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.
80 b: &AUX argument in a boa-constructor without a default value means
81 "do not initilize this slot" and does not cause type error. But
82 an error may be signalled at read time and it would be good if
88 And as long as we're wishing, it would be awfully nice if INSPECT could
89 also report on closures, telling about the values of the bound variables.
91 Currently INSPECT and DESCRIBE do show the values, but showing the
92 names of the bindings would be even nicer.
95 The compiler assumes that any time a function of declared FTYPE
96 doesn't signal an error, its arguments were of the declared type.
97 E.g. compiling and loading
98 (DECLAIM (OPTIMIZE (SAFETY 3)))
99 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
101 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
103 (COND ((> (FACTORIAL X) 1.0E6)
104 (FORMAT T "too big~%"))
106 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
108 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
111 will cause the INTEGERP case to be selected, giving bogus output a la
113 This violates the "declarations are assertions" principle.
114 According to the ANSI spec, in the section "System Class FUNCTION",
115 this is a case of "lying to the compiler", but the lying is done
116 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
117 not by the unexpectedly general definition of FACTORIAL. In any case,
118 "declarations are assertions" means that lying to the compiler should
119 cause an error to be signalled, and should not cause a bogus
120 result to be returned. Thus, the compiler should not assume
121 that arbitrary functions check their argument types. (It might
122 make sense to add another flag (CHECKED?) to DEFKNOWN to
123 identify functions which *do* check their argument types.)
124 (Also, verify that the compiler handles declared function
125 return types as assertions.)
128 The definitions of SIGCONTEXT-FLOAT-REGISTER and
129 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
130 supported on FreeBSD because the floating point state is not saved,
131 but at least as of FreeBSD 4.0, the floating point state *is* saved,
132 so they could be supported after all. Very likely
133 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
136 Compiling and loading
137 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
139 then requesting a BACKTRACE at the debugger prompt gives no information
140 about where in the user program the problem occurred.
142 (this is apparently mostly fixed on the SPARC, PPC, and x86 architectures:
143 while giving the backtrace the non-x86 systems complains about "unknown
144 source location: using block start", but apart from that the
145 backtrace seems reasonable. On x86 this is masked by bug 353. See
146 tests/debug.impure.lisp for a test case)
149 Using the pretty-printer from the command prompt gives funny
150 results, apparently because the pretty-printer doesn't know
151 about user's command input, including the user's carriage return
152 that the user, and therefore the pretty-printer thinks that
153 the new output block should start indented 2 or more characters
154 rightward of the correct location.
157 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
158 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
159 crashes SBCL. In general tracing anything which is used in the
160 implementation of TRACE is likely to have the same problem.
163 ANSI says in one place that type declarations can be abbreviated even
164 when the type name is not a symbol, e.g.
165 (DECLAIM ((VECTOR T) *FOOVECTOR*))
166 SBCL doesn't support this. But ANSI says in another place that this
167 isn't allowed. So it's not clear this is a bug after all. (See the
168 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
171 (Actually this has changed changed since, and types as above are
172 now supported. This may be a bug.)
175 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
176 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
177 the mean of the distribution can be systematically O(0.1%) wrong.
178 Just increasing R-I-E-B is probably not a good solution, since
179 it would decrease efficiency more than is probably necessary. Perhaps
180 using some sort of accept/reject method would be better.
183 Internally the compiler sometimes evaluates
184 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
185 (I stumbled across this when I added an
186 (assert (not (eq type1 *wild-type*)))
187 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
188 in a type context should probably be translated to T, and so it's
189 probably wrong to ask whether it's equal to the T type and then (using
190 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
191 (I haven't tried to investigate this bug enough to guess whether
192 there might be any user-level symptoms.)
194 In fact, the type system is likely to depend on this inequality not
195 holding... * is not equivalent to T in many cases, such as
196 (VECTOR *) /= (VECTOR T).
199 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
200 CL), out-of-line structure slot setters are horribly inefficient
201 whenever the type of the slot is declared, because out-of-line
202 structure slot setters are implemented as closures to save space,
203 so the compiler doesn't compile the type test into code, but
204 instead just saves the type in a lexical closure and interprets it
206 To exercise the problem, compile and load
207 (cl:in-package :cl-user)
209 (bar (error "missing") :type bar))
212 (loop (setf (foo-bar *foo*) x)))
214 (defvar *bar* (make-bar))
215 (defvar *foo* (make-foo :bar *bar*))
216 (defvar *setf-foo-bar* #'(setf foo-bar))
218 (loop (funcall *setf-foo-bar* x *foo*)))
219 then run (WASTREL1 *BAR*) or (WASTREL2 *BAR*), hit Ctrl-C, and
220 use BACKTRACE, to see it's spending all essentially all its time
221 in %TYPEP and VALUES-SPECIFIER-TYPE and so forth.
222 One possible solution would be simply to give up on
223 representing structure slot accessors as functions, and represent
224 them as macroexpansions instead. This can be inconvenient for users,
225 but it's not clear that it's worse than trying to help by expanding
226 into a horribly inefficient implementation.
227 As a workaround for the problem, #'(SETF FOO) expressions
228 can be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
229 (defmacro efficient-setf-function (place-function-name)
230 (or #+sbcl (and (sb-int:info :function :accessor-for place-function-name)
231 ;; a workaround for the problem, encouraging the
232 ;; inline expansion of the structure accessor, so
233 ;; that the compiler can optimize its type test
234 (let ((new-value (gensym "NEW-VALUE-"))
235 (structure-value (gensym "STRUCTURE-VALUE-")))
236 `(lambda (,new-value ,structure-value)
237 (setf (,place-function-name ,structure-value)
239 ;; no problem, can just use the ordinary expansion
240 `(function (setf ,place-function-name))))
243 There's apparently a bug in CEILING optimization which caused
244 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
245 applied the patches to SBCL and they didn't seem to cause problems
246 (as reported sbcl-devel 2001-05-04). However, since the patches
247 modify nontrivial code which was apparently written incorrectly
248 the first time around, until regression tests are written I'm not
249 comfortable merging the patches in the CVS version of SBCL.
254 a) ROOM works by walking over the heap linearly, instead of
255 following the object graph. Hence, it report garbage objects that
256 are unreachable. (Maybe this is a feature and not a bug?)
258 b) ROOM uses MAP-ALLOCATED-OBJECTS to walk the heap, which doesn't
259 check all pointers as well as it should, and can hence become
260 confused, leading to aver failures. As of 1.0.13.21 these (the
261 SAP= aver in particular) should be mostly under control, but push
262 ROOM hard enough and it still might croak.
265 When the compiler inline expands functions, it may be that different
266 kinds of return values are generated from different code branches.
267 E.g. an inline expansion of POSITION generates integer results
268 from one branch, and NIL results from another. When that inline
269 expansion is used in a context where only one of those results
272 (aref *a1* (position x *a2*)))
273 and the compiler can't prove that the unacceptable branch is
274 never taken, then bogus type mismatch warnings can be generated.
275 If you need to suppress the type mismatch warnings, you can
276 suppress the inline expansion,
278 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
279 (aref *a1* (position x *a2*)))
280 or, sometimes, suppress them by declaring the result to be of an
283 (aref *a1* (the integer (position x *a2*))))
285 This is not a new compiler problem in 0.7.0, but the new compiler
286 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
287 more conspicuous. If you don't need performance from these functions,
288 and the bogus warnings are a nuisance for you, you can return to
289 your pre-0.7.0 state of grace with
290 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
295 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
296 the entire lexical environment at the point of MACROLET available
297 in the bodies of the macroexpander functions. In particular, it
298 allows the function bodies (which run at compile time) to try to
299 access lexical variables (which are only defined at runtime).
300 It doesn't even issue a warning, which is bad.
302 The SBCL behavior arguably conforms to the ANSI spec (since the
303 spec says that the behavior is undefined, ergo anything conforms).
304 However, it would be better to issue a compile-time error.
305 Unfortunately I (WHN) don't see any simple way to detect this
306 condition in order to issue such an error, so for the meantime
307 SBCL just does this weird broken "conforming" thing.
309 The ANSI standard says, in the definition of the special operator
311 The macro-expansion functions defined by MACROLET are defined
312 in the lexical environment in which the MACROLET form appears.
