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 Sometimes (SB-EXT:QUIT) fails with
89 Argh! maximum interrupt nesting depth (4096) exceeded, exiting
90 Process inferior-lisp exited abnormally with code 1
91 I haven't noticed a repeatable case of this yet.
94 And as long as we're wishing, it would be awfully nice if INSPECT could
95 also report on closures, telling about the values of the bound variables.
97 Currently INSPECT and DESCRIBE do show the values, but showing the
98 names of the bindings would be even nicer.
101 The compiler assumes that any time a function of declared FTYPE
102 doesn't signal an error, its arguments were of the declared type.
103 E.g. compiling and loading
104 (DECLAIM (OPTIMIZE (SAFETY 3)))
105 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
107 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
109 (COND ((> (FACTORIAL X) 1.0E6)
110 (FORMAT T "too big~%"))
112 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
114 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
117 will cause the INTEGERP case to be selected, giving bogus output a la
119 This violates the "declarations are assertions" principle.
120 According to the ANSI spec, in the section "System Class FUNCTION",
121 this is a case of "lying to the compiler", but the lying is done
122 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
123 not by the unexpectedly general definition of FACTORIAL. In any case,
124 "declarations are assertions" means that lying to the compiler should
125 cause an error to be signalled, and should not cause a bogus
126 result to be returned. Thus, the compiler should not assume
127 that arbitrary functions check their argument types. (It might
128 make sense to add another flag (CHECKED?) to DEFKNOWN to
129 identify functions which *do* check their argument types.)
130 (Also, verify that the compiler handles declared function
131 return types as assertions.)
134 The definitions of SIGCONTEXT-FLOAT-REGISTER and
135 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
136 supported on FreeBSD because the floating point state is not saved,
137 but at least as of FreeBSD 4.0, the floating point state *is* saved,
138 so they could be supported after all. Very likely
139 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
142 Compiling and loading
143 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
145 then requesting a BACKTRACE at the debugger prompt gives no information
146 about where in the user program the problem occurred.
148 (this is apparently mostly fixed on the SPARC, PPC, and x86 architectures:
149 while giving the backtrace the non-x86 systems complains about "unknown
150 source location: using block start", but apart from that the
151 backtrace seems reasonable. On x86 this is masked by bug 353. See
152 tests/debug.impure.lisp for a test case)
155 Using the pretty-printer from the command prompt gives funny
156 results, apparently because the pretty-printer doesn't know
157 about user's command input, including the user's carriage return
158 that the user, and therefore the pretty-printer thinks that
159 the new output block should start indented 2 or more characters
160 rightward of the correct location.
163 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
164 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
165 crashes SBCL. In general tracing anything which is used in the
166 implementation of TRACE is likely to have the same problem.
169 ANSI says in one place that type declarations can be abbreviated even
170 when the type name is not a symbol, e.g.
171 (DECLAIM ((VECTOR T) *FOOVECTOR*))
172 SBCL doesn't support this. But ANSI says in another place that this
173 isn't allowed. So it's not clear this is a bug after all. (See the
174 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
178 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
179 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
180 the mean of the distribution can be systematically O(0.1%) wrong.
181 Just increasing R-I-E-B is probably not a good solution, since
182 it would decrease efficiency more than is probably necessary. Perhaps
183 using some sort of accept/reject method would be better.
186 Internally the compiler sometimes evaluates
187 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
188 (I stumbled across this when I added an
189 (assert (not (eq type1 *wild-type*)))
190 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
191 in a type context should probably be translated to T, and so it's
192 probably wrong to ask whether it's equal to the T type and then (using
193 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
194 (I haven't tried to investigate this bug enough to guess whether
195 there might be any user-level symptoms.)
197 In fact, the type system is likely to depend on this inequality not
198 holding... * is not equivalent to T in many cases, such as
199 (VECTOR *) /= (VECTOR T).
202 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
203 CL), out-of-line structure slot setters are horribly inefficient
204 whenever the type of the slot is declared, because out-of-line
205 structure slot setters are implemented as closures to save space,
206 so the compiler doesn't compile the type test into code, but
207 instead just saves the type in a lexical closure and interprets it
209 To exercise the problem, compile and load
210 (cl:in-package :cl-user)
212 (bar (error "missing") :type bar))
215 (loop (setf (foo-bar *foo*) x)))
217 (defvar *bar* (make-bar))
218 (defvar *foo* (make-foo :bar *bar*))
219 (defvar *setf-foo-bar* #'(setf foo-bar))
221 (loop (funcall *setf-foo-bar* x *foo*)))
222 then run (WASTREL1 *BAR*) or (WASTREL2 *BAR*), hit Ctrl-C, and
223 use BACKTRACE, to see it's spending all essentially all its time
224 in %TYPEP and VALUES-SPECIFIER-TYPE and so forth.
225 One possible solution would be simply to give up on
226 representing structure slot accessors as functions, and represent
227 them as macroexpansions instead. This can be inconvenient for users,
228 but it's not clear that it's worse than trying to help by expanding
229 into a horribly inefficient implementation.
230 As a workaround for the problem, #'(SETF FOO) expressions
231 can be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
232 (defmacro efficient-setf-function (place-function-name)
233 (or #+sbcl (and (sb-int:info :function :accessor-for place-function-name)
234 ;; a workaround for the problem, encouraging the
235 ;; inline expansion of the structure accessor, so
236 ;; that the compiler can optimize its type test
237 (let ((new-value (gensym "NEW-VALUE-"))
238 (structure-value (gensym "STRUCTURE-VALUE-")))
239 `(lambda (,new-value ,structure-value)
240 (setf (,place-function-name ,structure-value)
242 ;; no problem, can just use the ordinary expansion
243 `(function (setf ,place-function-name))))
246 There's apparently a bug in CEILING optimization which caused
247 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
248 applied the patches to SBCL and they didn't seem to cause problems
249 (as reported sbcl-devel 2001-05-04). However, since the patches
250 modify nontrivial code which was apparently written incorrectly
251 the first time around, until regression tests are written I'm not
252 comfortable merging the patches in the CVS version of SBCL.
255 (TIME (ROOM T)) reports more than 200 Mbytes consed even for
256 a clean, just-started SBCL system. And it seems to be right:
257 (ROOM T) can bring a small computer to its knees for a *long*
258 time trying to GC afterwards. Surely there's some more economical
259 way to implement (ROOM T).
261 Daniel Barlow doesn't know what fixed this, but observes that it
262 doesn't seem to be the case in 0.8.7.3 any more. Instead, (ROOM T)
263 in a fresh SBCL causes
265 debugger invoked on a SB-INT:BUG in thread 5911:
266 failed AVER: "(SAP= CURRENT END)"
268 unless a GC has happened beforehand.
271 When the compiler inline expands functions, it may be that different
272 kinds of return values are generated from different code branches.
273 E.g. an inline expansion of POSITION generates integer results
274 from one branch, and NIL results from another. When that inline
275 expansion is used in a context where only one of those results
278 (aref *a1* (position x *a2*)))
279 and the compiler can't prove that the unacceptable branch is
280 never taken, then bogus type mismatch warnings can be generated.
281 If you need to suppress the type mismatch warnings, you can
282 suppress the inline expansion,
284 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
285 (aref *a1* (position x *a2*)))
286 or, sometimes, suppress them by declaring the result to be of an
289 (aref *a1* (the integer (position x *a2*))))
291 This is not a new compiler problem in 0.7.0, but the new compiler
292 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
293 more conspicuous. If you don't need performance from these functions,
294 and the bogus warnings are a nuisance for you, you can return to
295 your pre-0.7.0 state of grace with
296 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
301 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
302 the entire lexical environment at the point of MACROLET available
303 in the bodies of the macroexpander functions. In particular, it
304 allows the function bodies (which run at compile time) to try to
305 access lexical variables (which are only defined at runtime).
306 It doesn't even issue a warning, which is bad.
308 The SBCL behavior arguably conforms to the ANSI spec (since the
309 spec says that the behavior is undefined, ergo anything conforms).
310 However, it would be better to issue a compile-time error.
311 Unfortunately I (WHN) don't see any simple way to detect this
312 condition in order to issue such an error, so for the meantime
313 SBCL just does this weird broken "conforming" thing.
315 The ANSI standard says, in the definition of the special operator
317 The macro-expansion functions defined by MACROLET are defined
318 in the lexical environment in which the MACROLET form appears.
