the
225 ;; inline expansion of the structure accessor, so
226 ;; that the compiler can optimize its type test
227 (let ((new-value (gensym "NEW-VALUE-"))
228 (structure-value (gensym "STRUCTURE-VALUE-")))
229 `(lambda (,new-value ,structure-value)
230 (setf (,place-function-name ,structure-value)
232 ;; no problem, can just use the ordinary expansion
233 `(function (setf ,place-function-name))))
236 There's apparently a bug in CEILING optimization which caused
237 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
238 applied the patches to SBCL and they didn't seem to cause problems
239 (as reported sbcl-devel 2001-05-04). However, since the patches
240 modify nontrivial code which was apparently written incorrectly
241 the first time around, until regression tests are written I'm not
242 comfortable merging the patches in the CVS version of SBCL.
247 a) ROOM works by walking over the heap linearly, instead of
248 following the object graph. Hence, it report garbage objects that
249 are unreachable. (Maybe this is a feature and not a bug?)
251 b) ROOM uses MAP-ALLOCATED-OBJECTS to walk the heap, which doesn't
252 check all pointers as well as it should, and can hence become
253 confused, leading to aver failures. As of 1.0.13.21 these (the
254 SAP= aver in particular) should be mostly under control, but push
255 ROOM hard enough and it still might croak.
258 When the compiler inline expands functions, it may be that different
259 kinds of return values are generated from different code branches.
260 E.g. an inline expansion of POSITION generates integer results
261 from one branch, and NIL results from another. When that inline
262 expansion is used in a context where only one of those results
265 (aref *a1* (position x *a2*)))
266 and the compiler can't prove that the unacceptable branch is
267 never taken, then bogus type mismatch warnings can be generated.
268 If you need to suppress the type mismatch warnings, you can
269 suppress the inline expansion,
271 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
272 (aref *a1* (position x *a2*)))
273 or, sometimes, suppress them by declaring the result to be of an
276 (aref *a1* (the integer (position x *a2*))))
278 This is not a new compiler problem in 0.7.0, but the new compiler
279 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
280 more conspicuous. If you don't need performance from these functions,
281 and the bogus warnings are a nuisance for you, you can return to
282 your pre-0.7.0 state of grace with
283 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
288 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
289 the entire lexical environment at the point of MACROLET available
290 in the bodies of the macroexpander functions. In particular, it
291 allows the function bodies (which run at compile time) to try to
292 access lexical variables (which are only defined at runtime).
293 It doesn't even issue a warning, which is bad.
295 The SBCL behavior arguably conforms to the ANSI spec (since the
296 spec says that the behavior is undefined, ergo anything conforms).
297 However, it would be better to issue a compile-time error.
298 Unfortunately I (WHN) don't see any simple way to detect this
299 condition in order to issue such an error, so for the meantime
300 SBCL just does this weird broken "conforming" thing.
302 The ANSI standard says, in the definition of the special operator
304 The macro-expansion functions defined by MACROLET are defined
305 in the lexical environment in which the MACROLET form appears.
306 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
307 the local macro definitions in a MACROLET, but the consequences
308 are undefined if the local macro definitions reference any
309 local variable or function bindings that are visible in that
311 Then it seems to contradict itself by giving the example
313 (macrolet ((fudge (z)
314 ;The parameters x and flag are not accessible
315 ; at this point; a reference to flag would be to
316 ; the global variable of that name.
317 ` (if flag (* ,z ,z) ,z)))
318 ;The parameters x and flag are accessible here.
322 The comment "a reference to flag would be to the global variable
323 of the same name" sounds like good behavior for the system to have.
324 but actual specification quoted above says that the actual behavior
327 (Since 0.7.8.23 macroexpanders are defined in a restricted version
328 of the lexical environment, containing no lexical variables and
329 functions, which seems to conform to ANSI and CLtL2, but signalling
330 a STYLE-WARNING for references to variables similar to locals might
334 Ideally, uninterning a symbol would allow it, and its associated
335 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
336 at least as of sbcl-0.7.0, this isn't the case. Information about
337 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
338 essentially in ordinary (non-weak) hash tables keyed by symbols.
339 Thus, once a system has an entry in this system, it tends to live
340 forever, even when it is uninterned and all other references to it
345 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
346 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
347 COMPLEX implementation didn't deal with this, and hasn't been
348 upgraded to do so. (This doesn't seem to be a high priority
349 conformance problem, since seems hard to construct useful code
352 [ partially fixed by CSR in 0.8.17.17 because of a PFD ansi-tests
353 report that (COMPLEX RATIO) was failing; still failing on types of
354 the form (AND NUMBER (SATISFIES REALP) (SATISFIES ZEROP)). ]
356 b. (fixed in 0.8.3.43)
359 Floating point errors are reported poorly. E.g. on x86 OpenBSD
362 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
363 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
364 No traps are enabled? How can this be?
365 It should be possible to be much more specific (overflow, division
366 by zero, etc.) and of course the "How can this be?" should be fixable.
368 See also bugs #45.c and #183
371 (reported by Robert E. Brown 2002-04-16)
372 When a function is called with too few arguments, causing the
373 debugger to be entered, the uninitialized slots in the bad call frame
374 seem to cause GCish problems, being interpreted as tagged data even
375 though they're not. In particular, executing ROOM in the
376 debugger at that point causes AVER failures:
379 * (lisp-implementation-version)
385 failed AVER: "(SAP= CURRENT END)"
386 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
387 isn't too surprising since there are many differences in stack
388 implementation and GC conservatism between the X86 and other ports.)
390 (Can't reproduce on x86 linux as of 1.0.20.23 - MGL)
392 This is probably the same bug as 216
395 The compiler sometimes tries to constant-fold expressions before
396 it checks to see whether they can be reached. This can lead to
397 bogus warnings about errors in the constant folding, e.g. in code
400 (WRITE-STRING (> X 0) "+" "0"))
401 compiled in a context where the compiler can prove that X is NIL,
402 and the compiler complains that (> X 0) causes a type error because
403 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
404 caused a full WARNING, which made the bug really annoying because then
405 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
406 code. Since then the warning has been downgraded to STYLE-WARNING,
407 so it's still a bug but at least it's a little less annoying.