313 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
314 the local macro definitions in a MACROLET, but the consequences
315 are undefined if the local macro definitions reference any
316 local variable or function bindings that are visible in that
318 Then it seems to contradict itself by giving the example
320 (macrolet ((fudge (z)
321 ;The parameters x and flag are not accessible
322 ; at this point; a reference to flag would be to
323 ; the global variable of that name.
324 ` (if flag (* ,z ,z) ,z)))
325 ;The parameters x and flag are accessible here.
329 The comment "a reference to flag would be to the global variable
330 of the same name" sounds like good behavior for the system to have.
331 but actual specification quoted above says that the actual behavior
334 (Since 0.7.8.23 macroexpanders are defined in a restricted version
335 of the lexical environment, containing no lexical variables and
336 functions, which seems to conform to ANSI and CLtL2, but signalling
337 a STYLE-WARNING for references to variables similar to locals might
341 Ideally, uninterning a symbol would allow it, and its associated
342 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
343 at least as of sbcl-0.7.0, this isn't the case. Information about
344 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
345 essentially in ordinary (non-weak) hash tables keyed by symbols.
346 Thus, once a system has an entry in this system, it tends to live
347 forever, even when it is uninterned and all other references to it
351 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
352 prominent SourceForge web/db bug tracking system, which is
353 unfortunately not a reliable way to get a timely response from
354 the SBCL maintainers)
355 In the course of trying to build a test case for an
356 application error, I encountered this behavior:
357 If you start up sbcl, and then lay on CTRL-C for a
358 minute or two, the lisp process will eventually say:
359 %PRIMITIVE HALT called; the party is over.
360 and throw you into the monitor. If I start up lisp,
361 attach to the process with strace, and then do the same
362 (abusive) thing, I get instead:
363 access failure in heap page not marked as write-protected
364 and the monitor again. I don't know enough to have the
365 faintest idea of what is going on here.
366 This is with sbcl 6.12, uname -a reports:
367 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
368 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
369 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
370 it took more than two minutes (but less than five) for me.
374 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
375 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
376 COMPLEX implementation didn't deal with this, and hasn't been
377 upgraded to do so. (This doesn't seem to be a high priority
378 conformance problem, since seems hard to construct useful code
381 [ partially fixed by CSR in 0.8.17.17 because of a PFD ansi-tests
382 report that (COMPLEX RATIO) was failing; still failing on types of
383 the form (AND NUMBER (SATISFIES REALP) (SATISFIES ZEROP)). ]
385 b. (fixed in 0.8.3.43)
388 Floating point errors are reported poorly. E.g. on x86 OpenBSD
391 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
392 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
393 No traps are enabled? How can this be?
394 It should be possible to be much more specific (overflow, division
395 by zero, etc.) and of course the "How can this be?" should be fixable.
397 See also bugs #45.c and #183
400 (reported by Robert E. Brown 2002-04-16)
401 When a function is called with too few arguments, causing the
402 debugger to be entered, the uninitialized slots in the bad call frame
403 seem to cause GCish problems, being interpreted as tagged data even
404 though they're not. In particular, executing ROOM in the
405 debugger at that point causes AVER failures:
408 * (lisp-implementation-version)
414 failed AVER: "(SAP= CURRENT END)"
415 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
416 isn't too surprising since there are many differences in stack
417 implementation and GC conservatism between the X86 and other ports.)
419 (Can't reproduce on x86 linux as of 1.0.20.23 - MGL)
421 This is probably the same bug as 216
424 The compiler sometimes tries to constant-fold expressions before
425 it checks to see whether they can be reached. This can lead to
426 bogus warnings about errors in the constant folding, e.g. in code
429 (WRITE-STRING (> X 0) "+" "0"))
430 compiled in a context where the compiler can prove that X is NIL,
431 and the compiler complains that (> X 0) causes a type error because
432 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
433 caused a full WARNING, which made the bug really annoying because then
434 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
435 code. Since then the warning has been downgraded to STYLE-WARNING,
436 so it's still a bug but at least it's a little less annoying.
438 183: "IEEE floating point issues"
439 Even where floating point handling is being dealt with relatively
440 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
441 accrued-exceptions and current-exceptions part of the fp control
442 word don't seem to bear much relation to reality. E.g. on
446 debugger invoked on condition of type DIVISION-BY-ZERO:
447 arithmetic error DIVISION-BY-ZERO signalled
448 0] (sb-vm::get-floating-point-modes)
450 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
451 :ROUNDING-MODE :NEAREST
452 :CURRENT-EXCEPTIONS NIL
453 :ACCRUED-EXCEPTIONS (:INEXACT)
456 * (sb-vm::get-floating-point-modes)
457 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
458 :ROUNDING-MODE :NEAREST
459 :CURRENT-EXCEPTIONS (:INEXACT)
460 :ACCRUED-EXCEPTIONS (:INEXACT)
463 188: "compiler performance fiasco involving type inference and UNION-TYPE"
467 (declare (optimize (safety 3)))
468 (declare (optimize (compilation-speed 2)))
469 (declare (optimize (speed 1) (debug 1) (space 1)))
471 (declare (type (integer 0) start))
472 (print (incf start 22))
473 (print (incf start 26))
474 (print (incf start 28)))
476 (declare (type (integer 0) start))
477 (print (incf start 22))
478 (print (incf start 26)))
480 (declare (type (integer 0) start))
481 (print (incf start 22))
482 (print (incf start 26))))))
484 [ Update: 1.0.14.36 improved this quite a bit (20-25%) by
485 eliminating useless work from PROPAGATE-FROM-SETS -- but as alluded
486 below, maybe we should be smarter about when to decide a derived
487 type is "good enough". ]
489 This example could be solved with clever enough constraint
490 propagation or with SSA, but consider
495 The careful type of X is {2k} :-(. Is it really important to be
496 able to work with unions of many intervals?
498 191: "Miscellaneous PCL deficiencies"
499 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
500 a. DEFCLASS does not inform the compiler about generated
501 functions. Compiling a file with
505 (WITH-SLOTS (A-CLASS-X) A
507 results in a STYLE-WARNING:
509 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
511 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
512 Mai points out that the declamation of functions is in fact
513 incorrect in some cases (most notably for structure
514 classes). This means that at present erroneous attempts to use
515 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
516 won't get the corresponding STYLE-WARNING.
518 [much later, in 2006-08] in fact it's no longer erroneous to use
519 WITH-SLOTS on structure-classes. However, including :METACLASS
520 STRUCTURE-CLASS in the class definition gives a whole bunch of
521 function redefinition warnings, so we're still not good to close
524 c. (fixed in 0.8.4.23)
526 201: "Incautious type inference from compound types"
527 a. (reported by APD sbcl-devel 2002-09-17)
529 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
531 (FORMAT NIL "~S IS ~S, Y = ~S"
538 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
542 (declare (type (array * (4 4)) x))
544 (setq x (make-array '(4 4)))
545 (adjust-array y '(3 5))
546 (= (array-dimension y 0) (eval `(array-dimension ,y 0)))))
548 * (foo (make-array '(4 4) :adjustable t))
551 205: "environment issues in cross compiler"
552 (These bugs have no impact on user code, but should be fixed or
554 a. Macroexpanders introduced with MACROLET are defined in the null
556 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
557 the null lexical environment.
558 c. The cross-compiler cannot inline functions defined in a non-null
561 207: "poorly distributed SXHASH results for compound data"
562 SBCL's SXHASH could probably try a little harder. ANSI: "the
563 intent is that an implementation should make a good-faith
564 effort to produce hash-codes that are well distributed
565 within the range of non-negative fixnums". But
566 (let ((hits (make-hash-table)))
569 (let* ((ij (cons i j))
570 (newlist (push ij (gethash (sxhash ij) hits))))
572 (format t "~&collision: ~S~%" newlist))))))
573 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
574 would be an obvious way of fix. Maybe it would be acceptably efficient
575 to redo MIX using a lookup into a 256-entry s-box containing
576 29-bit pseudorandom numbers?
578 211: "keywords processing"
579 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
580 number of keyword arguments.