319 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
320 the local macro definitions in a MACROLET, but the consequences
321 are undefined if the local macro definitions reference any
322 local variable or function bindings that are visible in that
324 Then it seems to contradict itself by giving the example
326 (macrolet ((fudge (z)
327 ;The parameters x and flag are not accessible
328 ; at this point; a reference to flag would be to
329 ; the global variable of that name.
330 ` (if flag (* ,z ,z) ,z)))
331 ;The parameters x and flag are accessible here.
335 The comment "a reference to flag would be to the global variable
336 of the same name" sounds like good behavior for the system to have.
337 but actual specification quoted above says that the actual behavior
340 (Since 0.7.8.23 macroexpanders are defined in a restricted version
341 of the lexical environment, containing no lexical variables and
342 functions, which seems to conform to ANSI and CLtL2, but signalling
343 a STYLE-WARNING for references to variables similar to locals might
347 Ideally, uninterning a symbol would allow it, and its associated
348 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
349 at least as of sbcl-0.7.0, this isn't the case. Information about
350 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
351 essentially in ordinary (non-weak) hash tables keyed by symbols.
352 Thus, once a system has an entry in this system, it tends to live
353 forever, even when it is uninterned and all other references to it
357 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
358 prominent SourceForge web/db bug tracking system, which is
359 unfortunately not a reliable way to get a timely response from
360 the SBCL maintainers)
361 In the course of trying to build a test case for an
362 application error, I encountered this behavior:
363 If you start up sbcl, and then lay on CTRL-C for a
364 minute or two, the lisp process will eventually say:
365 %PRIMITIVE HALT called; the party is over.
366 and throw you into the monitor. If I start up lisp,
367 attach to the process with strace, and then do the same
368 (abusive) thing, I get instead:
369 access failure in heap page not marked as write-protected
370 and the monitor again. I don't know enough to have the
371 faintest idea of what is going on here.
372 This is with sbcl 6.12, uname -a reports:
373 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
374 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
375 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
376 it took more than two minutes (but less than five) for me.
380 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
381 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
382 COMPLEX implementation didn't deal with this, and hasn't been
383 upgraded to do so. (This doesn't seem to be a high priority
384 conformance problem, since seems hard to construct useful code
387 [ partially fixed by CSR in 0.8.17.17 because of a PFD ansi-tests
388 report that (COMPLEX RATIO) was failing; still failing on types of
389 the form (AND NUMBER (SATISFIES REALP) (SATISFIES ZEROP)). ]
391 b. (fixed in 0.8.3.43)
394 Floating point errors are reported poorly. E.g. on x86 OpenBSD
397 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
398 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
399 No traps are enabled? How can this be?
400 It should be possible to be much more specific (overflow, division
401 by zero, etc.) and of course the "How can this be?" should be fixable.
403 See also bugs #45.c and #183
406 (reported by Robert E. Brown 2002-04-16)
407 When a function is called with too few arguments, causing the
408 debugger to be entered, the uninitialized slots in the bad call frame
409 seem to cause GCish problems, being interpreted as tagged data even
410 though they're not. In particular, executing ROOM in the
411 debugger at that point causes AVER failures:
414 * (lisp-implementation-version)
420 failed AVER: "(SAP= CURRENT END)"
421 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
422 isn't too surprising since there are many differences in stack
423 implementation and GC conservatism between the X86 and other ports.)
425 This is probably the same bug as 216
428 The compiler sometimes tries to constant-fold expressions before
429 it checks to see whether they can be reached. This can lead to
430 bogus warnings about errors in the constant folding, e.g. in code
433 (WRITE-STRING (> X 0) "+" "0"))
434 compiled in a context where the compiler can prove that X is NIL,
435 and the compiler complains that (> X 0) causes a type error because
436 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
437 caused a full WARNING, which made the bug really annoying because then
438 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
439 code. Since then the warning has been downgraded to STYLE-WARNING,
440 so it's still a bug but at least it's a little less annoying.
442 183: "IEEE floating point issues"
443 Even where floating point handling is being dealt with relatively
444 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
445 accrued-exceptions and current-exceptions part of the fp control
446 word don't seem to bear much relation to reality. E.g. on
450 debugger invoked on condition of type DIVISION-BY-ZERO:
451 arithmetic error DIVISION-BY-ZERO signalled
452 0] (sb-vm::get-floating-point-modes)
454 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
455 :ROUNDING-MODE :NEAREST
456 :CURRENT-EXCEPTIONS NIL
457 :ACCRUED-EXCEPTIONS (:INEXACT)
460 * (sb-vm::get-floating-point-modes)
461 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
462 :ROUNDING-MODE :NEAREST
463 :CURRENT-EXCEPTIONS (:INEXACT)
464 :ACCRUED-EXCEPTIONS (:INEXACT)
467 188: "compiler performance fiasco involving type inference and UNION-TYPE"
471 (declare (optimize (safety 3)))
472 (declare (optimize (compilation-speed 2)))
473 (declare (optimize (speed 1) (debug 1) (space 1)))
475 (declare (type (integer 0) start))
476 (print (incf start 22))
477 (print (incf start 26))
478 (print (incf start 28)))
480 (declare (type (integer 0) start))
481 (print (incf start 22))
482 (print (incf start 26)))
484 (declare (type (integer 0) start))
485 (print (incf start 22))
486 (print (incf start 26))))))
488 This example could be solved with clever enough constraint
489 propagation or with SSA, but consider
494 The careful type of X is {2k} :-(. Is it really important to be
495 able to work with unions of many intervals?
497 191: "Miscellaneous PCL deficiencies"
498 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
499 a. DEFCLASS does not inform the compiler about generated
500 functions. Compiling a file with
504 (WITH-SLOTS (A-CLASS-X) A
506 results in a STYLE-WARNING:
508 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
510 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
511 Mai points out that the declamation of functions is in fact
512 incorrect in some cases (most notably for structure
513 classes). This means that at present erroneous attempts to use
514 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
515 won't get the corresponding STYLE-WARNING.
517 [much later, in 2006-08] in fact it's no longer erroneous to use
518 WITH-SLOTS on structure-classes. However, including :METACLASS
519 STRUCTURE-CLASS in the class definition gives a whole bunch of
520 function redefinition warnings, so we're still not good to close
523 c. (fixed in 0.8.4.23)
525 201: "Incautious type inference from compound types"
526 a. (reported by APD sbcl-devel 2002-09-17)
528 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
530 (FORMAT NIL "~S IS ~S, Y = ~S"
537 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
541 (declare (type (array * (4 4)) x))
543 (setq x (make-array '(4 4)))
544 (adjust-array y '(3 5))
545 (= (array-dimension y 0) (eval `(array-dimension ,y 0)))))
547 * (foo (make-array '(4 4) :adjustable t))
550 205: "environment issues in cross compiler"
551 (These bugs have no impact on user code, but should be fixed or
553 a. Macroexpanders introduced with MACROLET are defined in the null
555 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
556 the null lexical environment.
557 c. The cross-compiler cannot inline functions defined in a non-null
560 206: ":SB-FLUID feature broken"
561 (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
562 Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
565 207: "poorly distributed SXHASH results for compound data"
566 SBCL's SXHASH could probably try a little harder. ANSI: "the
567 intent is that an implementation should make a good-faith
568 effort to produce hash-codes that are well distributed
569 within the range of non-negative fixnums". But
570 (let ((hits (make-hash-table)))
573 (let* ((ij (cons i j))
574 (newlist (push ij (gethash (sxhash ij) hits))))
576 (format t "~&collision: ~S~%" newlist))))))
577 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
578 would be an obvious way of fix. Maybe it would be acceptably efficient
579 to redo MIX using a lookup into a 256-entry s-box containing
580 29-bit pseudorandom numbers?
582 211: "keywords processing"
583 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
584 number of keyword arguments.
586 212: "Sequence functions and circular arguments"
587 COERCE, MERGE and CONCATENATE go into an infinite loop when given
588 circular arguments; it would be good for the user if they could be
589 given an error instead (ANSI 17.1.1 allows this behaviour on the part
590 of the implementation, as conforming code cannot give non-proper
591 sequences to these functions. MAP also has this problem (and
592 solution), though arguably the convenience of being able to do
593 (MAP 'LIST '+ FOO '#1=(1 . #1#))
594 might be classed as more important (though signalling an error when
595 all of the arguments are circular is probably desireable).
597 213: "Sequence functions and type checking"
598 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
599 check that it will return a sequence of the given type. Fixing
600 it along the same lines as the others (cf. work done around
601 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
602 entirely straightforward.
603 c. All of these functions will silently accept a type of the form
605 whether or not the return value is of this type. This is
606 probably permitted by ANSI (see "Exceptional Situations" under
607 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
608 know about this escape clause, so code of the form
609 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
610 can erroneously return T.