409 183: "IEEE floating point issues"
410 Even where floating point handling is being dealt with relatively
411 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
412 accrued-exceptions and current-exceptions part of the fp control
413 word don't seem to bear much relation to reality. E.g. on
417 debugger invoked on condition of type DIVISION-BY-ZERO:
418 arithmetic error DIVISION-BY-ZERO signalled
419 0] (sb-vm::get-floating-point-modes)
421 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
422 :ROUNDING-MODE :NEAREST
423 :CURRENT-EXCEPTIONS NIL
424 :ACCRUED-EXCEPTIONS (:INEXACT)
427 * (sb-vm::get-floating-point-modes)
428 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
429 :ROUNDING-MODE :NEAREST
430 :CURRENT-EXCEPTIONS (:INEXACT)
431 :ACCRUED-EXCEPTIONS (:INEXACT)
434 188: "compiler performance fiasco involving type inference and UNION-TYPE"
438 (declare (optimize (safety 3)))
439 (declare (optimize (compilation-speed 2)))
440 (declare (optimize (speed 1) (debug 1) (space 1)))
442 (declare (type (integer 0) start))
443 (print (incf start 22))
444 (print (incf start 26))
445 (print (incf start 28)))
447 (declare (type (integer 0) start))
448 (print (incf start 22))
449 (print (incf start 26)))
451 (declare (type (integer 0) start))
452 (print (incf start 22))
453 (print (incf start 26))))))
455 [ Update: 1.0.14.36 improved this quite a bit (20-25%) by
456 eliminating useless work from PROPAGATE-FROM-SETS -- but as alluded
457 below, maybe we should be smarter about when to decide a derived
458 type is "good enough". ]
460 This example could be solved with clever enough constraint
461 propagation or with SSA, but consider
466 The careful type of X is {2k} :-(. Is it really important to be
467 able to work with unions of many intervals?
469 191: "Miscellaneous PCL deficiencies"
470 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
471 a. DEFCLASS does not inform the compiler about generated
472 functions. Compiling a file with
476 (WITH-SLOTS (A-CLASS-X) A
478 results in a STYLE-WARNING:
480 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
482 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
483 Mai points out that the declamation of functions is in fact
484 incorrect in some cases (most notably for structure
485 classes). This means that at present erroneous attempts to use
486 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
487 won't get the corresponding STYLE-WARNING.
489 [much later, in 2006-08] in fact it's no longer erroneous to use
490 WITH-SLOTS on structure-classes. However, including :METACLASS
491 STRUCTURE-CLASS in the class definition gives a whole bunch of
492 function redefinition warnings, so we're still not good to close
495 c. (fixed in 0.8.4.23)
497 201: "Incautious type inference from compound types"
498 a. (reported by APD sbcl-devel 2002-09-17)
500 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
502 (FORMAT NIL "~S IS ~S, Y = ~S"
509 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
513 (declare (type (array * (4 4)) x))
515 (setq x (make-array '(4 4)))
516 (adjust-array y '(3 5))
517 (= (array-dimension y 0) (eval `(array-dimension ,y 0)))))
519 * (foo (make-array '(4 4) :adjustable t))
522 205: "environment issues in cross compiler"
523 (These bugs have no impact on user code, but should be fixed or
525 a. Macroexpanders introduced with MACROLET are defined in the null
527 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
528 the null lexical environment.
529 c. The cross-compiler cannot inline functions defined in a non-null
532 207: "poorly distributed SXHASH results for compound data"
533 SBCL's SXHASH could probably try a little harder. ANSI: "the
534 intent is that an implementation should make a good-faith
535 effort to produce hash-codes that are well distributed
536 within the range of non-negative fixnums". But
537 (let ((hits (make-hash-table)))
540 (let* ((ij (cons i j))
541 (newlist (push ij (gethash (sxhash ij) hits))))
543 (format t "~&collision: ~S~%" newlist))))))
544 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
545 would be an obvious way of fix. Maybe it would be acceptably efficient
546 to redo MIX using a lookup into a 256-entry s-box containing
547 29-bit pseudorandom numbers?
549 211: "keywords processing"
550 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
551 number of keyword arguments.
553 212: "Sequence functions and circular arguments"
554 COERCE, MERGE and CONCATENATE go into an infinite loop when given
555 circular arguments; it would be good for the user if they could be
556 given an error instead (ANSI 17.1.1 allows this behaviour on the part
557 of the implementation, as conforming code cannot give non-proper
558 sequences to these functions. MAP also has this problem (and
559 solution), though arguably the convenience of being able to do
560 (MAP 'LIST '+ FOO '#1=(1 . #1#))
561 might be classed as more important (though signalling an error when
562 all of the arguments are circular is probably desireable).
564 213: "Sequence functions and type checking"
565 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
566 check that it will return a sequence of the given type. Fixing
567 it along the same lines as the others (cf. work done around
568 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
569 entirely straightforward.
570 c. All of these functions will silently accept a type of the form
572 whether or not the return value is of this type. This is
573 probably permitted by ANSI (see "Exceptional Situations" under
574 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
575 know about this escape clause, so code of the form
576 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
577 can erroneously return T.
579 215: ":TEST-NOT handling by functions"
581 We should verify that our handling of :TEST-NOT and :TEST is consistent
582 for all functions that accept them: that is, signal an error if both
585 Similarly, a compile-time full warning for calls with both would be good.
587 We might also consider a compile-time style warning for :TEST-NOT.
589 216: "debugger confused by frames with invalid number of arguments"
590 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
591 leaves the system confused, enough so that (QUIT) no longer works.
592 It's as though the process of working with the uninitialized slot in
593 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
594 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
596 (Can't reproduce on x86 linux as of 1.0.20.22 - MGL)
598 This is probably the same bug as 162
600 235: "type system and inline expansion"
602 (declaim (ftype (function (cons) number) acc))
603 (declaim (inline acc))
605 (the number (car c)))
608 (values (locally (declare (optimize (safety 0)))
610 (locally (declare (optimize (safety 3)))
613 (foo '(nil) '(t)) => NIL, T.
615 As of 0.9.15.41 this seems to be due to ACC being inlined only once
616 inside FOO, which results in the second call reusing the FUNCTIONAL
617 resulting from the first -- which doesn't check the type.
619 237: "Environment arguments to type functions"
620 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
621 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
622 argument, but they ignore it completely. This is almost
623 certainly not correct.
624 b. Also, the compiler's optimizers for TYPEP have not been informed
625 about the new argument; consequently, they will not transform
626 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
627 just as optimizeable as (TYPEP 1 'INTEGER).
629 242: "WRITE-SEQUENCE suboptimality"
630 (observed from clx performance)
631 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
632 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
633 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
634 rather than writing the sequence in one go, leading to severe
635 performance degradation.
636 As of sbcl-0.9.0.36, this is solved for fd-streams, so is less of a
637 problem in practice. (Fully fixing this would require adding a
638 ansi-stream-n-bout slot and associated methods to write a byte
639 sequence to ansi-stream, similar to the existing ansi-stream-sout
642 243: "STYLE-WARNING overenthusiasm for unused variables"
643 (observed from clx compilation)
644 In sbcl-0.7.14, in the presence of the macros
645 (DEFMACRO FOO (X) `(BAR ,X))
646 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
647 somewhat surprising style warnings are emitted for
648 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
650 ; (LAMBDA (Y) (FOO Y))
652 ; caught STYLE-WARNING:
653 ; The variable Y is defined but never used.
655 245: bugs in disassembler
656 b. On X86 operand size prefix is not recognized.