582 212: "Sequence functions and circular arguments"
583 COERCE, MERGE and CONCATENATE go into an infinite loop when given
584 circular arguments; it would be good for the user if they could be
585 given an error instead (ANSI 17.1.1 allows this behaviour on the part
586 of the implementation, as conforming code cannot give non-proper
587 sequences to these functions. MAP also has this problem (and
588 solution), though arguably the convenience of being able to do
589 (MAP 'LIST '+ FOO '#1=(1 . #1#))
590 might be classed as more important (though signalling an error when
591 all of the arguments are circular is probably desireable).
593 213: "Sequence functions and type checking"
594 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
595 check that it will return a sequence of the given type. Fixing
596 it along the same lines as the others (cf. work done around
597 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
598 entirely straightforward.
599 c. All of these functions will silently accept a type of the form
601 whether or not the return value is of this type. This is
602 probably permitted by ANSI (see "Exceptional Situations" under
603 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
604 know about this escape clause, so code of the form
605 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
606 can erroneously return T.
608 215: ":TEST-NOT handling by functions"
609 a. FIND and POSITION currently signal errors when given non-NIL for
610 both their :TEST and (deprecated) :TEST-NOT arguments, but by
611 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
612 means that the effect is "unpredictable but harmless". It's not
613 clear what that actually means; it may preclude conforming
614 implementations from signalling errors.
615 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
616 when conflict occurs. As a quality of implementation issue, it
617 might be preferable to treat :TEST and :TEST-NOT as being in some
618 sense the same &KEY, and effectively take the first test function in
620 c. Again, a quality of implementation issue: it would be good to issue a
621 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
622 WARNING for calls with both :TEST and :TEST-NOT; possibly this
623 latter should be WARNed about at execute-time too.
625 216: "debugger confused by frames with invalid number of arguments"
626 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
627 leaves the system confused, enough so that (QUIT) no longer works.
628 It's as though the process of working with the uninitialized slot in
629 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
630 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
632 (Can't reproduce on x86 linux as of 1.0.20.22 - MGL)
634 This is probably the same bug as 162
636 235: "type system and inline expansion"
638 (declaim (ftype (function (cons) number) acc))
639 (declaim (inline acc))
641 (the number (car c)))
644 (values (locally (declare (optimize (safety 0)))
646 (locally (declare (optimize (safety 3)))
649 (foo '(nil) '(t)) => NIL, T.
651 As of 0.9.15.41 this seems to be due to ACC being inlined only once
652 inside FOO, which results in the second call reusing the FUNCTIONAL
653 resulting from the first -- which doesn't check the type.
655 237: "Environment arguments to type functions"
656 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
657 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
658 argument, but they ignore it completely. This is almost
659 certainly not correct.
660 b. Also, the compiler's optimizers for TYPEP have not been informed
661 about the new argument; consequently, they will not transform
662 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
663 just as optimizeable as (TYPEP 1 'INTEGER).
665 238: "REPL compiler overenthusiasm for CLOS code"
667 * (defclass foo () ())
668 * (defmethod bar ((x foo) (foo foo)) (call-next-method))
669 causes approximately 100 lines of code deletion notes. Some
670 discussion on this issue happened under the title 'Three "interesting"
671 bugs in PCL', resulting in a fix for this oververbosity from the
672 compiler proper; however, the problem persists in the interactor
673 because the notion of original source is not preserved: for the
674 compiler, the original source of the above expression is (DEFMETHOD
675 BAR ((X FOO) (FOO FOO)) (CALL-NEXT-METHOD)), while by the time the
676 compiler gets its hands on the code needing compilation from the REPL,
677 it has been macroexpanded several times.
679 A symptom of the same underlying problem, reported by Tony Martinez:
681 (with-input-from-string (*query-io* " no")
683 (simple-type-error () 'error))
685 ; (SB-KERNEL:FLOAT-WAIT)
687 ; note: deleting unreachable code
688 ; compilation unit finished
691 242: "WRITE-SEQUENCE suboptimality"
692 (observed from clx performance)
693 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
694 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
695 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
696 rather than writing the sequence in one go, leading to severe
697 performance degradation.
698 As of sbcl-0.9.0.36, this is solved for fd-streams, so is less of a
699 problem in practice. (Fully fixing this would require adding a
700 ansi-stream-n-bout slot and associated methods to write a byte
701 sequence to ansi-stream, similar to the existing ansi-stream-sout
704 243: "STYLE-WARNING overenthusiasm for unused variables"
705 (observed from clx compilation)
706 In sbcl-0.7.14, in the presence of the macros
707 (DEFMACRO FOO (X) `(BAR ,X))
708 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
709 somewhat surprising style warnings are emitted for
710 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
712 ; (LAMBDA (Y) (FOO Y))
714 ; caught STYLE-WARNING:
715 ; The variable Y is defined but never used.
717 245: bugs in disassembler
718 b. On X86 operand size prefix is not recognized.
721 (defun foo (&key (a :x))
725 does not cause a warning. (BTW: old SBCL issued a warning, but for a
726 function, which was never called!)
729 Compiler does not emit warnings for
731 a. (lambda () (svref (make-array 8 :adjustable t) 1))
733 b. (fixed at some point before 1.0.4.10)
736 (declare (optimize (debug 0)))
737 (declare (type vector x))
738 (list (fill-pointer x)
742 Complex array type does not have corresponding type specifier.
744 This is a problem because the compiler emits optimization notes when
745 you use a non-simple array, and without a type specifier for hairy
746 array types, there's no good way to tell it you're doing it
747 intentionally so that it should shut up and just compile the code.
749 Another problem is confusing error message "asserted type ARRAY
750 conflicts with derived type (VALUES SIMPLE-VECTOR &OPTIONAL)" during
751 compiling (LAMBDA (V) (VALUES (SVREF V 0) (VECTOR-POP V))).
753 The last problem is that when type assertions are converted to type
754 checks, types are represented with type specifiers, so we could lose
755 complex attribute. (Now this is probably not important, because
756 currently checks for complex arrays seem to be performed by
760 (compile nil '(lambda () (aref (make-array 0) 0))) compiles without
761 warning. Analogous cases with the index and length being equal and
762 greater than 0 are warned for; the problem here seems to be that the
763 type required for an array reference of this type is (INTEGER 0 (0))
764 which is canonicalized to NIL.
769 (t1 (specifier-type s)))
770 (eval `(defstruct ,s))
771 (type= t1 (specifier-type s)))
776 b. The same for CSUBTYPEP.
778 262: "yet another bug in inline expansion of local functions"
779 During inline expansion of a local function Python can try to
780 reference optimized away objects (functions, variables, CTRANs from
781 tags and blocks), which later may lead to problems. Some of the
782 cases are worked around by forbidding expansion in such cases, but
783 the better way would be to reimplement inline expansion by copying
787 David Lichteblau provided (sbcl-devel 2003-06-01) a patch to fix
788 behaviour of streams with element-type (SIGNED-BYTE 8). The patch
789 looks reasonable, if not obviously correct; however, it caused the
790 PPC/Linux port to segfault during warm-init while loading
791 src/pcl/std-class.fasl. A workaround patch was made, but it would
792 be nice to understand why the first patch caused problems, and to
793 fix the cause if possible.
795 268: "wrong free declaration scope"
796 The following code must signal type error:
798 (locally (declare (optimize (safety 3)))
799 (flet ((foo (x &optional (y (car x)))
800 (declare (optimize (safety 0)))
802 (funcall (eval #'foo) 1)))
805 In the following function constraint propagator optimizes nothing:
808 (declare (integer x))
809 (declare (optimize speed))
817 Compilation of the following two forms causes "X is unbound" error:
819 (symbol-macrolet ((x pi))
820 (macrolet ((foo (y) (+ x y)))
821 (declaim (inline bar))
827 (See (COERCE (CDR X) 'FUNCTION) in IR1-CONVERT-INLINE-LAMBDA.)
830 CLHS says that type declaration of a symbol macro should not affect
831 its expansion, but in SBCL it does. (If you like magic and want to
832 fix it, don't forget to change all uses of MACROEXPAND to
836 The following code (taken from CLOCC) takes a lot of time to compile:
839 (declare (type (integer 0 #.large-constant) n))
842 (fixed in 0.8.2.51, but a test case would be good)
844 279: type propagation error -- correctly inferred type goes astray?