612 215: ":TEST-NOT handling by functions"
613 a. FIND and POSITION currently signal errors when given non-NIL for
614 both their :TEST and (deprecated) :TEST-NOT arguments, but by
615 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
616 means that the effect is "unpredictable but harmless". It's not
617 clear what that actually means; it may preclude conforming
618 implementations from signalling errors.
619 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
620 when conflict occurs. As a quality of implementation issue, it
621 might be preferable to treat :TEST and :TEST-NOT as being in some
622 sense the same &KEY, and effectively take the first test function in
624 c. Again, a quality of implementation issue: it would be good to issue a
625 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
626 WARNING for calls with both :TEST and :TEST-NOT; possibly this
627 latter should be WARNed about at execute-time too.
629 216: "debugger confused by frames with invalid number of arguments"
630 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
631 leaves the system confused, enough so that (QUIT) no longer works.
632 It's as though the process of working with the uninitialized slot in
633 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
634 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
636 This is probably the same bug as 162
638 217: "Bad type operations with FUNCTION types"
641 * (values-type-union (specifier-type '(function (base-char)))
642 (specifier-type '(function (integer))))
644 #<FUN-TYPE (FUNCTION (BASE-CHAR) *)>
646 It causes insertion of wrong type assertions into generated
650 (let ((f (etypecase x
651 (character #'write-char)
652 (integer #'write-byte))))
655 (character (write-char x s))
656 (integer (write-byte x s)))))
658 Then (FOO #\1 *STANDARD-OUTPUT*) signals type error.
660 (In 0.7.9.1 the result type is (FUNCTION * *), so Python does not
661 produce invalid code, but type checking is not accurate.)
663 235: "type system and inline expansion"
665 (declaim (ftype (function (cons) number) acc))
666 (declaim (inline acc))
668 (the number (car c)))
671 (values (locally (declare (optimize (safety 0)))
673 (locally (declare (optimize (safety 3)))
676 (foo '(nil) '(t)) => NIL, T.
678 237: "Environment arguments to type functions"
679 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
680 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
681 argument, but they ignore it completely. This is almost
682 certainly not correct.
683 b. Also, the compiler's optimizers for TYPEP have not been informed
684 about the new argument; consequently, they will not transform
685 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
686 just as optimizeable as (TYPEP 1 'INTEGER).
688 238: "REPL compiler overenthusiasm for CLOS code"
690 * (defclass foo () ())
691 * (defmethod bar ((x foo) (foo foo)) (call-next-method))
692 causes approximately 100 lines of code deletion notes. Some
693 discussion on this issue happened under the title 'Three "interesting"
694 bugs in PCL', resulting in a fix for this oververbosity from the
695 compiler proper; however, the problem persists in the interactor
696 because the notion of original source is not preserved: for the
697 compiler, the original source of the above expression is (DEFMETHOD
698 BAR ((X FOO) (FOO FOO)) (CALL-NEXT-METHOD)), while by the time the
699 compiler gets its hands on the code needing compilation from the REPL,
700 it has been macroexpanded several times.
702 A symptom of the same underlying problem, reported by Tony Martinez:
704 (with-input-from-string (*query-io* " no")
706 (simple-type-error () 'error))
708 ; (SB-KERNEL:FLOAT-WAIT)
710 ; note: deleting unreachable code
711 ; compilation unit finished
714 242: "WRITE-SEQUENCE suboptimality"
715 (observed from clx performance)
716 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
717 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
718 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
719 rather than writing the sequence in one go, leading to severe
720 performance degradation.
721 As of sbcl-0.9.0.36, this is solved for fd-streams, so is less of a
722 problem in practice. (Fully fixing this would require adding a
723 ansi-stream-n-bout slot and associated methods to write a byte
724 sequence to ansi-stream, similar to the existing ansi-stream-sout
727 243: "STYLE-WARNING overenthusiasm for unused variables"
728 (observed from clx compilation)
729 In sbcl-0.7.14, in the presence of the macros
730 (DEFMACRO FOO (X) `(BAR ,X))
731 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
732 somewhat surprising style warnings are emitted for
733 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
735 ; (LAMBDA (Y) (FOO Y))
737 ; caught STYLE-WARNING:
738 ; The variable Y is defined but never used.
740 245: bugs in disassembler
741 b. On X86 operand size prefix is not recognized.
744 (defun foo (&key (a :x))
748 does not cause a warning. (BTW: old SBCL issued a warning, but for a
749 function, which was never called!)
752 Compiler does not emit warnings for
754 a. (lambda () (svref (make-array 8 :adjustable t) 1))
757 (list (let ((y (the real x)))
758 (unless (floatp y) (error ""))
763 (declare (optimize (debug 0)))
764 (declare (type vector x))
765 (list (fill-pointer x)
769 Complex array type does not have corresponding type specifier.
771 This is a problem because the compiler emits optimization notes when
772 you use a non-simple array, and without a type specifier for hairy
773 array types, there's no good way to tell it you're doing it
774 intentionally so that it should shut up and just compile the code.
776 Another problem is confusing error message "asserted type ARRAY
777 conflicts with derived type (VALUES SIMPLE-VECTOR &OPTIONAL)" during
778 compiling (LAMBDA (V) (VALUES (SVREF V 0) (VECTOR-POP V))).
780 The last problem is that when type assertions are converted to type
781 checks, types are represented with type specifiers, so we could lose
782 complex attribute. (Now this is probably not important, because
783 currently checks for complex arrays seem to be performed by
787 (compile nil '(lambda () (aref (make-array 0) 0))) compiles without
788 warning. Analogous cases with the index and length being equal and
789 greater than 0 are warned for; the problem here seems to be that the
790 type required for an array reference of this type is (INTEGER 0 (0))
791 which is canonicalized to NIL.
796 (t1 (specifier-type s)))
797 (eval `(defstruct ,s))
798 (type= t1 (specifier-type s)))
803 b. The same for CSUBTYPEP.
805 262: "yet another bug in inline expansion of local functions"
806 During inline expansion of a local function Python can try to
807 reference optimized away objects (functions, variables, CTRANs from
808 tags and blocks), which later may lead to problems. Some of the
809 cases are worked around by forbidding expansion in such cases, but
810 the better way would be to reimplement inline expansion by copying
814 David Lichteblau provided (sbcl-devel 2003-06-01) a patch to fix
815 behaviour of streams with element-type (SIGNED-BYTE 8). The patch
816 looks reasonable, if not obviously correct; however, it caused the
817 PPC/Linux port to segfault during warm-init while loading
818 src/pcl/std-class.fasl. A workaround patch was made, but it would
819 be nice to understand why the first patch caused problems, and to
820 fix the cause if possible.
822 268: "wrong free declaration scope"
823 The following code must signal type error:
825 (locally (declare (optimize (safety 3)))
826 (flet ((foo (x &optional (y (car x)))
827 (declare (optimize (safety 0)))
829 (funcall (eval #'foo) 1)))
832 In the following function constraint propagator optimizes nothing:
835 (declare (integer x))
836 (declare (optimize speed))
844 Compilation of the following two forms causes "X is unbound" error:
846 (symbol-macrolet ((x pi))
847 (macrolet ((foo (y) (+ x y)))
848 (declaim (inline bar))
854 (See (COERCE (CDR X) 'FUNCTION) in IR1-CONVERT-INLINE-LAMBDA.)
857 CLHS says that type declaration of a symbol macro should not affect
858 its expansion, but in SBCL it does. (If you like magic and want to
859 fix it, don't forget to change all uses of MACROEXPAND to
863 The following code (taken from CLOCC) takes a lot of time to compile:
866 (declare (type (integer 0 #.large-constant) n))
869 (fixed in 0.8.2.51, but a test case would be good)
872 b. The same as in a., but using MULTIPLE-VALUE-SETQ instead of SETQ.
874 (defmethod faa ((*faa* double-float))
875 (set '*faa* (when (< *faa* 0) (- *faa*)))
877 (faa 1d0) => type error
879 279: type propagation error -- correctly inferred type goes astray?
880 In sbcl-0.8.3 and sbcl-0.8.1.47, the warning
881 The binding of ABS-FOO is a (VALUES (INTEGER 0 0)
882 &OPTIONAL), not a (INTEGER 1 536870911)
883 is emitted when compiling this file:
884 (declaim (ftype (function ((integer 0 #.most-positive-fixnum))
885 (integer #.most-negative-fixnum 0))
890 (let* (;; Uncomment this for a type mismatch warning indicating
891 ;; that the type of (FOO X) is correctly understood.