659 (defun foo (&key (a :x))
663 does not cause a warning. (BTW: old SBCL issued a warning, but for a
664 function, which was never called!)
667 Compiler does not emit warnings for
669 a. (lambda () (svref (make-array 8 :adjustable t) 1))
671 b. (fixed at some point before 1.0.4.10)
674 (declare (optimize (debug 0)))
675 (declare (type vector x))
676 (list (fill-pointer x)
680 Complex array type does not have corresponding type specifier.
682 This is a problem because the compiler emits optimization notes when
683 you use a non-simple array, and without a type specifier for hairy
684 array types, there's no good way to tell it you're doing it
685 intentionally so that it should shut up and just compile the code.
687 Another problem is confusing error message "asserted type ARRAY
688 conflicts with derived type (VALUES SIMPLE-VECTOR &OPTIONAL)" during
689 compiling (LAMBDA (V) (VALUES (SVREF V 0) (VECTOR-POP V))).
691 The last problem is that when type assertions are converted to type
692 checks, types are represented with type specifiers, so we could lose
693 complex attribute. (Now this is probably not important, because
694 currently checks for complex arrays seem to be performed by
698 (compile nil '(lambda () (aref (make-array 0) 0))) compiles without
699 warning. Analogous cases with the index and length being equal and
700 greater than 0 are warned for; the problem here seems to be that the
701 type required for an array reference of this type is (INTEGER 0 (0))
702 which is canonicalized to NIL.
707 (t1 (specifier-type s)))
708 (eval `(defstruct ,s))
709 (type= t1 (specifier-type s)))
714 b. The same for CSUBTYPEP.
716 262: "yet another bug in inline expansion of local functions"
717 During inline expansion of a local function Python can try to
718 reference optimized away objects (functions, variables, CTRANs from
719 tags and blocks), which later may lead to problems. Some of the
720 cases are worked around by forbidding expansion in such cases, but
721 the better way would be to reimplement inline expansion by copying
725 David Lichteblau provided (sbcl-devel 2003-06-01) a patch to fix
726 behaviour of streams with element-type (SIGNED-BYTE 8). The patch
727 looks reasonable, if not obviously correct; however, it caused the
728 PPC/Linux port to segfault during warm-init while loading
729 src/pcl/std-class.fasl. A workaround patch was made, but it would
730 be nice to understand why the first patch caused problems, and to
731 fix the cause if possible.
733 268: "wrong free declaration scope"
734 The following code must signal type error:
736 (locally (declare (optimize (safety 3)))
737 (flet ((foo (x &optional (y (car x)))
738 (declare (optimize (safety 0)))
740 (funcall (eval #'foo) 1)))
743 In the following function constraint propagator optimizes nothing:
746 (declare (integer x))
747 (declare (optimize speed))
755 Compilation of the following two forms causes "X is unbound" error:
757 (symbol-macrolet ((x pi))
758 (macrolet ((foo (y) (+ x y)))
759 (declaim (inline bar))
765 (See (COERCE (CDR X) 'FUNCTION) in IR1-CONVERT-INLINE-LAMBDA.)
768 CLHS says that type declaration of a symbol macro should not affect
769 its expansion, but in SBCL it does. (If you like magic and want to
770 fix it, don't forget to change all uses of MACROEXPAND to
774 The following code (taken from CLOCC) takes a lot of time to compile:
777 (declare (type (integer 0 #.large-constant) n))
780 (fixed in 0.8.2.51, but a test case would be good)
782 279: type propagation error -- correctly inferred type goes astray?
783 In sbcl-0.8.3 and sbcl-0.8.1.47, the warning
784 The binding of ABS-FOO is a (VALUES (INTEGER 0 0)
785 &OPTIONAL), not a (INTEGER 1 536870911)
786 is emitted when compiling this file:
787 (declaim (ftype (function ((integer 0 #.most-positive-fixnum))
788 (integer #.most-negative-fixnum 0))
793 (let* (;; Uncomment this for a type mismatch warning indicating
794 ;; that the type of (FOO X) is correctly understood.
795 #+nil (fs-foo (float-sign (foo x)))
796 ;; Uncomment this for a type mismatch warning
797 ;; indicating that the type of (ABS (FOO X)) is
798 ;; correctly understood.
799 #+nil (fs-abs-foo (float-sign (abs (foo x))))
800 ;; something wrong with this one though
801 (abs-foo (abs (foo x))))
802 (declare (type (integer 1 100) abs-foo))
807 283: Thread safety: libc functions
808 There are places that we call unsafe-for-threading libc functions
809 that we should find alternatives for, or put locks around. Known or
810 strongly suspected problems, as of 1.0.3.13: please update this
811 bug instead of creating new ones
813 284: Thread safety: special variables
814 There are lots of special variables in SBCL, and I feel sure that at
815 least some of them are indicative of potentially thread-unsafe
816 parts of the system. See doc/internals/notes/threading-specials
818 286: "recursive known functions"
819 Self-call recognition conflicts with known function
820 recognition. Currently cross compiler and target COMPILE do not
821 recognize recursion, and in target compiler it can be disabled. We
822 can always disable it for known functions with RECURSIVE attribute,
823 but there remains a possibility of a function with a
824 (tail)-recursive simplification pass and transforms/VOPs for base
827 288: fundamental cross-compilation issues (from old UGLINESS file)
828 Using host floating point numbers to represent target floating point
829 numbers, or host characters to represent target characters, is
830 theoretically shaky. (The characters are OK as long as the characters
831 are in the ANSI-guaranteed character set, though, so they aren't a
832 real problem as long as the sources don't need anything but that;
833 the floats are a real problem.)
835 289: "type checking and source-transforms"
837 (block nil (let () (funcall #'+ (eval 'nil) (eval '1) (return :good))))
840 Our policy is to check argument types at the moment of a call. It
841 disagrees with ANSI, which says that type assertions are put
842 immediately onto argument expressions, but is easier to implement in
843 IR1 and is more compatible to type inference, inline expansion,
844 etc. IR1-transforms automatically keep this policy, but source
845 transforms for associative functions (such as +), being applied
846 during IR1-convertion, do not. It may be tolerable for direct calls
847 (+ x y z), but for (FUNCALL #'+ x y z) it is non-conformant.
849 b. Another aspect of this problem is efficiency. [x y + z +]
850 requires less registers than [x y z + +]. This transformation is
851 currently performed with source transforms, but it would be good to
852 also perform it in IR1 optimization phase.
854 290: Alpha floating point and denormalized traps
855 In SBCL 0.8.3.6x on the alpha, we work around what appears to be a
856 hardware or kernel deficiency: the status of the enable/disable
857 denormalized-float traps bit seems to be ambiguous; by the time we
858 get to os_restore_fp_control after a trap, denormalized traps seem
859 to be enabled. Since we don't want a trap every time someone uses a
860 denormalized float, in general, we mask out that bit when we restore
861 the control word; however, this clobbers any change the user might
865 LOOP with non-constant arithmetic step clauses suffers from overzealous
866 type constraint: code of the form
867 (loop for d of-type double-float from 0d0 to 10d0 by x collect d)
868 compiles to a type restriction on X of (AND DOUBLE-FLOAT (REAL
869 (0))). However, an integral value of X should be legal, because
870 successive adds of integers to double-floats produces double-floats,
871 so none of the type restrictions in the code is violated.