845 In sbcl-0.8.3 and sbcl-0.8.1.47, the warning
846 The binding of ABS-FOO is a (VALUES (INTEGER 0 0)
847 &OPTIONAL), not a (INTEGER 1 536870911)
848 is emitted when compiling this file:
849 (declaim (ftype (function ((integer 0 #.most-positive-fixnum))
850 (integer #.most-negative-fixnum 0))
855 (let* (;; Uncomment this for a type mismatch warning indicating
856 ;; that the type of (FOO X) is correctly understood.
857 #+nil (fs-foo (float-sign (foo x)))
858 ;; Uncomment this for a type mismatch warning
859 ;; indicating that the type of (ABS (FOO X)) is
860 ;; correctly understood.
861 #+nil (fs-abs-foo (float-sign (abs (foo x))))
862 ;; something wrong with this one though
863 (abs-foo (abs (foo x))))
864 (declare (type (integer 1 100) abs-foo))
869 283: Thread safety: libc functions
870 There are places that we call unsafe-for-threading libc functions
871 that we should find alternatives for, or put locks around. Known or
872 strongly suspected problems, as of 1.0.3.13: please update this
873 bug instead of creating new ones
875 284: Thread safety: special variables
876 There are lots of special variables in SBCL, and I feel sure that at
877 least some of them are indicative of potentially thread-unsafe
878 parts of the system. See doc/internals/notes/threading-specials
880 286: "recursive known functions"
881 Self-call recognition conflicts with known function
882 recognition. Currently cross compiler and target COMPILE do not
883 recognize recursion, and in target compiler it can be disabled. We
884 can always disable it for known functions with RECURSIVE attribute,
885 but there remains a possibility of a function with a
886 (tail)-recursive simplification pass and transforms/VOPs for base
889 288: fundamental cross-compilation issues (from old UGLINESS file)
890 Using host floating point numbers to represent target floating point
891 numbers, or host characters to represent target characters, is
892 theoretically shaky. (The characters are OK as long as the characters
893 are in the ANSI-guaranteed character set, though, so they aren't a
894 real problem as long as the sources don't need anything but that;
895 the floats are a real problem.)
897 289: "type checking and source-transforms"
899 (block nil (let () (funcall #'+ (eval 'nil) (eval '1) (return :good))))
902 Our policy is to check argument types at the moment of a call. It
903 disagrees with ANSI, which says that type assertions are put
904 immediately onto argument expressions, but is easier to implement in
905 IR1 and is more compatible to type inference, inline expansion,
906 etc. IR1-transforms automatically keep this policy, but source
907 transforms for associative functions (such as +), being applied
908 during IR1-convertion, do not. It may be tolerable for direct calls
909 (+ x y z), but for (FUNCALL #'+ x y z) it is non-conformant.
911 b. Another aspect of this problem is efficiency. [x y + z +]
912 requires less registers than [x y z + +]. This transformation is
913 currently performed with source transforms, but it would be good to
914 also perform it in IR1 optimization phase.
916 290: Alpha floating point and denormalized traps
917 In SBCL 0.8.3.6x on the alpha, we work around what appears to be a
918 hardware or kernel deficiency: the status of the enable/disable
919 denormalized-float traps bit seems to be ambiguous; by the time we
920 get to os_restore_fp_control after a trap, denormalized traps seem
921 to be enabled. Since we don't want a trap every time someone uses a
922 denormalized float, in general, we mask out that bit when we restore
923 the control word; however, this clobbers any change the user might
927 LOOP with non-constant arithmetic step clauses suffers from overzealous
928 type constraint: code of the form
929 (loop for d of-type double-float from 0d0 to 10d0 by x collect d)
930 compiles to a type restriction on X of (AND DOUBLE-FLOAT (REAL
931 (0))). However, an integral value of X should be legal, because
932 successive adds of integers to double-floats produces double-floats,
933 so none of the type restrictions in the code is violated.
935 300: (reported by Peter Graves) Function PEEK-CHAR checks PEEK-TYPE
936 argument type only after having read a character. This is caused
937 with EXPLICIT-CHECK attribute in DEFKNOWN. The similar problem
938 exists with =, /=, <, >, <=, >=. They were fixed, but it is probably
939 less error prone to have EXPLICIT-CHECK be a local declaration,
940 being put into the definition, instead of an attribute being kept in
941 a separate file; maybe also put it into SB-EXT?
943 301: ARRAY-SIMPLE-=-TYPE-METHOD breaks on corner cases which can arise
944 in NOTE-ASSUMED-TYPES
945 In sbcl-0.8.7.32, compiling the file
947 (declare (type integer x))
948 (declare (type (vector (or hash-table bit)) y))
951 (declare (type integer x))
952 (declare (type (simple-array base (2)) y))
955 failed AVER: "(NOT (AND (NOT EQUALP) CERTAINP))"
957 303: "nonlinear LVARs" (aka MISC.293)
959 (multiple-value-call #'list
961 (multiple-value-prog1
962 (eval '(values :a :b :c))
968 (throw 'bar (values 3 4)))))))))))
970 (BUU 1) returns garbage.
972 The problem is that both EVALs sequentially write to the same LVAR.
974 306: "Imprecise unions of array types"
976 a. fixed in SBCL 0.9.15.48
981 ,@(loop for x across sb-vm:*specialized-array-element-type-properties*
982 collect `(array ,(sb-vm:saetp-specifier x)))))
983 => NIL, T (when it should be T, T)
985 309: "Dubious values for implementation limits"
986 (reported by Bruno Haible sbcl-devel "Incorrect value of
987 multiple-values-limit" 2004-04-19)
988 (values-list (make-list 1000000)), on x86/linux, signals a stack
989 exhaustion condition, despite MULTIPLE-VALUES-LIMIT being
990 significantly larger than 1000000. There are probably similar
991 dubious values for CALL-ARGUMENTS-LIMIT (see cmucl-help/cmucl-imp
992 around the same time regarding a call to LIST on sparc with 1000
993 arguments) and other implementation limit constants.
995 314: "LOOP :INITIALLY clauses and scope of initializers"
996 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
997 test suite, originally by Thomas F. Burdick.
998 ;; <http://www.lisp.org/HyperSpec/Body/sec_6-1-7-2.html>
999 ;; According to the HyperSpec 6.1.2.1.4, in for-as-equals-then, var is
1000 ;; initialized to the result of evaluating form1. 6.1.7.2 says that
1001 ;; initially clauses are evaluated in the loop prologue, which precedes all
1002 ;; loop code except for the initial settings provided by with, for, or as.
1003 (loop :for x = 0 :then (1+ x)
1004 :for y = (1+ x) :then (ash y 1)
1005 :for z :across #(1 3 9 27 81 243)
1007 :initially (assert (zerop x)) :initially (assert (= 2 w))
1008 :until (>= w 100) :collect w)
1009 Expected: (2 6 15 38)
1012 318: "stack overflow in compiler warning with redefined class"
1013 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
1016 (setf (find-class 'foo) nil)
1017 (defstruct foo slot-1)
1018 This used to give a stack overflow from within the printer, which has
1019 been fixed as of 0.8.16.11. Current result:
1021 ; can't compile TYPEP of anonymous or undefined class:
1022 ; #<SB-KERNEL:STRUCTURE-CLASSOID FOO>
1024 debugger invoked on a TYPE-ERROR in thread 19973:
1025 The value NIL is not of type FUNCTION.
1027 CSR notes: it's not really clear what it should give: is (SETF FIND-CLASS)
1028 meant to be enough to delete structure classes from the system?
1030 319: "backquote with comma inside array"
1031 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
1033 (read-from-string "`#1A(1 2 ,(+ 2 2) 4)")
1035 #(1 2 ((SB-IMPL::|,|) + 2 2) 4)
1036 which probably isn't intentional.
1038 324: "STREAMs and :ELEMENT-TYPE with large bytesize"
1039 In theory, (open foo :element-type '(unsigned-byte <x>)) should work
1040 for all positive integral <x>. At present, it only works for <x> up
1041 to about 1024 (and similarly for signed-byte), so
1042 (open "/dev/zero" :element-type '(unsigned-byte 1025))
1043 gives an error in sbcl-0.8.10.