892 #+nil (fs-foo (float-sign (foo x)))
893 ;; Uncomment this for a type mismatch warning
894 ;; indicating that the type of (ABS (FOO X)) is
895 ;; correctly understood.
896 #+nil (fs-abs-foo (float-sign (abs (foo x))))
897 ;; something wrong with this one though
898 (abs-foo (abs (foo x))))
899 (declare (type (integer 1 100) abs-foo))
904 283: Thread safety: libc functions
905 There are places that we call unsafe-for-threading libc functions
906 that we should find alternatives for, or put locks around. Known or
907 strongly suspected problems, as of 0.8.3.10: please update this
908 bug instead of creating new ones
910 gethostbyname, gethostbyaddr in sb-bsd-sockets
912 284: Thread safety: special variables
913 There are lots of special variables in SBCL, and I feel sure that at
914 least some of them are indicative of potentially thread-unsafe
915 parts of the system. See doc/internals/notes/threading-specials
917 286: "recursive known functions"
918 Self-call recognition conflicts with known function
919 recognition. Currently cross compiler and target COMPILE do not
920 recognize recursion, and in target compiler it can be disabled. We
921 can always disable it for known functions with RECURSIVE attribute,
922 but there remains a possibility of a function with a
923 (tail)-recursive simplification pass and transforms/VOPs for base
926 288: fundamental cross-compilation issues (from old UGLINESS file)
927 Using host floating point numbers to represent target floating point
928 numbers, or host characters to represent target characters, is
929 theoretically shaky. (The characters are OK as long as the characters
930 are in the ANSI-guaranteed character set, though, so they aren't a
931 real problem as long as the sources don't need anything but that;
932 the floats are a real problem.)
934 289: "type checking and source-transforms"
936 (block nil (let () (funcall #'+ (eval 'nil) (eval '1) (return :good))))
939 Our policy is to check argument types at the moment of a call. It
940 disagrees with ANSI, which says that type assertions are put
941 immediately onto argument expressions, but is easier to implement in
942 IR1 and is more compatible to type inference, inline expansion,
943 etc. IR1-transforms automatically keep this policy, but source
944 transforms for associative functions (such as +), being applied
945 during IR1-convertion, do not. It may be tolerable for direct calls
946 (+ x y z), but for (FUNCALL #'+ x y z) it is non-conformant.
948 b. Another aspect of this problem is efficiency. [x y + z +]
949 requires less registers than [x y z + +]. This transformation is
950 currently performed with source transforms, but it would be good to
951 also perform it in IR1 optimization phase.
953 290: Alpha floating point and denormalized traps
954 In SBCL 0.8.3.6x on the alpha, we work around what appears to be a
955 hardware or kernel deficiency: the status of the enable/disable
956 denormalized-float traps bit seems to be ambiguous; by the time we
957 get to os_restore_fp_control after a trap, denormalized traps seem
958 to be enabled. Since we don't want a trap every time someone uses a
959 denormalized float, in general, we mask out that bit when we restore
960 the control word; however, this clobbers any change the user might
964 LOOP with non-constant arithmetic step clauses suffers from overzealous
965 type constraint: code of the form
966 (loop for d of-type double-float from 0d0 to 10d0 by x collect d)
967 compiles to a type restriction on X of (AND DOUBLE-FLOAT (REAL
968 (0))). However, an integral value of X should be legal, because
969 successive adds of integers to double-floats produces double-floats,
970 so none of the type restrictions in the code is violated.
972 300: (reported by Peter Graves) Function PEEK-CHAR checks PEEK-TYPE
973 argument type only after having read a character. This is caused
974 with EXPLICIT-CHECK attribute in DEFKNOWN. The similar problem
975 exists with =, /=, <, >, <=, >=. They were fixed, but it is probably
976 less error prone to have EXPLICIT-CHECK be a local declaration,
977 being put into the definition, instead of an attribute being kept in
978 a separate file; maybe also put it into SB-EXT?
980 301: ARRAY-SIMPLE-=-TYPE-METHOD breaks on corner cases which can arise
981 in NOTE-ASSUMED-TYPES
982 In sbcl-0.8.7.32, compiling the file
984 (declare (type integer x))
985 (declare (type (vector (or hash-table bit)) y))
988 (declare (type integer x))
989 (declare (type (simple-array base (2)) y))
992 failed AVER: "(NOT (AND (NOT EQUALP) CERTAINP))"
994 303: "nonlinear LVARs" (aka MISC.293)
996 (multiple-value-call #'list
998 (multiple-value-prog1
999 (eval '(values :a :b :c))
1005 (throw 'bar (values 3 4)))))))))))
1007 (BUU 1) returns garbage.
1009 The problem is that both EVALs sequentially write to the same LVAR.
1011 306: "Imprecise unions of array types"
1013 (declare (optimize speed)
1014 (type (or (array cons) (array vector)) x))
1016 (foo #((0))) => TYPE-ERROR
1023 ,@(loop for x across sb-vm:*specialized-array-element-type-properties*
1024 collect `(array ,(sb-vm:saetp-specifier x)))))
1025 => NIL, T (when it should be T, T)
1027 309: "Dubious values for implementation limits"
1028 (reported by Bruno Haible sbcl-devel "Incorrect value of
1029 multiple-values-limit" 2004-04-19)
1030 (values-list (make-list 1000000)), on x86/linux, signals a stack
1031 exhaustion condition, despite MULTIPLE-VALUES-LIMIT being
1032 significantly larger than 1000000. There are probably similar
1033 dubious values for CALL-ARGUMENTS-LIMIT (see cmucl-help/cmucl-imp
1034 around the same time regarding a call to LIST on sparc with 1000
1035 arguments) and other implementation limit constants.
1037 314: "LOOP :INITIALLY clauses and scope of initializers"
1038 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
1039 test suite, originally by Thomas F. Burdick.
1040 ;; <http://www.lisp.org/HyperSpec/Body/sec_6-1-7-2.html>
1041 ;; According to the HyperSpec 6.1.2.1.4, in for-as-equals-then, var is
1042 ;; initialized to the result of evaluating form1. 6.1.7.2 says that
1043 ;; initially clauses are evaluated in the loop prologue, which precedes all
1044 ;; loop code except for the initial settings provided by with, for, or as.
1045 (loop :for x = 0 :then (1+ x)
1046 :for y = (1+ x) :then (ash y 1)
1047 :for z :across #(1 3 9 27 81 243)
1049 :initially (assert (zerop x)) :initially (assert (= 2 w))
1050 :until (>= w 100) :collect w)
1051 Expected: (2 6 15 38)
1054 318: "stack overflow in compiler warning with redefined class"
1055 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
1058 (setf (find-class 'foo) nil)
1059 (defstruct foo slot-1)
1060 This used to give a stack overflow from within the printer, which has
1061 been fixed as of 0.8.16.11. Current result:
1063 ; can't compile TYPEP of anonymous or undefined class:
1064 ; #<SB-KERNEL:STRUCTURE-CLASSOID FOO>
1066 debugger invoked on a TYPE-ERROR in thread 19973:
1067 The value NIL is not of type FUNCTION.
1069 CSR notes: it's not really clear what it should give: is (SETF FIND-CLASS)
1070 meant to be enough to delete structure classes from the system?
1072 319: "backquote with comma inside array"
1073 reported by Bruno Haible sbcl-devel "various SBCL bugs" from CLISP
1075 (read-from-string "`#1A(1 2 ,(+ 2 2) 4)")
1077 #(1 2 ((SB-IMPL::|,|) + 2 2) 4)
1078 which probably isn't intentional.
1080 324: "STREAMs and :ELEMENT-TYPE with large bytesize"
1081 In theory, (open foo :element-type '(unsigned-byte <x>)) should work
1082 for all positive integral <x>. At present, it only works for <x> up
1083 to about 1024 (and similarly for signed-byte), so
1084 (open "/dev/zero" :element-type '(unsigned-byte 1025))
1085 gives an error in sbcl-0.8.10.
1087 325: "CLOSE :ABORT T on supeseding streams"
1088 Closing a stream opened with :IF-EXISTS :SUPERSEDE with :ABORT T leaves no
1089 file on disk, even if one existed before opening.