873 300: (reported by Peter Graves) Function PEEK-CHAR checks PEEK-TYPE
874 argument type only after having read a character. This is caused
875 with EXPLICIT-CHECK attribute in DEFKNOWN. The similar problem
876 exists with =, /=, <, >, <=, >=. They were fixed, but it is probably
877 less error prone to have EXPLICIT-CHECK be a local declaration,
878 being put into the definition, instead of an attribute being kept in
879 a separate file; maybe also put it into SB-EXT?
881 301: ARRAY-SIMPLE-=-TYPE-METHOD breaks on corner cases which can arise
882 in NOTE-ASSUMED-TYPES
883 In sbcl-0.8.7.32, compiling the file
885 (declare (type integer x))
886 (declare (type (vector (or hash-table bit)) y))
889 (declare (type integer x))
890 (declare (type (simple-array base (2)) y))
893 failed AVER: "(NOT (AND (NOT EQUALP) CERTAINP))"
895 303: "nonlinear LVARs" (aka MISC.293)
897 (multiple-value-call #'list
899 (multiple-value-prog1
900 (eval '(values :a :b :c))
906 (throw 'bar (values 3 4)))))))))))
908 (BUU 1) returns garbage.
910 The problem is that both EVALs sequentially write to the same LVAR.
912 306: "Imprecise unions of array types"
914 a. fixed in SBCL 0.9.15.48
919 ,@(loop for x across sb-vm:*specialized-array-element-type-properties*
920 collect `(array ,(sb-vm:saetp-specifier x)))))
921 => NIL, T (when it should be T, T)
923 309: "Dubious values for implementation limits"
924 (reported by Bruno Haible sbcl-devel "Incorrect value of
925 multiple-values-limit" 2004-04-19)
926 (values-list (make-list 1000000)), on x86/linux, signals a stack
927 exhaustion condition, despite MULTIPLE-VALUES-LIMIT being
928 significantly larger than 1000000. There are probably similar
929 dubious values for CALL-ARGUMENTS-LIMIT (see cmucl-help/cmucl-imp
930 around the same time regarding a call to LIST on sparc with 1000
931 arguments) and other implementation limit constants.
933 314: "LOOP :INITIALLY clauses and scope of initializers"
934 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
935 test suite, originally by Thomas F. Burdick.
936 ;; <http://www.lisp.org/HyperSpec/Body/sec_6-1-7-2.html>
937 ;; According to the HyperSpec 6.1.2.1.4, in for-as-equals-then, var is
938 ;; initialized to the result of evaluating form1. 6.1.7.2 says that
939 ;; initially clauses are evaluated in the loop prologue, which precedes all
940 ;; loop code except for the initial settings provided by with, for, or as.
941 (loop :for x = 0 :then (1+ x)
942 :for y = (1+ x) :then (ash y 1)
943 :for z :across #(1 3 9 27 81 243)
945 :initially (assert (zerop x)) :initially (assert (= 2 w))
946 :until (>= w 100) :collect w)
947 Expected: (2 6 15 38)
950 318: "stack overflow in compiler warning with redefined class"
951 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
954 (setf (find-class 'foo) nil)
955 (defstruct foo slot-1)
956 This used to give a stack overflow from within the printer, which has
957 been fixed as of 0.8.16.11. Current result:
959 ; can't compile TYPEP of anonymous or undefined class:
960 ; #<SB-KERNEL:STRUCTURE-CLASSOID FOO>
962 debugger invoked on a TYPE-ERROR in thread 19973:
963 The value NIL is not of type FUNCTION.
965 CSR notes: it's not really clear what it should give: is (SETF FIND-CLASS)
966 meant to be enough to delete structure classes from the system?
968 319: "backquote with comma inside array"
969 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
971 (read-from-string "`#1A(1 2 ,(+ 2 2) 4)")
973 #(1 2 ((SB-IMPL::|,|) + 2 2) 4)
974 which probably isn't intentional.
976 324: "STREAMs and :ELEMENT-TYPE with large bytesize"
977 In theory, (open foo :element-type '(unsigned-byte <x>)) should work
978 for all positive integral <x>. At present, it only works for <x> up
979 to about 1024 (and similarly for signed-byte), so
980 (open "/dev/zero" :element-type '(unsigned-byte 1025))
981 gives an error in sbcl-0.8.10.
983 325: "CLOSE :ABORT T on superseding streams"
984 Closing a stream opened with :IF-EXISTS :SUPERSEDE with :ABORT T leaves no
985 file on disk, even if one existed before opening.
987 The illegality of this is not crystal clear, as the ANSI dictionary
988 entry for CLOSE says that when :ABORT is T superseded files are not
989 superseded (ie. the original should be restored), whereas the OPEN
990 entry says about :IF-EXISTS :SUPERSEDE "If possible, the
991 implementation should not destroy the old file until the new stream
992 is closed." -- implying that even though undesirable, early deletion
993 is legal. Restoring the original would none the less be the polite
996 326: "*PRINT-CIRCLE* crosstalk between streams"
997 In sbcl-0.8.10.48 it's possible for *PRINT-CIRCLE* references to be
998 mixed between streams when output operations are intermingled closely
999 enough (as by doing output on S2 from within (PRINT-OBJECT X S1) in the
1000 test case below), so that e.g. the references #2# appears on a stream
1001 with no preceding #2= on that stream to define it (because the #2= was
1002 sent to another stream).
1003 (cl:in-package :cl-user)
1004 (defstruct foo index)
1005 (defparameter *foo* (make-foo :index 4))
1007 (defparameter *bar* (make-bar))
1008 (defparameter *tangle* (list *foo* *bar* *foo*))
1009 (defmethod print-object ((foo foo) stream)
1010 (let ((index (foo-index foo)))
1011 (format *trace-output*
1012 "~&-$- emitting FOO ~D, ambient *BAR*=~S~%"
1014 (format stream "[FOO ~D]" index))
1016 (let ((tsos (make-string-output-stream))
1017 (ssos (make-string-output-stream)))
1018 (let ((*print-circle* t)
1019 (*trace-output* tsos)
1020 (*standard-output* ssos))
1021 (prin1 *tangle* *standard-output*))
1022 (let ((string (get-output-stream-string ssos)))
1023 (unless (string= string "(#1=[FOO 4] #S(BAR) #1#)")
1024 ;; In sbcl-0.8.10.48 STRING was "(#1=[FOO 4] #2# #1#)".:-(
1025 (error "oops: ~S" string)))))
1026 It might be straightforward to fix this by turning the
1027 *CIRCULARITY-HASH-TABLE* and *CIRCULARITY-COUNTER* variables into
1028 per-stream slots, but (1) it would probably be sort of messy faking
1029 up the special variable binding semantics using UNWIND-PROTECT and
1030 (2) it might be sort of a pain to test that no other bugs had been
1033 328: "Profiling generic functions", transplanted from #241
1034 (from tonyms on #lisp IRC 2003-02-25)
1035 In sbcl-0.7.12.55, typing
1036 (defclass foo () ((bar :accessor foo-bar)))
1039 (defclass foo () ((bar :accessor foo-bar)))
1040 gives the error message
1041 "#:FOO-BAR already names an ordinary function or a macro."