1045 325: "CLOSE :ABORT T on superseding streams"
1046 Closing a stream opened with :IF-EXISTS :SUPERSEDE with :ABORT T leaves no
1047 file on disk, even if one existed before opening.
1049 The illegality of this is not crystal clear, as the ANSI dictionary
1050 entry for CLOSE says that when :ABORT is T superseded files are not
1051 superseded (ie. the original should be restored), whereas the OPEN
1052 entry says about :IF-EXISTS :SUPERSEDE "If possible, the
1053 implementation should not destroy the old file until the new stream
1054 is closed." -- implying that even though undesirable, early deletion
1055 is legal. Restoring the original would none the less be the polite
1058 326: "*PRINT-CIRCLE* crosstalk between streams"
1059 In sbcl-0.8.10.48 it's possible for *PRINT-CIRCLE* references to be
1060 mixed between streams when output operations are intermingled closely
1061 enough (as by doing output on S2 from within (PRINT-OBJECT X S1) in the
1062 test case below), so that e.g. the references #2# appears on a stream
1063 with no preceding #2= on that stream to define it (because the #2= was
1064 sent to another stream).
1065 (cl:in-package :cl-user)
1066 (defstruct foo index)
1067 (defparameter *foo* (make-foo :index 4))
1069 (defparameter *bar* (make-bar))
1070 (defparameter *tangle* (list *foo* *bar* *foo*))
1071 (defmethod print-object ((foo foo) stream)
1072 (let ((index (foo-index foo)))
1073 (format *trace-output*
1074 "~&-$- emitting FOO ~D, ambient *BAR*=~S~%"
1076 (format stream "[FOO ~D]" index))
1078 (let ((tsos (make-string-output-stream))
1079 (ssos (make-string-output-stream)))
1080 (let ((*print-circle* t)
1081 (*trace-output* tsos)
1082 (*standard-output* ssos))
1083 (prin1 *tangle* *standard-output*))
1084 (let ((string (get-output-stream-string ssos)))
1085 (unless (string= string "(#1=[FOO 4] #S(BAR) #1#)")
1086 ;; In sbcl-0.8.10.48 STRING was "(#1=[FOO 4] #2# #1#)".:-(
1087 (error "oops: ~S" string)))))
1088 It might be straightforward to fix this by turning the
1089 *CIRCULARITY-HASH-TABLE* and *CIRCULARITY-COUNTER* variables into
1090 per-stream slots, but (1) it would probably be sort of messy faking
1091 up the special variable binding semantics using UNWIND-PROTECT and
1092 (2) it might be sort of a pain to test that no other bugs had been
1095 328: "Profiling generic functions", transplanted from #241
1096 (from tonyms on #lisp IRC 2003-02-25)
1097 In sbcl-0.7.12.55, typing
1098 (defclass foo () ((bar :accessor foo-bar)))
1101 (defclass foo () ((bar :accessor foo-bar)))
1102 gives the error message
1103 "#:FOO-BAR already names an ordinary function or a macro."
1105 Problem: when a generic function is profiled, it appears as an ordinary
1106 function to PCL. (Remembering the uninterned accessor is OK, as the
1107 redefinition must be able to remove old accessors from their generic
1110 329: "Sequential class redefinition"
1111 reported by Bruno Haible:
1112 (defclass reactor () ((max-temp :initform 10000000)))
1113 (defvar *r1* (make-instance 'reactor))
1114 (defvar *r2* (make-instance 'reactor))
1115 (slot-value *r1* 'max-temp)
1116 (slot-value *r2* 'max-temp)
1117 (defclass reactor () ((uptime :initform 0)))
1118 (slot-value *r1* 'uptime)
1119 (defclass reactor () ((uptime :initform 0) (max-temp :initform 10000)))
1120 (slot-value *r1* 'max-temp) ; => 10000
1121 (slot-value *r2* 'max-temp) ; => 10000000 oops...
1124 The method effective when the wrapper is obsoleted can be saved
1125 in the wrapper, and then to update the instance just run through
1126 all the old wrappers in order from oldest to newest.
1128 336: "slot-definitions must retain the generic functions of accessors"
1129 reported by Tony Martinez:
1130 (defclass foo () ((bar :reader foo-bar)))
1131 (defun foo-bar (x) x)
1132 (defclass foo () ((bar :reader get-bar))) ; => error, should work
1134 Note: just punting the accessor removal if the fdefinition
1135 is not a generic function is not enough:
1137 (defclass foo () ((bar :reader foo-bar)))
1138 (defvar *reader* #'foo-bar)
1139 (defun foo-bar (x) x)
1140 (defclass foo () ((bar :initform 'ok :reader get-bar)))
1141 (funcall *reader* (make-instance 'foo)) ; should be an error, since
1142 ; the method must be removed
1143 ; by the class redefinition
1145 Fixing this should also fix a subset of #328 -- update the
1146 description with a new test-case then.
1148 339: "DEFINE-METHOD-COMBINATION bugs"
1149 (reported by Bruno Haible via the clisp test suite)
1151 a. Syntax checking laxity (should produce errors):
1152 i. (define-method-combination foo :documentation :operator)
1153 ii. (define-method-combination foo :documentation nil)
1154 iii. (define-method-combination foo nil)
1155 iv. (define-method-combination foo nil nil
1156 (:arguments order &aux &key))
1157 v. (define-method-combination foo nil nil (:arguments &whole))
1158 vi. (define-method-combination foo nil nil (:generic-function))
1159 vii. (define-method-combination foo nil nil (:generic-function bar baz))
1160 viii. (define-method-combination foo nil nil (:generic-function (bar)))
1161 ix. (define-method-combination foo nil ((3)))
1162 x. (define-method-combination foo nil ((a)))
1164 b. define-method-combination arguments lambda list badness
1165 i. &aux args are currently unsupported;
1166 ii. default values of &optional and &key arguments are ignored;
1167 iii. supplied-p variables for &optional and &key arguments are not
1170 c. (fixed in sbcl-0.9.15.15)
1172 344: more (?) ROOM T problems (possibly part of bug 108)
1173 In sbcl-0.8.12.51, and off and on leading up to it, the
1174 SB!VM:MEMORY-USAGE operations in ROOM T caused
1175 unhandled condition (of type SB-INT:BUG):
1176 failed AVER: "(SAP= CURRENT END)"
1177 Several clever people have taken a shot at this without fixing
1178 it; this time around (before sbcl-0.8.13 release) I (WHN) just
1179 commented out the SB!VM:MEMORY-USAGE calls until someone figures
1180 out how to make them work reliably with the rest of the GC.
1182 (Note: there's at least one dubious thing in room.lisp: see the
1183 comment in VALID-OBJ)
1185 346: alpha backtrace
1186 In sbcl-0.8.13, all backtraces from errors caused by internal errors
1187 on the alpha seem to have a "bogus stack frame".
1189 349: PPRINT-INDENT rounding implementation decisions
1190 At present, pprint-indent (and indeed the whole pretty printer)
1191 more-or-less assumes that it's using a monospace font. That's
1192 probably not too silly an assumption, but one piece of information
1193 the current implementation loses is from requests to indent by a
1194 non-integral amount. As of sbcl-0.8.15.9, the system silently
1195 truncates the indentation to an integer at the point of request, but
1196 maybe the non-integral value should be propagated through the
1197 pprinter and only truncated at output? (So that indenting by 1/2
1198 then 3/2 would indent by two spaces, not one?)
1200 352: forward-referenced-class trouble
1201 reported by Bruno Haible on sbcl-devel
1203 (setf (class-name (find-class 'a)) 'b)
1207 Expected: an instance of c, with a slot named x
1208 Got: debugger invoked on a SIMPLE-ERROR in thread 78906:
1209 While computing the class precedence list of the class named C.
1210 The class named B is a forward referenced class.
1211 The class named B is a direct superclass of the class named C.
1213 [ Is this actually a bug? DEFCLASS only replaces an existing class
1214 when the class name is the proper name of that class, and in the
1215 above code the class found by (FIND-CLASS 'A) does not have a
1216 proper name. CSR, 2006-08-07 ]
1218 353: debugger suboptimalities on x86
1219 On x86 backtraces for undefined functions start with a bogus stack
1220 frame, and backtraces for throws to unknown catch tags with a "no
1221 debug information" frame. These are both due to CODE-COMPONENT-FROM-BITS
1222 (used on non-x86 platforms) being a more complete solution then what
1225 On x86/linux large portions of tests/debug.impure.lisp have been commented
1226 out as failures. The probable culprit for these problems is in x86-call-context
1227 (things work fine on x86/freebsd).