1091 The illegality of this is not crystal clear, as the ANSI dictionary
1092 entry for CLOSE says that when :ABORT is T superseded files are not
1093 superseded (ie. the original should be restored), whereas the OPEN
1094 entry says about :IF-EXISTS :SUPERSEDE "If possible, the
1095 implementation should not destroy the old file until the new stream
1096 is closed." -- implying that even though undesirable, early deletion
1097 is legal. Restoring the original would none the less be the polite
1100 326: "*PRINT-CIRCLE* crosstalk between streams"
1101 In sbcl-0.8.10.48 it's possible for *PRINT-CIRCLE* references to be
1102 mixed between streams when output operations are intermingled closely
1103 enough (as by doing output on S2 from within (PRINT-OBJECT X S1) in the
1104 test case below), so that e.g. the references #2# appears on a stream
1105 with no preceding #2= on that stream to define it (because the #2= was
1106 sent to another stream).
1107 (cl:in-package :cl-user)
1108 (defstruct foo index)
1109 (defparameter *foo* (make-foo :index 4))
1111 (defparameter *bar* (make-bar))
1112 (defparameter *tangle* (list *foo* *bar* *foo*))
1113 (defmethod print-object ((foo foo) stream)
1114 (let ((index (foo-index foo)))
1115 (format *trace-output*
1116 "~&-$- emitting FOO ~D, ambient *BAR*=~S~%"
1118 (format stream "[FOO ~D]" index))
1120 (let ((tsos (make-string-output-stream))
1121 (ssos (make-string-output-stream)))
1122 (let ((*print-circle* t)
1123 (*trace-output* tsos)
1124 (*standard-output* ssos))
1125 (prin1 *tangle* *standard-output*))
1126 (let ((string (get-output-stream-string ssos)))
1127 (unless (string= string "(#1=[FOO 4] #S(BAR) #1#)")
1128 ;; In sbcl-0.8.10.48 STRING was "(#1=[FOO 4] #2# #1#)".:-(
1129 (error "oops: ~S" string)))))
1130 It might be straightforward to fix this by turning the
1131 *CIRCULARITY-HASH-TABLE* and *CIRCULARITY-COUNTER* variables into
1132 per-stream slots, but (1) it would probably be sort of messy faking
1133 up the special variable binding semantics using UNWIND-PROTECT and
1134 (2) it might be sort of a pain to test that no other bugs had been
1137 328: "Profiling generic functions", transplanted from #241
1138 (from tonyms on #lisp IRC 2003-02-25)
1139 In sbcl-0.7.12.55, typing
1140 (defclass foo () ((bar :accessor foo-bar)))
1143 (defclass foo () ((bar :accessor foo-bar)))
1144 gives the error message
1145 "#:FOO-BAR already names an ordinary function or a macro."
1147 Problem: when a generic function is profiled, it appears as an ordinary
1148 function to PCL. (Remembering the uninterned accessor is OK, as the
1149 redefinition must be able to remove old accessors from their generic
1152 329: "Sequential class redefinition"
1153 reported by Bruno Haible:
1154 (defclass reactor () ((max-temp :initform 10000000)))
1155 (defvar *r1* (make-instance 'reactor))
1156 (defvar *r2* (make-instance 'reactor))
1157 (slot-value *r1* 'max-temp)
1158 (slot-value *r2* 'max-temp)
1159 (defclass reactor () ((uptime :initform 0)))
1160 (slot-value *r1* 'uptime)
1161 (defclass reactor () ((uptime :initform 0) (max-temp :initform 10000)))
1162 (slot-value *r1* 'max-temp) ; => 10000
1163 (slot-value *r2* 'max-temp) ; => 10000000 oops...
1166 The method effective when the wrapper is obsoleted can be saved
1167 in the wrapper, and then to update the instance just run through
1168 all the old wrappers in order from oldest to newest.
1170 332: "fasl stack inconsistency in structure redefinition"
1171 (reported by Tim Daly Jr sbcl-devel 2004-05-06)
1172 Even though structure redefinition is undefined by the standard, the
1173 following behaviour is suboptimal: running
1174 (defun stimulate-sbcl ()
1175 (let ((filename (format nil "/tmp/~A.lisp" (gensym))))
1176 ;;create a file which redefines a structure incompatibly
1177 (with-open-file (f filename :direction :output :if-exists :supersede)
1178 (print '(defstruct astruct foo) f)
1179 (print '(defstruct astruct foo bar) f))
1180 ;;compile and load the file, then invoke the continue restart on
1181 ;;the structure redefinition error
1182 (handler-bind ((error (lambda (c) (continue c))))
1183 (load (compile-file filename)))))
1185 and choosing the CONTINUE restart yields the message
1186 debugger invoked on a SB-INT:BUG in thread 27726:
1187 fasl stack not empty when it should be
1189 336: "slot-definitions must retain the generic functions of accessors"
1190 reported by Tony Martinez:
1191 (defclass foo () ((bar :reader foo-bar)))
1192 (defun foo-bar (x) x)
1193 (defclass foo () ((bar :reader get-bar))) ; => error, should work
1195 Note: just punting the accessor removal if the fdefinition
1196 is not a generic function is not enough:
1198 (defclass foo () ((bar :reader foo-bar)))
1199 (defvar *reader* #'foo-bar)
1200 (defun foo-bar (x) x)
1201 (defclass foo () ((bar :initform 'ok :reader get-bar)))
1202 (funcall *reader* (make-instance 'foo)) ; should be an error, since
1203 ; the method must be removed
1204 ; by the class redefinition
1206 Fixing this should also fix a subset of #328 -- update the
1207 description with a new test-case then.
1209 339: "DEFINE-METHOD-COMBINATION bugs"
1210 (reported by Bruno Haible via the clisp test suite)
1212 a. Syntax checking laxity (should produce errors):
1213 i. (define-method-combination foo :documentation :operator)
1214 ii. (define-method-combination foo :documentation nil)
1215 iii. (define-method-combination foo nil)
1216 iv. (define-method-combination foo nil nil
1217 (:arguments order &aux &key))
1218 v. (define-method-combination foo nil nil (:arguments &whole))
1219 vi. (define-method-combination foo nil nil (:generic-function))
1220 vii. (define-method-combination foo nil nil (:generic-function bar baz))
1221 viii. (define-method-combination foo nil nil (:generic-function (bar)))
1222 ix. (define-method-combination foo nil ((3)))
1223 x. (define-method-combination foo nil ((a)))
1225 b. define-method-combination arguments lambda list badness
1226 i. &aux args are currently unsupported;
1227 ii. default values of &optional and &key arguments are ignored;
1228 iii. supplied-p variables for &optional and &key arguments are not
1231 c. (fixed in sbcl-0.9.15.15)
1233 344: more (?) ROOM T problems (possibly part of bug 108)
1234 In sbcl-0.8.12.51, and off and on leading up to it, the
1235 SB!VM:MEMORY-USAGE operations in ROOM T caused
1236 unhandled condition (of type SB-INT:BUG):
1237 failed AVER: "(SAP= CURRENT END)"
1238 Several clever people have taken a shot at this without fixing
1239 it; this time around (before sbcl-0.8.13 release) I (WHN) just
1240 commented out the SB!VM:MEMORY-USAGE calls until someone figures
1241 out how to make them work reliably with the rest of the GC.
1243 (Note: there's at least one dubious thing in room.lisp: see the
1244 comment in VALID-OBJ)
1246 346: alpha backtrace
1247 In sbcl-0.8.13, all backtraces from errors caused by internal errors
1248 on the alpha seem to have a "bogus stack frame".
1250 349: PPRINT-INDENT rounding implementation decisions
1251 At present, pprint-indent (and indeed the whole pretty printer)
1252 more-or-less assumes that it's using a monospace font. That's
1253 probably not too silly an assumption, but one piece of information
1254 the current implementation loses is from requests to indent by a
1255 non-integral amount. As of sbcl-0.8.15.9, the system silently
1256 truncates the indentation to an integer at the point of request, but
1257 maybe the non-integral value should be propagated through the
1258 pprinter and only truncated at output? (So that indenting by 1/2
1259 then 3/2 would indent by two spaces, not one?)
1261 352: forward-referenced-class trouble
1262 reported by Bruno Haible on sbcl-devel
1264 (setf (class-name (find-class 'a)) 'b)
1268 Expected: an instance of c, with a slot named x
1269 Got: debugger invoked on a SIMPLE-ERROR in thread 78906:
1270 While computing the class precedence list of the class named C.
1271 The class named B is a forward referenced class.
1272 The class named B is a direct superclass of the class named C.
1274 [ Is this actually a bug? DEFCLASS only replaces an existing class
1275 when the class name is the proper name of that class, and in the
1276 above code the class found by (FIND-CLASS 'A) does not have a
1277 proper name. CSR, 2006-08-07 ]
1279 353: debugger suboptimalities on x86
1280 On x86 backtraces for undefined functions start with a bogus stack
1281 frame, and backtraces for throws to unknown catch tags with a "no
1282 debug information" frame. These are both due to CODE-COMPONENT-FROM-BITS
1283 (used on non-x86 platforms) being a more complete solution then what
1286 On x86/linux large portions of tests/debug.impure.lisp have been commented
1287 out as failures. The probable culprit for these problems is in x86-call-context
1288 (things work fine on x86/freebsd).