1043 Problem: when a generic function is profiled, it appears as an ordinary
1044 function to PCL. (Remembering the uninterned accessor is OK, as the
1045 redefinition must be able to remove old accessors from their generic
1048 329: "Sequential class redefinition"
1049 reported by Bruno Haible:
1050 (defclass reactor () ((max-temp :initform 10000000)))
1051 (defvar *r1* (make-instance 'reactor))
1052 (defvar *r2* (make-instance 'reactor))
1053 (slot-value *r1* 'max-temp)
1054 (slot-value *r2* 'max-temp)
1055 (defclass reactor () ((uptime :initform 0)))
1056 (slot-value *r1* 'uptime)
1057 (defclass reactor () ((uptime :initform 0) (max-temp :initform 10000)))
1058 (slot-value *r1* 'max-temp) ; => 10000
1059 (slot-value *r2* 'max-temp) ; => 10000000 oops...
1062 The method effective when the wrapper is obsoleted can be saved
1063 in the wrapper, and then to update the instance just run through
1064 all the old wrappers in order from oldest to newest.
1066 336: "slot-definitions must retain the generic functions of accessors"
1067 reported by Tony Martinez:
1068 (defclass foo () ((bar :reader foo-bar)))
1069 (defun foo-bar (x) x)
1070 (defclass foo () ((bar :reader get-bar))) ; => error, should work
1072 Note: just punting the accessor removal if the fdefinition
1073 is not a generic function is not enough:
1075 (defclass foo () ((bar :reader foo-bar)))
1076 (defvar *reader* #'foo-bar)
1077 (defun foo-bar (x) x)
1078 (defclass foo () ((bar :initform 'ok :reader get-bar)))
1079 (funcall *reader* (make-instance 'foo)) ; should be an error, since
1080 ; the method must be removed
1081 ; by the class redefinition
1083 Fixing this should also fix a subset of #328 -- update the
1084 description with a new test-case then.
1086 339: "DEFINE-METHOD-COMBINATION bugs"
1087 (reported by Bruno Haible via the clisp test suite)
1089 a. Syntax checking laxity (should produce errors):
1090 i. (define-method-combination foo :documentation :operator)
1091 ii. (define-method-combination foo :documentation nil)
1092 iii. (define-method-combination foo nil)
1093 iv. (define-method-combination foo nil nil
1094 (:arguments order &aux &key))
1095 v. (define-method-combination foo nil nil (:arguments &whole))
1096 vi. (define-method-combination foo nil nil (:generic-function))
1097 vii. (define-method-combination foo nil nil (:generic-function bar baz))
1098 viii. (define-method-combination foo nil nil (:generic-function (bar)))
1099 ix. (define-method-combination foo nil ((3)))
1100 x. (define-method-combination foo nil ((a)))
1102 b. define-method-combination arguments lambda list badness
1103 i. &aux args are currently unsupported;
1104 ii. default values of &optional and &key arguments are ignored;
1105 iii. supplied-p variables for &optional and &key arguments are not
1108 c. (fixed in sbcl-0.9.15.15)
1110 344: more (?) ROOM T problems (possibly part of bug 108)
1111 In sbcl-0.8.12.51, and off and on leading up to it, the
1112 SB!VM:MEMORY-USAGE operations in ROOM T caused
1113 unhandled condition (of type SB-INT:BUG):
1114 failed AVER: "(SAP= CURRENT END)"
1115 Several clever people have taken a shot at this without fixing
1116 it; this time around (before sbcl-0.8.13 release) I (WHN) just
1117 commented out the SB!VM:MEMORY-USAGE calls until someone figures
1118 out how to make them work reliably with the rest of the GC.
1120 (Note: there's at least one dubious thing in room.lisp: see the
1121 comment in VALID-OBJ)
1123 346: alpha backtrace
1124 In sbcl-0.8.13, all backtraces from errors caused by internal errors
1125 on the alpha seem to have a "bogus stack frame".
1127 349: PPRINT-INDENT rounding implementation decisions
1128 At present, pprint-indent (and indeed the whole pretty printer)
1129 more-or-less assumes that it's using a monospace font. That's
1130 probably not too silly an assumption, but one piece of information
1131 the current implementation loses is from requests to indent by a
1132 non-integral amount. As of sbcl-0.8.15.9, the system silently
1133 truncates the indentation to an integer at the point of request, but
1134 maybe the non-integral value should be propagated through the
1135 pprinter and only truncated at output? (So that indenting by 1/2
1136 then 3/2 would indent by two spaces, not one?)
1138 352: forward-referenced-class trouble
1139 reported by Bruno Haible on sbcl-devel
1141 (setf (class-name (find-class 'a)) 'b)
1145 Expected: an instance of c, with a slot named x
1146 Got: debugger invoked on a SIMPLE-ERROR in thread 78906:
1147 While computing the class precedence list of the class named C.
1148 The class named B is a forward referenced class.
1149 The class named B is a direct superclass of the class named C.
1151 [ Is this actually a bug? DEFCLASS only replaces an existing class
1152 when the class name is the proper name of that class, and in the
1153 above code the class found by (FIND-CLASS 'A) does not have a
1154 proper name. CSR, 2006-08-07 ]
1156 353: debugger suboptimalities on x86
1157 On x86 backtraces for undefined functions start with a bogus stack
1158 frame, and backtraces for throws to unknown catch tags with a "no
1159 debug information" frame. These are both due to CODE-COMPONENT-FROM-BITS
1160 (used on non-x86 platforms) being a more complete solution then what
1163 On x86/linux large portions of tests/debug.impure.lisp have been commented
1164 out as failures. The probable culprit for these problems is in x86-call-context
1165 (things work fine on x86/freebsd).
1167 More generally, the debugger internals suffer from excessive x86/non-x86
1168 conditionalization and OAOOMization: refactoring the common parts would
1172 (reported by Bruno Haible)
1173 After the "layout depth conflict" error, the CLOS is left in a state where
1174 it's not possible to define new standard-class subclasses any more.