1229 More generally, the debugger internals suffer from excessive x86/non-x86
1230 conditionalization and OAOOMization: refactoring the common parts would
1233 354: XEPs in backtraces
1234 Under default compilation policy
1238 Has the XEP for TEST in the backtrace, not the TEST frame itself.
1239 (sparc and x86 at least)
1241 Since SBCL 0.8.20.1 this is hidden unless *SHOW-ENTRY-POINT-DETAILS*
1242 is true (instead there appear two TEST frames at least on ppc). The
1243 underlying cause seems to be that SB-C::TAIL-ANNOTATE will not merge
1244 the tail-call for the XEP, since Python has by that time proved that
1245 the function can never return; same happens if the function holds an
1246 unconditional call to ERROR.
1249 (reported by Bruno Haible)
1250 After the "layout depth conflict" error, the CLOS is left in a state where
1251 it's not possible to define new standard-class subclasses any more.
1253 (defclass prioritized-dispatcher ()
1254 ((dependents :type list :initform nil)))
1255 (defmethod sb-pcl:validate-superclass ((c1 sb-pcl:funcallable-standard-class)
1256 (c2 (eql (find-class 'prioritized-dispatcher))))
1258 (defclass prioritized-generic-function (prioritized-dispatcher standard-generic-function)
1260 (:metaclass sb-pcl:funcallable-standard-class))
1261 ;; ERROR, Quit the debugger with ABORT
1262 (defclass typechecking-reader-class (standard-class)
1264 Expected: #<STANDARD-CLASS TYPECHECKING-READER-CLASS>
1265 Got: ERROR "The assertion SB-PCL::WRAPPERS failed."
1267 [ This test case does not cause the error any more. However,
1268 similar problems can be observed with
1270 (defclass foo (standard-class) ()
1271 (:metaclass sb-mop:funcallable-standard-class))
1272 (sb-mop:finalize-inheritance (find-class 'foo))
1274 (defclass bar (standard-class) ())
1275 (make-instance 'bar)
1278 357: defstruct inheritance of initforms
1279 (reported by Bruno Haible)
1280 When defstruct and defclass (with :metaclass structure-class) are mixed,
1281 1. some slot initforms are ignored by the DEFSTRUCT generated constructor
1283 2. all slot initforms are ignored by MAKE-INSTANCE. (This can be arguably
1284 OK for initforms that were given in a DEFSTRUCT form, but for those
1285 given in a DEFCLASS form, I think it qualifies as a bug.)
1287 (defstruct structure02a
1291 (defclass structure02b (structure02a)
1292 ((slot4 :initform -44)
1295 (slot7 :initform (floor (* pi pi)))
1296 (slot8 :initform 88))
1297 (:metaclass structure-class))
1298 (defstruct (structure02c (:include structure02b (slot8 -88)))
1301 (slot11 (floor (exp 3))))
1303 (let ((a (make-structure02c)))
1304 (list (structure02c-slot4 a)
1305 (structure02c-slot5 a)
1306 (structure02c-slot6 a)
1307 (structure02c-slot7 a)))
1308 Expected: (-44 nil t 9)
1309 Got: (SB-PCL::..SLOT-UNBOUND.. SB-PCL::..SLOT-UNBOUND..
1310 SB-PCL::..SLOT-UNBOUND.. SB-PCL::..SLOT-UNBOUND..)
1312 (let ((b (make-instance 'structure02c)))
1313 (list (structure02c-slot2 b)
1314 (structure02c-slot3 b)
1315 (structure02c-slot4 b)
1316 (structure02c-slot6 b)
1317 (structure02c-slot7 b)
1318 (structure02c-slot8 b)
1319 (structure02c-slot10 b)
1320 (structure02c-slot11 b)))
1321 Expected: (t 3 -44 t 9 -88 t 20)
1322 Got: (0 0 0 0 0 0 0 0)
1324 359: wrong default value for ensure-generic-function's :generic-function-class argument
1325 (reported by Bruno Haible)
1326 ANSI CL is silent on this, but the MOP's specification of ENSURE-GENERIC-FUNCTION says:
1327 "The remaining arguments are the complete set of keyword arguments
1328 received by ENSURE-GENERIC-FUNCTION."
1329 and the spec of ENSURE-GENERIC-FUNCTION-USING-CLASS:
1330 ":GENERIC-FUNCTION-CLASS - a class metaobject or a class name. If it is not
1331 supplied, it defaults to the class named STANDARD-GENERIC-FUNCTION."
1332 This is not the case in SBCL. Test case:
1333 (defclass my-generic-function (standard-generic-function)
1335 (:metaclass sb-pcl:funcallable-standard-class))
1336 (setf (fdefinition 'foo1)
1337 (make-instance 'my-generic-function :name 'foo1))
1338 (ensure-generic-function 'foo1
1339 :generic-function-class (find-class 'standard-generic-function))
1341 ; => #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1342 (setf (fdefinition 'foo2)
1343 (make-instance 'my-generic-function :name 'foo2))
1344 (ensure-generic-function 'foo2)
1346 Expected: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1347 Got: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS MY-GENERIC-FUNCTION>
1349 362: missing error when a slot-definition is created without a name
1350 (reported by Bruno Haible)
1351 The MOP says about slot-definition initialization:
1352 "The :NAME argument is a slot name. An ERROR is SIGNALled if this argument
1353 is not a symbol which can be used as a variable name. An ERROR is SIGNALled
1354 if this argument is not supplied."
1356 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition))
1358 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION NIL>
1360 363: missing error when a slot-definition is created with a wrong documentation object
1361 (reported by Bruno Haible)
1362 The MOP says about slot-definition initialization:
1363 "The :DOCUMENTATION argument is a STRING or NIL. An ERROR is SIGNALled
1364 if it is not. This argument default to NIL during initialization."
1366 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition)
1368 :documentation 'not-a-string)
1370 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION FOO>
1372 370: reader misbehaviour on large-exponent floats
1373 (read-from-string "1.0s1000000000000000000000000000000000000000")
1374 causes the reader to attempt to create a very large bignum (which it
1375 will then attempt to coerce to a rational). While this isn't
1376 completely wrong, it is probably not ideal -- checking the floating
1377 point control word state and then returning the relevant float
1378 (most-positive-short-float or short-float-infinity) or signalling an
1379 error immediately would seem to make more sense.
1381 372: floating-point overflow not signalled on ppc/darwin
1382 The following assertions in float.pure.lisp fail on ppc/darwin
1383 (Mac OS X version 10.3.7):
1384 (assert (raises-error? (scale-float 1.0 most-positive-fixnum)
1385 floating-point-overflow))
1386 (assert (raises-error? (scale-float 1.0d0 (1+ most-positive-fixnum))
1387 floating-point-overflow)))
1388 as the SCALE-FLOAT just returns
1389 #.SB-EXT:SINGLE/DOUBLE-FLOAT-POSITIVE-INFINITY. These tests have been
1390 disabled on Darwin for now.
1392 377: Memory fault error reporting
1393 On those architectures where :C-STACK-IS-CONTROL-STACK is in
1394 *FEATURES*, we handle SIG_MEMORY_FAULT (SEGV or BUS) on an altstack,
1395 so we cannot handle the signal directly (as in interrupt_handle_now())
1396 in the case when the signal comes from some external agent (the user
1397 using kill(1), or a fault in some foreign code, for instance). As
1398 of sbcl-0.8.20.20, this is fixed by calling
1399 arrange_return_to_lisp_function() to a new error-signalling
1400 function, but as a result the error reporting is poor: we cannot
1401 even tell the user at which address the fault occurred. We should
1402 arrange such that arguments can be passed to the function called from
1403 arrange_return_to_lisp_function(), but this looked hard to do in
1404 general without suffering from memory leaks.