1290 More generally, the debugger internals suffer from excessive x86/non-x86
1291 conditionalization and OAOOMization: refactoring the common parts would
1294 354: XEPs in backtraces
1295 Under default compilation policy
1299 Has the XEP for TEST in the backtrace, not the TEST frame itself.
1300 (sparc and x86 at least)
1302 Since SBCL 0.8.20.1 this is hidden unless *SHOW-ENTRY-POINT-DETAILS*
1303 is true (instead there appear two TEST frames at least on ppc). The
1304 underlying cause seems to be that SB-C::TAIL-ANNOTATE will not merge
1305 the tail-call for the XEP, since Python has by that time proved that
1306 the function can never return; same happens if the function holds an
1307 unconditional call to ERROR.
1310 (reported by Bruno Haible)
1311 After the "layout depth conflict" error, the CLOS is left in a state where
1312 it's not possible to define new standard-class subclasses any more.
1314 (defclass prioritized-dispatcher ()
1315 ((dependents :type list :initform nil)))
1316 (defmethod sb-pcl:validate-superclass ((c1 sb-pcl:funcallable-standard-class)
1317 (c2 (eql (find-class 'prioritized-dispatcher))))
1319 (defclass prioritized-generic-function (prioritized-dispatcher standard-generic-function)
1321 (:metaclass sb-pcl:funcallable-standard-class))
1322 ;; ERROR, Quit the debugger with ABORT
1323 (defclass typechecking-reader-class (standard-class)
1325 Expected: #<STANDARD-CLASS TYPECHECKING-READER-CLASS>
1326 Got: ERROR "The assertion SB-PCL::WRAPPERS failed."
1328 [ This test case does not cause the error any more. However,
1329 similar problems can be observed with
1331 (defclass foo (standard-class) ()
1332 (:metaclass sb-mop:funcallable-standard-class))
1333 (sb-mop:finalize-inheritance (find-class 'foo))
1335 (defclass bar (standard-class) ())
1336 (make-instance 'bar)
1339 357: defstruct inheritance of initforms
1340 (reported by Bruno Haible)
1341 When defstruct and defclass (with :metaclass structure-class) are mixed,
1342 1. some slot initforms are ignored by the DEFSTRUCT generated constructor
1344 2. all slot initforms are ignored by MAKE-INSTANCE. (This can be arguably
1345 OK for initforms that were given in a DEFSTRUCT form, but for those
1346 given in a DEFCLASS form, I think it qualifies as a bug.)
1348 (defstruct structure02a
1352 (defclass structure02b (structure02a)
1353 ((slot4 :initform -44)
1356 (slot7 :initform (floor (* pi pi)))
1357 (slot8 :initform 88))
1358 (:metaclass structure-class))
1359 (defstruct (structure02c (:include structure02b (slot8 -88)))
1362 (slot11 (floor (exp 3))))
1364 (let ((a (make-structure02c)))
1365 (list (structure02c-slot4 a)
1366 (structure02c-slot5 a)
1367 (structure02c-slot6 a)
1368 (structure02c-slot7 a)))
1369 Expected: (-44 nil t 9)
1370 Got: (SB-PCL::..SLOT-UNBOUND.. SB-PCL::..SLOT-UNBOUND..
1371 SB-PCL::..SLOT-UNBOUND.. SB-PCL::..SLOT-UNBOUND..)
1373 (let ((b (make-instance 'structure02c)))
1374 (list (structure02c-slot2 b)
1375 (structure02c-slot3 b)
1376 (structure02c-slot4 b)
1377 (structure02c-slot6 b)
1378 (structure02c-slot7 b)
1379 (structure02c-slot8 b)
1380 (structure02c-slot10 b)
1381 (structure02c-slot11 b)))
1382 Expected: (t 3 -44 t 9 -88 t 20)
1383 Got: (0 0 0 0 0 0 0 0)
1385 359: wrong default value for ensure-generic-function's :generic-function-class argument
1386 (reported by Bruno Haible)
1387 ANSI CL is silent on this, but the MOP's specification of ENSURE-GENERIC-FUNCTION says:
1388 "The remaining arguments are the complete set of keyword arguments
1389 received by ENSURE-GENERIC-FUNCTION."
1390 and the spec of ENSURE-GENERIC-FUNCTION-USING-CLASS:
1391 ":GENERIC-FUNCTION-CLASS - a class metaobject or a class name. If it is not
1392 supplied, it defaults to the class named STANDARD-GENERIC-FUNCTION."
1393 This is not the case in SBCL. Test case:
1394 (defclass my-generic-function (standard-generic-function)
1396 (:metaclass sb-pcl:funcallable-standard-class))
1397 (setf (fdefinition 'foo1)
1398 (make-instance 'my-generic-function :name 'foo1))
1399 (ensure-generic-function 'foo1
1400 :generic-function-class (find-class 'standard-generic-function))
1402 ; => #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1403 (setf (fdefinition 'foo2)
1404 (make-instance 'my-generic-function :name 'foo2))
1405 (ensure-generic-function 'foo2)
1407 Expected: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS STANDARD-GENERIC-FUNCTION>
1408 Got: #<SB-MOP:FUNCALLABLE-STANDARD-CLASS MY-GENERIC-FUNCTION>
1410 361: initialize-instance of standard-reader-method ignores :function argument
1411 (reported by Bruno Haible)
1412 Pass a custom :function argument to initialize-instance of a
1413 standard-reader-method instance, but it has no effect.
1414 ;; Check that it's possible to define reader methods that do typechecking.
1416 (defclass typechecking-reader-method (sb-pcl:standard-reader-method)
1418 (defmethod initialize-instance ((method typechecking-reader-method) &rest initargs
1419 &key slot-definition)
1420 (let ((name (sb-pcl:slot-definition-name slot-definition))
1421 (type (sb-pcl:slot-definition-type slot-definition)))
1422 (apply #'call-next-method method
1423 :function #'(lambda (args next-methods)
1424 (declare (ignore next-methods))
1425 (apply #'(lambda (instance)
1426 (let ((value (slot-value instance name)))
1427 (unless (typep value type)
1428 (error "Slot ~S of ~S is not of type ~S: ~S"
1429 name instance type value))
1433 (defclass typechecking-reader-class (standard-class)
1435 (defmethod sb-pcl:validate-superclass ((c1 typechecking-reader-class) (c2 standard-class))
1437 (defmethod reader-method-class ((class typechecking-reader-class) direct-slot &rest args)
1438 (find-class 'typechecking-reader-method))
1439 (defclass testclass25 ()
1440 ((pair :type (cons symbol (cons symbol null)) :initarg :pair :accessor testclass25-pair))
1441 (:metaclass typechecking-reader-class))
1442 (macrolet ((succeeds (form)
1443 `(not (nth-value 1 (ignore-errors ,form)))))
1444 (let ((p (list 'abc 'def))
1445 (x (make-instance 'testclass25)))
1446 (list (succeeds (make-instance 'testclass25 :pair '(seventeen 17)))
1447 (succeeds (setf (testclass25-pair x) p))
1448 (succeeds (setf (second p) 456))
1449 (succeeds (testclass25-pair x))
1450 (succeeds (slot-value x 'pair))))))
1451 Expected: (t t t nil t)
1454 (inspect (first (sb-pcl:generic-function-methods #'testclass25-pair)))
1455 shows that the method was created with a FAST-FUNCTION slot but with a
1456 FUNCTION slot of NIL.
1458 362: missing error when a slot-definition is created without a name
1459 (reported by Bruno Haible)
1460 The MOP says about slot-definition initialization:
1461 "The :NAME argument is a slot name. An ERROR is SIGNALled if this argument
1462 is not a symbol which can be used as a variable name. An ERROR is SIGNALled
1463 if this argument is not supplied."
1465 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition))
1467 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION NIL>
1469 363: missing error when a slot-definition is created with a wrong documentation object
1470 (reported by Bruno Haible)
1471 The MOP says about slot-definition initialization:
1472 "The :DOCUMENTATION argument is a STRING or NIL. An ERROR is SIGNALled
1473 if it is not. This argument default to NIL during initialization."