1176 (defclass prioritized-dispatcher ()
1177 ((dependents :type list :initform nil)))
1178 (defmethod sb-pcl:validate-superclass ((c1 sb-pcl:funcallable-standard-class)
1179 (c2 (eql (find-class 'prioritized-dispatcher))))
1181 (defclass prioritized-generic-function (prioritized-dispatcher standard-generic-function)
1183 (:metaclass sb-pcl:funcallable-standard-class))
1184 ;; ERROR, Quit the debugger with ABORT
1185 (defclass typechecking-reader-class (standard-class)
1187 Expected: #<STANDARD-CLASS TYPECHECKING-READER-CLASS>
1188 Got: ERROR "The assertion SB-PCL::WRAPPERS failed."
1190 [ This test case does not cause the error any more. However,
1191 similar problems can be observed with
1193 (defclass foo (standard-class) ()
1194 (:metaclass sb-mop:funcallable-standard-class))
1195 (sb-mop:finalize-inheritance (find-class 'foo))
1197 (defclass bar (standard-class) ())
1198 (make-instance 'bar)
1201 359: wrong default value for ensure-generic-function's :generic-function-class argument
1202 (reported by Bruno Haible)
1203 ANSI CL is silent on this, but the MOP's specification of ENSURE-GENERIC-FUNCTION says:
1204 "The remaining arguments are the complete set of keyword arguments
1205 received by ENSURE-GENERIC-FUNCTION."
1206 and the spec of ENSURE-GENERIC-FUNCTION-USING-CLASS:
1207 ":GENERIC-FUNCTION-CLASS - a class metaobject or a class name. If it is not
1208 supplied, it defaults to the class named STANDARD-GENERIC-FUNCTION."
1209 This is not the case in SBCL. Test case:
1210 (defclass my-generic-function (standard-generic-function)
1212 (:metaclass sb-pcl:funcallable-standard-class))
1213 (setf (fdefinition 'foo1)
1214 (make-instance 'my-generic-function :name 'foo1))
1215 (ensure-generic-function 'foo1
1216 :generic-function-class (find-class 'standard-generic-function))
1218 ; => #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1219 (setf (fdefinition 'foo2)
1220 (make-instance 'my-generic-function :name 'foo2))
1221 (ensure-generic-function 'foo2)
1223 Expected: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1224 Got: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS MY-GENERIC-FUNCTION>
1226 362: missing error when a slot-definition is created without a name
1227 (reported by Bruno Haible)
1228 The MOP says about slot-definition initialization:
1229 "The :NAME argument is a slot name. An ERROR is SIGNALled if this argument
1230 is not a symbol which can be used as a variable name. An ERROR is SIGNALled
1231 if this argument is not supplied."
1233 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition))
1235 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION NIL>
1237 363: missing error when a slot-definition is created with a wrong documentation object
1238 (reported by Bruno Haible)
1239 The MOP says about slot-definition initialization:
1240 "The :DOCUMENTATION argument is a STRING or NIL. An ERROR is SIGNALled
1241 if it is not. This argument default to NIL during initialization."
1243 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition)
1245 :documentation 'not-a-string)
1247 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION FOO>
1249 370: reader misbehaviour on large-exponent floats
1250 (read-from-string "1.0s1000000000000000000000000000000000000000")
1251 causes the reader to attempt to create a very large bignum (which it
1252 will then attempt to coerce to a rational). While this isn't
1253 completely wrong, it is probably not ideal -- checking the floating
1254 point control word state and then returning the relevant float
1255 (most-positive-short-float or short-float-infinity) or signalling an
1256 error immediately would seem to make more sense.
1258 372: floating-point overflow not signalled on ppc/darwin
1259 The following assertions in float.pure.lisp fail on ppc/darwin
1260 (Mac OS X version 10.3.7):
1261 (assert (raises-error? (scale-float 1.0 most-positive-fixnum)
1262 floating-point-overflow))
1263 (assert (raises-error? (scale-float 1.0d0 (1+ most-positive-fixnum))
1264 floating-point-overflow)))
1265 as the SCALE-FLOAT just returns
1266 #.SB-EXT:SINGLE/DOUBLE-FLOAT-POSITIVE-INFINITY. These tests have been
1267 disabled on Darwin for now.
1269 377: Memory fault error reporting
1270 On those architectures where :C-STACK-IS-CONTROL-STACK is in
1271 *FEATURES*, we handle SIG_MEMORY_FAULT (SEGV or BUS) on an altstack,
1272 so we cannot handle the signal directly (as in interrupt_handle_now())
1273 in the case when the signal comes from some external agent (the user
1274 using kill(1), or a fault in some foreign code, for instance). As
1275 of sbcl-0.8.20.20, this is fixed by calling
1276 arrange_return_to_lisp_function() to a new error-signalling
1277 function, but as a result the error reporting is poor: we cannot
1278 even tell the user at which address the fault occurred. We should
1279 arrange such that arguments can be passed to the function called from
1280 arrange_return_to_lisp_function(), but this looked hard to do in
1281 general without suffering from memory leaks.
1283 379: TRACE :ENCAPSULATE NIL broken on ppc/darwin
1284 See commented-out test-case in debug.impure.lisp.
1286 382: externalization unexpectedly changes array simplicity
1287 COMPILE-FILE and LOAD
1289 (let ((x #.(make-array 4 :fill-pointer 0)))
1290 (values (eval `(typep ',x 'simple-array))
1291 (typep x 'simple-array))))
1292 then (FOO) => T, NIL.
1294 Similar problems exist with SIMPLE-ARRAY-P, ARRAY-HEADER accessors
1295 and all array dimension functions.
1297 383: ASH'ing non-constant zeros
1300 (declare (type (integer -2 14) b))
1301 (declare (ignorable b))
1302 (ash (imagpart b) 57))
1303 on PPC (and other platforms, presumably) gives an error during the
1304 emission of FASH-ASH-LEFT/FIXNUM=>FIXNUM as the assembler attempts to
1305 stuff a too-large constant into the immediate field of a PPC
1306 instruction. Either the VOP should be fixed or the compiler should be
1307 taught how to transform this case away, paying particular attention
1308 to side-effects that might occur in the arguments to ASH.
1310 384: Compiler runaway on very large character types
1312 (compile nil '(lambda (x)
1313 (declare (type (member #\a 1) x))
1314 (the (member 1 nil) x)))
1316 The types apparently normalize into a very large type, and the compiler
1317 gets lost in REMOVE-DUPLICATES. Perhaps the latter should use
1318 a better algorithm (one based on hash tables, say) on very long lists
1319 when :TEST has its default value?
1323 (compile nil '(lambda (x) (the (not (eql #\a)) x)))
1325 (partially fixed in 0.9.3.1, but a better representation for these
1329 (format nil "~4,1F" 0.001) => "0.00" (should be " 0.0");
1330 (format nil "~4,1@F" 0.001) => "+.00" (should be "+0.0").