1406 379: TRACE :ENCAPSULATE NIL broken on ppc/darwin
1407 See commented-out test-case in debug.impure.lisp.
1409 382: externalization unexpectedly changes array simplicity
1410 COMPILE-FILE and LOAD
1412 (let ((x #.(make-array 4 :fill-pointer 0)))
1413 (values (eval `(typep ',x 'simple-array))
1414 (typep x 'simple-array))))
1415 then (FOO) => T, NIL.
1417 Similar problems exist with SIMPLE-ARRAY-P, ARRAY-HEADER accessors
1418 and all array dimension functions.
1420 383: ASH'ing non-constant zeros
1423 (declare (type (integer -2 14) b))
1424 (declare (ignorable b))
1425 (ash (imagpart b) 57))
1426 on PPC (and other platforms, presumably) gives an error during the
1427 emission of FASH-ASH-LEFT/FIXNUM=>FIXNUM as the assembler attempts to
1428 stuff a too-large constant into the immediate field of a PPC
1429 instruction. Either the VOP should be fixed or the compiler should be
1430 taught how to transform this case away, paying particular attention
1431 to side-effects that might occur in the arguments to ASH.
1433 384: Compiler runaway on very large character types
1435 (compile nil '(lambda (x)
1436 (declare (type (member #\a 1) x))
1437 (the (member 1 nil) x)))
1439 The types apparently normalize into a very large type, and the compiler
1440 gets lost in REMOVE-DUPLICATES. Perhaps the latter should use
1441 a better algorithm (one based on hash tables, say) on very long lists
1442 when :TEST has its default value?
1446 (compile nil '(lambda (x) (the (not (eql #\a)) x)))
1448 (partially fixed in 0.9.3.1, but a better representation for these
1452 (format nil "~4,1F" 0.001) => "0.00" (should be " 0.0");
1453 (format nil "~4,1@F" 0.001) => "+.00" (should be "+0.0").
1454 (format nil "~E" 0.01) => "10.e-3" (should be "1.e-2");
1455 (format nil "~G" 0.01) => "10.e-3" (should be "1.e-2");
1457 386: SunOS/x86 stack exhaustion handling broken
1458 According to <http://alfa.s145.xrea.com/sbcl/solaris-x86.html>, the
1459 stack exhaustion checking (implemented with a write-protected guard
1460 page) does not work on SunOS/x86.
1463 (found by Dmitry Bogomolov)
1465 (defclass foo () ((x :type (unsigned-byte 8))))
1466 (defclass bar () ((x :type symbol)))
1467 (defclass baz (foo bar) ())
1471 SB-PCL::SPECIALIZER-APPLICABLE-USING-TYPE-P cannot handle the second argument
1474 [ Can't trigger this any more, as of 2006-08-07 ]
1477 (reported several times on sbcl-devel, by Rick Taube, Brian Rowe and
1480 ROUND-NUMERIC-BOUND assumes that float types always have a FORMAT
1481 specifying whether they're SINGLE or DOUBLE. This is true for types
1482 computed by the type system itself, but the compiler type derivation
1483 short-circuits this and constructs non-canonical types. A temporary
1484 fix was made to ROUND-NUMERIC-BOUND for the sbcl-0.9.6 release, but
1485 the right fix is to remove the abstraction violation in the
1486 compiler's type deriver.
1488 393: Wrong error from methodless generic function
1489 (DEFGENERIC FOO (X))
1491 gives NO-APPLICABLE-METHOD rather than an argument count error.
1493 395: Unicode and streams
1494 One of the remaining problems in SBCL's Unicode support is the lack
1495 of generality in certain streams.
1496 a. FILL-POINTER-STREAMs: SBCL refuses to write (e.g. using FORMAT)
1497 to streams made from strings that aren't character strings with
1499 (let ((v (make-array 5 :fill-pointer 0 :element-type 'standard-char)))
1502 should return a non-simple base string containing "foo" but
1505 (reported on sbcl-help by "tichy")
1507 396: block-compilation bug
1511 (when (funcall (eval #'(lambda (x) (eql x 2))) y)
1513 (incf x (incf y z))))))
1517 (bar 1) => 11, should be 4.
1520 The more interrupts arrive the less accurate SLEEP's timing gets.
1521 (time (sb-thread:terminate-thread
1522 (prog1 (sb-thread:make-thread (lambda ()
1529 398: GC-unsafe SB-ALIEN string deporting
1530 Translating a Lisp string to an alien string by taking a SAP to it
1531 as done by the :DEPORT-GEN methods for C-STRING and UTF8-STRING
1532 is not safe, since the Lisp string can move. For example the
1533 following code will fail quickly on both cheneygc and pre-0.9.8.19
1536 (setf (bytes-consed-between-gcs) 4096)
1537 (define-alien-routine "strcmp" int (s1 c-string) (s2 c-string))
1540 (let ((string "hello, world"))
1541 (assert (zerop (strcmp string string)))))
1543 (This will appear to work on post-0.9.8.19 GENCGC, since
1544 the GC no longer zeroes memory immediately after releasing
1545 it after a minor GC. Either enabling the READ_PROTECT_FREE_PAGES
1546 #define in gencgc.c or modifying the example so that a major
1547 GC will occasionally be triggered would unmask the bug.)
1549 On cheneygc the only solution would seem to be allocating some alien
1550 memory, copying the data over, and arranging that it's freed once we
1551 return. For GENCGC we could instead try to arrange that the string
1552 from which the SAP is taken is always pinned.
1554 For some more details see comments for (define-alien-type-method
1555 (c-string :deport-gen) ...) in host-c-call.lisp.
1557 403: FORMAT/PPRINT-LOGICAL-BLOCK of CONDITIONs ignoring *PRINT-CIRCLE*
1560 (make-condition 'simple-error
1561 :format-control "ow... ~S"
1562 :format-arguments '(#1=(#1#))))
1563 (setf *print-circle* t *print-level* 4)
1564 (format nil "~@<~A~:@>" *c*)
1567 where I (WHN) believe the correct result is "ow... #1=(#1#)",
1568 like the result from (PRINC-TO-STRING *C*). The question of
1569 what the correct result is is complicated by the hairy text in
1570 the Hyperspec "22.3.5.2 Tilde Less-Than-Sign: Logical Block",
1571 Other than the difference in its argument, ~@<...~:> is
1572 exactly the same as ~<...~:> except that circularity detection
1573 is not applied if ~@<...~:> is encountered at top level in a
1575 But because the odd behavior happens even without the at-sign,
1576 (format nil "~<~A~:@>" (list *c*)) ; => "ow... (((#)))"
1577 and because something seemingly similar can happen even in
1578 PPRINT-LOGICAL-BLOCK invoked directly without FORMAT,
1579 (pprint-logical-block (*standard-output* '(some nonempty list))
1580 (format *standard-output* "~A" '#1=(#1#)))
1581 (which prints "(((#)))" to *STANDARD-OUTPUT*), I don't think
1582 that the 22.3.5.2 trickiness is fundamental to the problem.
1584 My guess is that the problem is related to the logic around the MODE
1585 argument to CHECK-FOR-CIRCULARITY, but I haven't reverse-engineered
1586 enough of the intended meaning of the different MODE values to be
1589 404: nonstandard DWIMness in LOOP with unportably-ordered clauses
1590 In sbcl-0.9.13, the code
1591 (loop with stack = (make-array 2 :fill-pointer 2 :initial-element t)
1592 for length = (length stack)
1593 while (plusp length)
1594 for element = (vector-pop stack)
1596 compiles without error or warning and returns (T T). Unfortunately,
1597 it is inconsistent with the ANSI definition of the LOOP macro,
1598 because it mixes up VARIABLE-CLAUSEs with MAIN-CLAUSEs. Furthermore,
1599 SBCL's interpretation of the intended meaning is only one possible,
1600 unportable interpretation of the noncompliant code; in CLISP 2.33.2,
1601 the code compiles with a warning
1602 LOOP: FOR clauses should occur before the loop's main body
1603 and then fails at runtime with
1604 VECTOR-POP: #() has length zero
1605 perhaps because CLISP has shuffled the clauses into an
1606 ANSI-compliant order before proceeding.
1608 405: a TYPE-ERROR in MERGE-LETS exercised at DEBUG 3
1609 In sbcl-0.9.16.21 on linux/86, compiling
1610 (declaim (optimize (debug 3)))
1613 (flet ((i (x) (frob x (foo-bar foo))))
1616 The value NIL is not of type SB-C::PHYSENV.
1619 406: functional has external references -- failed aver
1620 Given the following food in a single file
1621 (eval-when (:compile-toplevel :load-toplevel :execute)
1624 (foo #.(make-foo3)))
1625 as of 0.9.18.11 the file compiler breaks on it:
1626 failed AVER: "(NOT (FUNCTIONAL-HAS-EXTERNAL-REFERENCES-P CLAMBDA))"
1627 Defining the missing MAKE-LOAD-FORM method makes the error go away.
1629 407: misoptimization of loop, COERCE 'FLOAT, and HANDLER-CASE for bignums
1630 (reported by Ariel Badichi on sbcl-devel 2007-01-09)
1631 407a: In sbcl-1.0.1 on Linux x86,
1633 (loop for n from (expt 2 1024) do
1635 (coerce n 'single-float)
1636 (simple-type-error ()
1637 (format t "Got here.~%")
1638 (return-from foo)))))
1640 causes an infinite loop, where handling the error would be expected.