1475 (make-instance (find-class 'sb-pcl:standard-direct-slot-definition)
1477 :documentation 'not-a-string)
1479 Got: #<SB-MOP:STANDARD-DIRECT-SLOT-DEFINITION FOO>
1481 367: TYPE-ERROR at compile time, undetected TYPE-ERROR at runtime
1483 (declaim (optimize (safety 3) (debug 2) (speed 2) (space 1)))
1487 (i367s (make-array 0 :fill-pointer t) :type (or (vector i367) null)))
1489 (g367 (error "missing :G367") :type g367 :read-only t))
1490 ;;; In sbcl-0.8.18, commenting out this (DECLAIM (FTYPE ... R367))
1491 ;;; gives an internal error at compile time:
1492 ;;; The value #<SB-KERNEL:NAMED-TYPE NIL> is not of
1493 ;;; type SB-KERNEL:VALUES-TYPE.
1494 (declaim (ftype (function ((vector i367) e367) (or s367 null)) r367))
1495 (declaim (ftype (function ((vector e367)) (values)) h367))
1497 (let ((x (g367-i367s (make-g367))))
1498 (let* ((y (or (r367 x w)
1501 (format t "~&Y=~S Z=~S~%" y z)
1503 (defun r367 (x y) (declare (ignore x y)) nil)
1504 (defun h367 (x) (declare (ignore x)) (values))
1505 ;;; In sbcl-0.8.18, executing this form causes an low-level error
1506 ;;; segmentation violation at #X9B0E1F4
1507 ;;; (instead of the TYPE-ERROR that one might like).
1508 (frob 0 (make-e367))
1509 can be made to cause two different problems, as noted in the comments:
1510 bug 367a: Compile and load the file. No TYPE-ERROR is signalled at
1511 run time (in the (S367-G367 Y) form of FROB, when Y is NIL
1512 instead of an instance of S367). Instead (on x86/Linux at least)
1513 we end up with a segfault.
1514 bug 367b: Comment out the (DECLAIM (FTYPE ... R367)), and compile
1515 the file. The compiler fails with TYPE-ERROR at compile time.
1517 368: miscompiled OR (perhaps related to bug 367)
1518 Trying to relax type declarations to find a workaround for bug 367,
1519 it turns out that even when the return type isn't declared (or
1520 declared to be T, anyway) the system remains confused about type
1521 inference in code similar to that for bug 367:
1522 (in-package :cl-user)
1523 (declaim (optimize (safety 3) (debug 2) (speed 2) (space 1)))
1527 (i368s (make-array 0 :fill-pointer t) :type (or (vector i368) null)))
1529 (g368 (error "missing :G368") :type g368 :read-only t))
1530 (declaim (ftype (function (fixnum (vector i368) e368) t) r368))
1531 (declaim (ftype (function (fixnum (vector e368)) t) h368))
1532 (defparameter *h368-was-called-p* nil)
1533 (defun nsu (vertices e368)
1534 (let ((i368s (g368-i368s (make-g368))))
1535 (let ((fuis (r368 0 i368s e368)))
1536 (format t "~&FUIS=~S~%" fuis)
1537 (or fuis (h368 0 i368s)))))
1539 (declare (ignore w x y))
1542 (declare (ignore w x))
1543 (setf *h368-was-called-p* t)
1544 (make-s368 :g368 (make-g368)))
1546 (format t "~&calling NSU~%")
1547 (let ((nsu (nsu #() (make-e368))))
1548 (format t "~&NSU returned ~S~%" nsu)
1549 (format t "~&*H368-WAS-CALLED-P*=~S~%" *h368-was-called-p*)
1550 (assert (s368-p nsu))
1551 (assert *h368-was-called-p*))
1552 In sbcl-0.8.18, both ASSERTs fail, and (DISASSEMBLE 'NSU) shows
1553 that no call to H368 is compiled.
1555 369: unlike-an-intersection behavior of VALUES-TYPE-INTERSECTION
1556 In sbcl-0.8.18.2, the identity $(x \cap y \cap y)=(x \cap y)$
1557 does not hold for VALUES-TYPE-INTERSECTION, even for types which
1558 can be intersected exactly, so that ASSERTs fail in this test case:
1559 (in-package :cl-user)
1560 (let ((types (mapcar #'sb-c::values-specifier-type
1561 '((values (vector package) &optional)
1562 (values (vector package) &rest t)
1563 (values (vector hash-table) &rest t)
1564 (values (vector hash-table) &optional)
1565 (values t &optional)
1567 (values nil &optional)
1568 (values nil &rest t)
1569 (values sequence &optional)
1570 (values sequence &rest t)
1571 (values list &optional)
1572 (values list &rest t)))))
1575 (let ((i (sb-c::values-type-intersection x y)))
1576 (assert (sb-c::type= i (sb-c::values-type-intersection i x)))
1577 (assert (sb-c::type= i (sb-c::values-type-intersection i y)))))))
1579 370: reader misbehaviour on large-exponent floats
1580 (read-from-string "1.0s1000000000000000000000000000000000000000")
1581 causes the reader to attempt to create a very large bignum (which it
1582 will then attempt to coerce to a rational). While this isn't
1583 completely wrong, it is probably not ideal -- checking the floating
1584 point control word state and then returning the relevant float
1585 (most-positive-short-float or short-float-infinity) or signalling an
1586 error immediately would seem to make more sense.
1588 372: floating-point overflow not signalled on ppc/darwin
1589 The following assertions in float.pure.lisp fail on ppc/darwin
1590 (Mac OS X version 10.3.7):
1591 (assert (raises-error? (scale-float 1.0 most-positive-fixnum)
1592 floating-point-overflow))
1593 (assert (raises-error? (scale-float 1.0d0 (1+ most-positive-fixnum))
1594 floating-point-overflow)))
1595 as the SCALE-FLOAT just returns
1596 #.SB-EXT:SINGLE/DOUBLE-FLOAT-POSITIVE-INFINITY. These tests have been
1597 disabled on Darwin for now.
1599 377: Memory fault error reporting
1600 On those architectures where :C-STACK-IS-CONTROL-STACK is in
1601 *FEATURES*, we handle SIG_MEMORY_FAULT (SEGV or BUS) on an altstack,
1602 so we cannot handle the signal directly (as in interrupt_handle_now())
1603 in the case when the signal comes from some external agent (the user
1604 using kill(1), or a fault in some foreign code, for instance). As
1605 of sbcl-0.8.20.20, this is fixed by calling
1606 arrange_return_to_lisp_function() to a new error-signalling
1607 function, but as a result the error reporting is poor: we cannot
1608 even tell the user at which address the fault occurred. We should
1609 arrange such that arguments can be passed to the function called from
1610 arrange_return_to_lisp_function(), but this looked hard to do in
1611 general without suffering from memory leaks.
1613 379: TRACE :ENCAPSULATE NIL broken on ppc/darwin
1614 See commented-out test-case in debug.impure.lisp.
1616 380: Accessor redefinition fails because of old accessor name
1617 When redefining an accessor, SB-PCL::FIX-SLOT-ACCESSORS may try to
1618 find the generic function named by the old accessor name using
1619 ENSURE-GENERIC-FUNCTION and then remove the old accessor's method in
1620 the GF. If the old name does not name a function, or if the old name
1621 does not name a generic function, no attempt to find the GF or remove
1622 any methods is made.
1624 However, if an unrelated GF with an incompatible lambda list exists,
1625 the class redefinition will fail when SB-PCL::REMOVE-READER-METHOD
1626 tries to find and remove a method with an incompatible lambda list
1627 from the unrelated generic function.
1629 381: incautious calls to EQUAL in fasl dumping
1631 (frob #(#1=(a #1#)))
1632 (frob #(#1=(b #1#)))
1633 (frob #(#1=(a #1#)))
1634 in sbcl-0.9.0 causes CONTROL-STACK-EXHAUSTED. My (WHN) impression
1635 is that this follows from the use of (MAKE-HASH-TABLE :TEST 'EQUAL)
1636 to detect sharing, in which case fixing it might require either
1637 getting less ambitious about detecting shared list structure, or
1638 implementing the moral equivalent of EQUAL hash tables in a
1641 382: externalization unexpectedly changes array simplicity
1642 COMPILE-FILE and LOAD
1644 (let ((x #.(make-array 4 :fill-pointer 0)))
1645 (values (eval `(typep ',x 'simple-array))
1646 (typep x 'simple-array))))
1647 then (FOO) => T, NIL.
1649 Similar problems exist with SIMPLE-ARRAY-P, ARRAY-HEADER accessors
1650 and all array dimension functions.