1331 (format nil "~E" 0.01) => "10.e-3" (should be "1.e-2");
1332 (format nil "~G" 0.01) => "10.e-3" (should be "1.e-2");
1334 386: SunOS/x86 stack exhaustion handling broken
1335 According to <http://alfa.s145.xrea.com/sbcl/solaris-x86.html>, the
1336 stack exhaustion checking (implemented with a write-protected guard
1337 page) does not work on SunOS/x86.
1340 (found by Dmitry Bogomolov)
1342 (defclass foo () ((x :type (unsigned-byte 8))))
1343 (defclass bar () ((x :type symbol)))
1344 (defclass baz (foo bar) ())
1348 SB-PCL::SPECIALIZER-APPLICABLE-USING-TYPE-P cannot handle the second argument
1351 [ Can't trigger this any more, as of 2006-08-07 ]
1354 (reported several times on sbcl-devel, by Rick Taube, Brian Rowe and
1357 ROUND-NUMERIC-BOUND assumes that float types always have a FORMAT
1358 specifying whether they're SINGLE or DOUBLE. This is true for types
1359 computed by the type system itself, but the compiler type derivation
1360 short-circuits this and constructs non-canonical types. A temporary
1361 fix was made to ROUND-NUMERIC-BOUND for the sbcl-0.9.6 release, but
1362 the right fix is to remove the abstraction violation in the
1363 compiler's type deriver.
1365 393: Wrong error from methodless generic function
1366 (DEFGENERIC FOO (X))
1368 gives NO-APPLICABLE-METHOD rather than an argument count error.
1370 396: block-compilation bug
1374 (when (funcall (eval #'(lambda (x) (eql x 2))) y)
1376 (incf x (incf y z))))))
1380 (bar 1) => 11, should be 4.
1383 The more interrupts arrive the less accurate SLEEP's timing gets.
1384 (time (sb-thread:terminate-thread
1385 (prog1 (sb-thread:make-thread (lambda ()
1392 398: GC-unsafe SB-ALIEN string deporting
1393 Translating a Lisp string to an alien string by taking a SAP to it
1394 as done by the :DEPORT-GEN methods for C-STRING and UTF8-STRING
1395 is not safe, since the Lisp string can move. For example the
1396 following code will fail quickly on both cheneygc and pre-0.9.8.19
1399 (setf (bytes-consed-between-gcs) 4096)
1400 (define-alien-routine "strcmp" int (s1 c-string) (s2 c-string))
1403 (let ((string "hello, world"))
1404 (assert (zerop (strcmp string string)))))
1406 (This will appear to work on post-0.9.8.19 GENCGC, since
1407 the GC no longer zeroes memory immediately after releasing
1408 it after a minor GC. Either enabling the READ_PROTECT_FREE_PAGES
1409 #define in gencgc.c or modifying the example so that a major
1410 GC will occasionally be triggered would unmask the bug.)
1412 On cheneygc the only solution would seem to be allocating some alien
1413 memory, copying the data over, and arranging that it's freed once we
1414 return. For GENCGC we could instead try to arrange that the string
1415 from which the SAP is taken is always pinned.
1417 For some more details see comments for (define-alien-type-method
1418 (c-string :deport-gen) ...) in host-c-call.lisp.
1420 403: FORMAT/PPRINT-LOGICAL-BLOCK of CONDITIONs ignoring *PRINT-CIRCLE*
1423 (make-condition 'simple-error
1424 :format-control "ow... ~S"
1425 :format-arguments '(#1=(#1#))))
1426 (setf *print-circle* t *print-level* 4)
1427 (format nil "~@<~A~:@>" *c*)
1430 where I (WHN) believe the correct result is "ow... #1=(#1#)",
1431 like the result from (PRINC-TO-STRING *C*). The question of
1432 what the correct result is is complicated by the hairy text in
1433 the Hyperspec "22.3.5.2 Tilde Less-Than-Sign: Logical Block",
1434 Other than the difference in its argument, ~@<...~:> is
1435 exactly the same as ~<...~:> except that circularity detection
1436 is not applied if ~@<...~:> is encountered at top level in a
1438 But because the odd behavior happens even without the at-sign,
1439 (format nil "~<~A~:@>" (list *c*)) ; => "ow... (((#)))"
1440 and because something seemingly similar can happen even in
1441 PPRINT-LOGICAL-BLOCK invoked directly without FORMAT,
1442 (pprint-logical-block (*standard-output* '(some nonempty list))
1443 (format *standard-output* "~A" '#1=(#1#)))
1444 (which prints "(((#)))" to *STANDARD-OUTPUT*), I don't think
1445 that the 22.3.5.2 trickiness is fundamental to the problem.
1447 My guess is that the problem is related to the logic around the MODE
1448 argument to CHECK-FOR-CIRCULARITY, but I haven't reverse-engineered
1449 enough of the intended meaning of the different MODE values to be
1452 404: nonstandard DWIMness in LOOP with unportably-ordered clauses
1453 In sbcl-0.9.13, the code
1454 (loop with stack = (make-array 2 :fill-pointer 2 :initial-element t)
1455 for length = (length stack)
1456 while (plusp length)
1457 for element = (vector-pop stack)
1459 compiles without error or warning and returns (T T). Unfortunately,
1460 it is inconsistent with the ANSI definition of the LOOP macro,
1461 because it mixes up VARIABLE-CLAUSEs with MAIN-CLAUSEs. Furthermore,
1462 SBCL's interpretation of the intended meaning is only one possible,
1463 unportable interpretation of the noncompliant code; in CLISP 2.33.2,
1464 the code compiles with a warning
1465 LOOP: FOR clauses should occur before the loop's main body
1466 and then fails at runtime with
1467 VECTOR-POP: #() has length zero
1468 perhaps because CLISP has shuffled the clauses into an
1469 ANSI-compliant order before proceeding.
1471 406: functional has external references -- failed aver
1472 Given the following food in a single file
1473 (eval-when (:compile-toplevel :load-toplevel :execute)
1476 (foo #.(make-foo3)))
1477 as of 0.9.18.11 the file compiler breaks on it:
1478 failed AVER: "(NOT (FUNCTIONAL-HAS-EXTERNAL-REFERENCES-P CLAMBDA))"
1479 Defining the missing MAKE-LOAD-FORM method makes the error go away.
1481 407: misoptimization of loop, COERCE 'FLOAT, and HANDLER-CASE for bignums
1482 (reported by Ariel Badichi on sbcl-devel 2007-01-09)
1483 407a: In sbcl-1.0.1 on Linux x86,
1485 (loop for n from (expt 2 1024) do
1487 (coerce n 'single-float)
1488 (simple-type-error ()
1489 (format t "Got here.~%")
1490 (return-from foo)))))
1492 causes an infinite loop, where handling the error would be expected.
1493 407b: In sbcl-1.0.1 on Linux x86,
1495 (loop for n from (expt 2 1024) do
1497 (format t "~E~%" (coerce n 'single-float))
1498 (simple-type-error ()
1499 (format t "Got here.~%")
1500 (return-from bar)))))
1501 fails to compile, with
1502 Too large to be represented as a SINGLE-FLOAT: ...
1504 0: ((LABELS SB-BIGNUM::CHECK-EXPONENT) ...)