1641 407b: In sbcl-1.0.1 on Linux x86,
1643 (loop for n from (expt 2 1024) do
1645 (format t "~E~%" (coerce n 'single-float))
1646 (simple-type-error ()
1647 (format t "Got here.~%")
1648 (return-from bar)))))
1649 fails to compile, with
1650 Too large to be represented as a SINGLE-FLOAT: ...
1652 0: ((LABELS SB-BIGNUM::CHECK-EXPONENT) ...)
1653 1: ((LABELS SB-BIGNUM::FLOAT-FROM-BITS) ...)
1654 2: (SB-KERNEL:%SINGLE-FLOAT ...)
1655 3: (SB-C::BOUND-FUNC ...)
1656 4: (SB-C::%SINGLE-FLOAT-DERIVE-TYPE-AUX ...)
1658 These are now fixed, but (COERCE HUGE 'SINGLE-FLOAT) still signals a
1659 type-error at runtime. The question is, should it instead signal a
1660 floating-point overflow, or return an infinity?
1662 408: SUBTYPEP confusion re. OR of SATISFIES of not-yet-defined predicate
1663 As reported by Levente M\'{e}sz\'{a}ros sbcl-devel 2006-02-20,
1664 (aver (equal (multiple-value-list
1665 (subtypep '(or (satisfies x) string)
1666 '(or (satisfies x) integer)))
1668 fails. Also, beneath that failure lurks another failure,
1669 (aver (equal (multiple-value-list
1671 '(or (satisfies x) integer)))
1673 Having looked at this for an hour or so in sbcl-1.0.2, and
1674 specifically having looked at the output from
1677 (y '(or (satisfies x) integer)))
1678 (trace sb-kernel::union-complex-subtypep-arg2
1679 sb-kernel::invoke-complex-subtypep-arg1-method
1680 sb-kernel::type-union
1681 sb-kernel::type-intersection
1684 my (WHN) impression is that the problem is that the semantics of TYPE=
1685 are wrong for what the UNION-COMPLEX-SUBTYPEP-ARG2 code is trying
1686 to use it for. The comments on the definition of TYPE= probably
1687 date back to CMU CL and seem to define it as a confusing thing:
1688 its primary value is something like "certainly equal," and its
1689 secondary value is something like "certain about that certainty."
1690 I'm left uncertain how to fix UNION-COMPLEX-SUBTYPEP-ARG2 without
1691 reducing its generality by removing the TYPE= cleverness. Possibly
1692 the tempting TYPE/= relative defined next to it might be a
1693 suitable replacement for the purpose. Probably, though, it would
1694 be best to start by reverse engineering exactly what TYPE= and
1695 TYPE/= do, and writing an explanation which is so clear that one
1696 can see immediately what it's supposed to mean in odd cases like
1697 (TYPE= '(SATISFIES X) 'INTEGER) when X isn't defined yet.
1699 409: MORE TYPE SYSTEM PROBLEMS
1700 Found while investigating an optimization failure for extended
1701 sequences. The extended sequence type implementation was altered to
1702 work around the problem, but the fundamental problem remains, to wit:
1703 (sb-kernel:type= (sb-kernel:specifier-type '(or float ratio))
1704 (sb-kernel:specifier-type 'single-float))
1705 returns NIL, NIL on sbcl-1.0.3.
1706 (probably related to bug #408)
1708 410: read circularities and type declarations
1709 Consider the definition
1710 (defstruct foo (a 0 :type (not symbol)))
1712 (setf *print-circle* t) ; just in case
1713 (read-from-string "#1=#s(foo :a #1#)")
1714 This gives a type error (#:G1 is not a (NOT SYMBOL)) because of the
1715 implementation of read circularity, using a symbol as a marker for
1716 the previously-referenced object.
1718 416: backtrace confusion
1729 gives the correct error, but the backtrace shows
1730 1: (SB-KERNEL:FDEFINITION-OBJECT 13 NIL)
1731 as the second frame.
1733 418: SUBSEQ on lists doesn't support bignum indexes
1735 LIST-SUBSEQ* now has all the works necessary to support bignum indexes,
1736 but it needs to be verified that changing the DEFKNOWN doesn't kill
1737 performance elsewhere.
1739 Other generic sequence functions have this problem as well.
1741 419: stack-allocated indirect closure variables are not popped
1744 (multiple-value-call #'list
1745 (eval '(values 1 2 3))
1747 (declare (sb-int:truly-dynamic-extent x))
1752 (declare (dynamic-extent #'mget #'mset))
1753 ((lambda (f g) (eval `(progn ,f ,g (values 4 5 6)))) #'mget #'mset)))))
1755 (ASSERT (EQUAL (BUG419 42) '(1 2 3 4 5 6))) => failure
1757 Note: as of SBCL 1.0.16.29 this bug no longer affects user code, as
1758 SB-INT:TRULY-DYNAMIC-EXTENT needs to be used instead of
1759 DYNAMIC-EXTENT for this to happen. Proper fix for this bug requires
1760 (Nikodemus thinks) storing the relevant LAMBDA-VARs in a
1761 :DYNAMIC-EXTENT cleanup, and teaching stack analysis how to deal
1764 421: READ-CHAR-NO-HANG misbehaviour on Windows Console:
1766 It seems that on Windows READ-CHAR-NO-HANG hangs if the user
1767 has pressed a key, but not yet enter (ie. SYSREAD-MAY-BLOCK-P
1768 seems to lie if the OS is buffering input for us on Console.)
1770 reported by Elliot Slaughter on sbcl-devel 2008/1/10.
1772 422: out-of-extent return not checked in safe code
1774 (declaim (optimize safety))
1775 (funcall (catch 't (block nil (throw 't (lambda () (return))))))
1777 behaves ...erratically. Reported by Kevin Reid on sbcl-devel
1778 2007-07-06. (We don't _have_ to check things like this, but we
1779 generally try to check returns in safe code, so we should here too.)
1781 424: toplevel closures and *CHECK-CONSISTENCY*
1783 The following breaks under COMPILE-FILE if *CHECK-CONSISTENCY* is true.
1785 (let ((exported-symbols-alist
1786 (loop for symbol being the external-symbols of :cl
1787 collect (cons symbol
1788 (concatenate 'string
1790 (string-downcase symbol))))))
1791 (defun hyperdoc-lookup (symbol)
1792 (cdr (assoc symbol exported-symbols-alist))))
1794 (Test-case adapted from CL-PPCRE.)
1796 428: TIMER SCHEDULE-STRESS and PARALLEL-UNSCHEDULE in
1797 timer.impure.lisp fails
1799 Failure modes vary. Core problem seems to be (?) recursive entry to
1804 Compiling a file with this contents makes the compiler loop in
1807 (declaim (inline storage))
1809 (the (simple-array flt (*)) (unknown x)))
1811 (defun test1 (lumps &key cg)
1812 (let ((nodes (map 'list (lambda (lump) (storage lump))
1814 (setf (aref nodes 0) 2)
1815 (assert (every #'~= (apply #'concatenate 'list nodes) '(2 3 6 9)))))
1817 431: alien strucure redefinition doesn't work as expected