1652 383: ASH'ing non-constant zeros
1655 (declare (type (integer -2 14) b))
1656 (declare (ignorable b))
1657 (ash (imagpart b) 57))
1658 on PPC (and other platforms, presumably) gives an error during the
1659 emission of FASH-ASH-LEFT/FIXNUM=>FIXNUM as the assembler attempts to
1660 stuff a too-large constant into the immediate field of a PPC
1661 instruction. Either the VOP should be fixed or the compiler should be
1662 taught how to transform this case away, paying particular attention
1663 to side-effects that might occur in the arguments to ASH.
1665 384: Compiler runaway on very large character types
1667 (compile nil '(lambda (x)
1668 (declare (type (member #\a 1) x))
1669 (the (member 1 nil) x)))
1671 The types apparently normalize into a very large type, and the compiler
1672 gets lost in REMOVE-DUPLICATES. Perhaps the latter should use
1673 a better algorithm (one based on hash tables, say) on very long lists
1674 when :TEST has its default value?
1678 (compile nil '(lambda (x) (the (not (eql #\a)) x)))
1680 (partially fixed in 0.9.3.1, but a better representation for these
1684 (format nil "~4,1F" 0.001) => "0.00" (should be " 0.0");
1685 (format nil "~4,1@F" 0.001) => "+.00" (should be "+0.0").
1687 386: SunOS/x86 stack exhaustion handling broken
1688 According to <http://alfa.s145.xrea.com/sbcl/solaris-x86.html>, the
1689 stack exhaustion checking (implemented with a write-protected guard
1690 page) does not work on SunOS/x86.
1693 12:10 < jsnell> the package-lock test is basically due to a change in the test
1694 behaviour when you install a handler for error around it. I
1695 thought I'd disabled the test for now, but apparently that was
1697 12:19 < Xophe> jsnell: ah, I see the problem in the package-locks stuff
1698 12:19 < Xophe> it's the same problem as we had with compiler-error conditions
1699 12:19 < Xophe> the thing that's signalled up and down the stack is a subtype of
1700 ERROR, where it probably shouldn't be
1703 (found by Dmitry Bogomolov)
1705 (defclass foo () ((x :type (unsigned-byte 8))))
1706 (defclass bar () ((x :type symbol)))
1707 (defclass baz (foo bar) ())
1711 SB-PCL::SPECIALIZER-APPLICABLE-USING-TYPE-P cannot handle the second argument
1714 [ Can't trigger this any more, as of 2006-08-07 ]
1717 (reported several times on sbcl-devel, by Rick Taube, Brian Rowe and
1720 ROUND-NUMERIC-BOUND assumes that float types always have a FORMAT
1721 specifying whether they're SINGLE or DOUBLE. This is true for types
1722 computed by the type system itself, but the compiler type derivation
1723 short-circuits this and constructs non-canonical types. A temporary
1724 fix was made to ROUND-NUMERIC-BOUND for the sbcl-0.9.6 release, but
1725 the right fix is to remove the abstraction violation in the
1726 compiler's type deriver.
1728 393: Wrong error from methodless generic function
1729 (DEFGENERIC FOO (X))
1731 gives NO-APPLICABLE-METHOD rather than an argument count error.
1733 395: Unicode and streams
1734 One of the remaining problems in SBCL's Unicode support is the lack
1735 of generality in certain streams.
1736 a. FILL-POINTER-STREAMs: SBCL refuses to write (e.g. using FORMAT)
1737 to streams made from strings that aren't character strings with
1739 (let ((v (make-array 5 :fill-pointer 0 :element-type 'standard-char)))
1742 should return a non-simple base string containing "foo" but
1745 (reported on sbcl-help by "tichy")
1747 396: block-compilation bug
1751 (when (funcall (eval #'(lambda (x) (eql x 2))) y)
1753 (incf x (incf y z))))))
1757 (bar 1) => 11, should be 4.
1760 The more interrupts arrive the less accurate SLEEP's timing gets.
1761 (time (sb-thread:terminate-thread
1762 (prog1 (sb-thread:make-thread (lambda ()
1769 398: GC-unsafe SB-ALIEN string deporting
1770 Translating a Lisp string to an alien string by taking a SAP to it
1771 as done by the :DEPORT-GEN methods for C-STRING and UTF8-STRING
1772 is not safe, since the Lisp string can move. For example the
1773 following code will fail quickly on both cheneygc and pre-0.9.8.19
1776 (setf (bytes-consed-between-gcs) 4096)
1777 (define-alien-routine "strcmp" int (s1 c-string) (s2 c-string))
1780 (let ((string "hello, world"))
1781 (assert (zerop (strcmp string string)))))
1783 (This will appear to work on post-0.9.8.19 GENCGC, since
1784 the GC no longer zeroes memory immediately after releasing
1785 it after a minor GC. Either enabling the READ_PROTECT_FREE_PAGES
1786 #define in gencgc.c or modifying the example so that a major
1787 GC will occasionally be triggered would unmask the bug.)
1789 On cheneygc the only solution would seem to be allocating some alien
1790 memory, copying the data over, and arranging that it's freed once we
1791 return. For GENCGC we could instead try to arrange that the string
1792 from which the SAP is taken is always pinned.
1794 For some more details see comments for (define-alien-type-method
1795 (c-string :deport-gen) ...) in host-c-call.lisp.
1797 402: "DECLAIM DECLARATION does not inform the PCL code-walker"
1798 reported by Vincent Arkesteijn:
1800 (declaim (declaration foo))
1801 (defgeneric bar (x))
1806 ==> WARNING: The declaration FOO is not understood by
1807 SB-PCL::SPLIT-DECLARATIONS.
1808 Please put FOO on one of the lists SB-PCL::*NON-VAR-DECLARATIONS*,
1809 SB-PCL::*VAR-DECLARATIONS-WITH-ARG*, or
1810 SB-PCL::*VAR-DECLARATIONS-WITHOUT-ARG*.
1811 (Assuming it is a variable declaration without argument).
1813 403: FORMAT/PPRINT-LOGICAL-BLOCK of CONDITIONs ignoring *PRINT-CIRCLE*
1816 (make-condition 'simple-error
1817 :format-control "ow... ~S"
1818 :format-arguments '(#1=(#1#))))
1819 (setf *print-circle* t *print-level* 4)
1820 (format nil "~@<~A~:@>" *c*)
1823 where I (WHN) believe the correct result is "ow... #1=(#1#)",
1824 like the result from (PRINC-TO-STRING *C*). The question of
1825 what the correct result is is complicated by the hairy text in
1826 the Hyperspec "22.3.5.2 Tilde Less-Than-Sign: Logical Block",
1827 Other than the difference in its argument, ~@<...~:> is
1828 exactly the same as ~<...~:> except that circularity detection
1829 is not applied if ~@<...~:> is encountered at top level in a
1831 But because the odd behavior happens even without the at-sign,
1832 (format nil "~<~A~:@>" (list *c*)) ; => "ow... (((#)))"
1833 and because something seemingly similar can happen even in
1834 PPRINT-LOGICAL-BLOCK invoked directly without FORMAT,
1835 (pprint-logical-block (*standard-output* '(some nonempty list))
1836 (format *standard-output* "~A" '#1=(#1#)))
1837 (which prints "(((#)))" to *STANDARD-OUTPUT*), I don't think
1838 that the 22.3.5.2 trickiness is fundamental to the problem.
1840 My guess is that the problem is related to the logic around the MODE
1841 argument to CHECK-FOR-CIRCULARITY, but I haven't reverse-engineered
1842 enough of the intended meaning of the different MODE values to be
1845 404: nonstandard DWIMness in LOOP with unportably-ordered clauses
1846 In sbcl-0.9.13, the code
1847 (loop with stack = (make-array 2 :fill-pointer 2 :initial-element t)
1848 for length = (length stack)
1849 while (plusp length)
1850 for element = (vector-pop stack)
1852 compiles without error or warning and returns (T T). Unfortunately,
1853 it is inconsistent with the ANSI definition of the LOOP macro,
1854 because it mixes up VARIABLE-CLAUSEs with MAIN-CLAUSEs. Furthermore,
1855 SBCL's interpretation of the intended meaning is only one possible,
1856 unportable interpretation of the noncompliant code; in CLISP 2.33.2,
1857 the code compiles with a warning
1858 LOOP: FOR clauses should occur before the loop's main body
1859 and then fails at runtime with
1860 VECTOR-POP: #() has length zero
1861 perhaps because CLISP has shuffled the clauses into an
1862 ANSI-compliant order before proceeding.