1505 1: ((LABELS SB-BIGNUM::FLOAT-FROM-BITS) ...)
1506 2: (SB-KERNEL:%SINGLE-FLOAT ...)
1507 3: (SB-C::BOUND-FUNC ...)
1508 4: (SB-C::%SINGLE-FLOAT-DERIVE-TYPE-AUX ...)
1510 These are now fixed, but (COERCE HUGE 'SINGLE-FLOAT) still signals a
1511 type-error at runtime. The question is, should it instead signal a
1512 floating-point overflow, or return an infinity?
1514 408: SUBTYPEP confusion re. OR of SATISFIES of not-yet-defined predicate
1515 As reported by Levente M\'{e}sz\'{a}ros sbcl-devel 2006-02-20,
1516 (aver (equal (multiple-value-list
1517 (subtypep '(or (satisfies x) string)
1518 '(or (satisfies x) integer)))
1520 fails. Also, beneath that failure lurks another failure,
1521 (aver (equal (multiple-value-list
1523 '(or (satisfies x) integer)))
1525 Having looked at this for an hour or so in sbcl-1.0.2, and
1526 specifically having looked at the output from
1529 (y '(or (satisfies x) integer)))
1530 (trace sb-kernel::union-complex-subtypep-arg2
1531 sb-kernel::invoke-complex-subtypep-arg1-method
1532 sb-kernel::type-union
1533 sb-kernel::type-intersection
1536 my (WHN) impression is that the problem is that the semantics of TYPE=
1537 are wrong for what the UNION-COMPLEX-SUBTYPEP-ARG2 code is trying
1538 to use it for. The comments on the definition of TYPE= probably
1539 date back to CMU CL and seem to define it as a confusing thing:
1540 its primary value is something like "certainly equal," and its
1541 secondary value is something like "certain about that certainty."
1542 I'm left uncertain how to fix UNION-COMPLEX-SUBTYPEP-ARG2 without
1543 reducing its generality by removing the TYPE= cleverness. Possibly
1544 the tempting TYPE/= relative defined next to it might be a
1545 suitable replacement for the purpose. Probably, though, it would
1546 be best to start by reverse engineering exactly what TYPE= and
1547 TYPE/= do, and writing an explanation which is so clear that one
1548 can see immediately what it's supposed to mean in odd cases like
1549 (TYPE= '(SATISFIES X) 'INTEGER) when X isn't defined yet.
1551 409: MORE TYPE SYSTEM PROBLEMS
1552 Found while investigating an optimization failure for extended
1553 sequences. The extended sequence type implementation was altered to
1554 work around the problem, but the fundamental problem remains, to wit:
1555 (sb-kernel:type= (sb-kernel:specifier-type '(or float ratio))
1556 (sb-kernel:specifier-type 'single-float))
1557 returns NIL, NIL on sbcl-1.0.3.
1558 (probably related to bug #408)
1560 410: read circularities and type declarations
1561 Consider the definition
1562 (defstruct foo (a 0 :type (not symbol)))
1564 (setf *print-circle* t) ; just in case
1565 (read-from-string "#1=#s(foo :a #1#)")
1566 This gives a type error (#:G1 is not a (NOT SYMBOL)) because of the
1567 implementation of read circularity, using a symbol as a marker for
1568 the previously-referenced object.
1570 416: backtrace confusion
1581 gives the correct error, but the backtrace shows
1582 1: (SB-KERNEL:FDEFINITION-OBJECT 13 NIL)
1583 as the second frame.
1585 418: SUBSEQ on lists doesn't support bignum indexes
1587 LIST-SUBSEQ* now has all the works necessary to support bignum indexes,
1588 but it needs to be verified that changing the DEFKNOWN doesn't kill
1589 performance elsewhere.
1591 Other generic sequence functions have this problem as well.
1593 419: stack-allocated indirect closure variables are not popped
1596 (multiple-value-call #'list
1597 (eval '(values 1 2 3))
1599 (declare (sb-int:truly-dynamic-extent x))
1604 (declare (dynamic-extent #'mget #'mset))
1605 ((lambda (f g) (eval `(progn ,f ,g (values 4 5 6)))) #'mget #'mset)))))
1607 (ASSERT (EQUAL (BUG419 42) '(1 2 3 4 5 6))) => failure
1609 Note: as of SBCL 1.0.16.29 this bug no longer affects user code, as
1610 SB-INT:TRULY-DYNAMIC-EXTENT needs to be used instead of
1611 DYNAMIC-EXTENT for this to happen. Proper fix for this bug requires
1612 (Nikodemus thinks) storing the relevant LAMBDA-VARs in a
1613 :DYNAMIC-EXTENT cleanup, and teaching stack analysis how to deal
1616 421: READ-CHAR-NO-HANG misbehaviour on Windows Console:
1618 It seems that on Windows READ-CHAR-NO-HANG hangs if the user
1619 has pressed a key, but not yet enter (ie. SYSREAD-MAY-BLOCK-P
1620 seems to lie if the OS is buffering input for us on Console.)
1622 reported by Elliot Slaughter on sbcl-devel 2008/1/10.
1624 422: out-of-extent return not checked in safe code
1626 (declaim (optimize safety))
1627 (funcall (catch 't (block nil (throw 't (lambda () (return))))))
1629 behaves ...erratically. Reported by Kevin Reid on sbcl-devel
1630 2007-07-06. (We don't _have_ to check things like this, but we
1631 generally try to check returns in safe code, so we should here too.)
1633 424: toplevel closures and *CHECK-CONSISTENCY*
1635 The following breaks under COMPILE-FILE if *CHECK-CONSISTENCY* is true.
1637 (let ((exported-symbols-alist
1638 (loop for symbol being the external-symbols of :cl
1639 collect (cons symbol
1640 (concatenate 'string
1642 (string-downcase symbol))))))
1643 (defun hyperdoc-lookup (symbol)
1644 (cdr (assoc symbol exported-symbols-alist))))
1646 (Test-case adapted from CL-PPCRE.)
1648 428: TIMER SCHEDULE-STRESS and PARALLEL-UNSCHEDULE in
1649 timer.impure.lisp fails
1651 Failure modes vary. Core problem seems to be (?) recursive entry to
1656 Compiling a file with this contents makes the compiler loop in
1659 (declaim (inline storage))
1661 (the (simple-array flt (*)) (unknown x)))
1663 (defun test1 (lumps &key cg)
1664 (let ((nodes (map 'list (lambda (lump) (storage lump))
1666 (setf (aref nodes 0) 2)
1667 (assert (every #'~= (apply #'concatenate 'list nodes) '(2 3 6 9)))))
1669 431: alien strucure redefinition doesn't work as expected