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 The "compiling top-level form:" output ought to be condensed.
89 Perhaps any number of such consecutive lines ought to turn into a
90 single "compiling top-level forms:" line.
93 It would be nice if the
95 (during macroexpansion)
96 said what macroexpansion was at fault, e.g.
98 (during macroexpansion of IN-PACKAGE,
99 during macroexpansion of DEFFOO)
102 (I *think* this is a bug. It certainly seems like strange behavior. But
103 the ANSI spec is scary, dark, and deep.. -- WHN)
104 (FORMAT NIL "~,1G" 1.4) => "1. "
105 (FORMAT NIL "~3,1G" 1.4) => "1. "
108 Sometimes (SB-EXT:QUIT) fails with
109 Argh! maximum interrupt nesting depth (4096) exceeded, exiting
110 Process inferior-lisp exited abnormally with code 1
111 I haven't noticed a repeatable case of this yet.
114 The printer doesn't report closures very well. This is true in
118 #<Closure Over Function "DEFUN STRUCTURE-SLOT-ACCESSOR" {134D1A1}>
119 It would be nice to make closures have a settable name slot,
120 and make things like DEFSTRUCT and FLET, which create closures,
121 set helpful values into this slot.
124 And as long as we're wishing, it would be awfully nice if INSPECT could
125 also report on closures, telling about the values of the bound variables.
128 The compiler assumes that any time a function of declared FTYPE
129 doesn't signal an error, its arguments were of the declared type.
130 E.g. compiling and loading
131 (DECLAIM (OPTIMIZE (SAFETY 3)))
132 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
134 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
136 (COND ((> (FACTORIAL X) 1.0E6)
137 (FORMAT T "too big~%"))
139 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
141 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
144 will cause the INTEGERP case to be selected, giving bogus output a la
146 This violates the "declarations are assertions" principle.
147 According to the ANSI spec, in the section "System Class FUNCTION",
148 this is a case of "lying to the compiler", but the lying is done
149 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
150 not by the unexpectedly general definition of FACTORIAL. In any case,
151 "declarations are assertions" means that lying to the compiler should
152 cause an error to be signalled, and should not cause a bogus
153 result to be returned. Thus, the compiler should not assume
154 that arbitrary functions check their argument types. (It might
155 make sense to add another flag (CHECKED?) to DEFKNOWN to
156 identify functions which *do* check their argument types.)
157 (Also, verify that the compiler handles declared function
158 return types as assertions.)
161 The definitions of SIGCONTEXT-FLOAT-REGISTER and
162 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
163 supported on FreeBSD because the floating point state is not saved,
164 but at least as of FreeBSD 4.0, the floating point state *is* saved,
165 so they could be supported after all. Very likely
166 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
169 a slew of floating-point-related errors reported by Peter Van Eynde
171 c: Many expressions generate floating infinity on x86/Linux:
176 PVE's regression tests want them to raise errors. sbcl-0.7.0.5
177 on x86/Linux generates the infinities instead. That might or
178 might not be conforming behavior, but it's also inconsistent,
179 which is almost certainly wrong. (Inconsistency: (/ 1 0.0)
180 should give the same result as (/ 1.0 0.0), but instead (/ 1 0.0)
181 generates SINGLE-FLOAT-POSITIVE-INFINITY and (/ 1.0 0.0)
183 d: (in section12.erg) various forms a la
184 (FLOAT 1 DOUBLE-FLOAT-EPSILON)
185 don't give the right behavior.
188 type safety errors reported by Peter Van Eynde July 25, 2000:
189 k: READ-BYTE is supposed to signal TYPE-ERROR when its argument is
190 not a binary input stream, but instead cheerfully reads from
191 string-input streams, e.g. (MAKE-STRING-INPUT-STREAM "abc").
192 [ Bug was reported as "from character streams", but in 0.8.3.10 we
193 get correct behaviour from (WITH-OPEN-FILE (i "/dev/zero") (READ-BYTE i)) ]
197 The debugger LIST-LOCATIONS command doesn't work properly.
198 (How should it work properly?)
201 Compiling and loading
202 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
204 then requesting a BACKTRACE at the debugger prompt gives no information
205 about where in the user program the problem occurred.
208 Using the pretty-printer from the command prompt gives funny
209 results, apparently because the pretty-printer doesn't know
210 about user's command input, including the user's carriage return
211 that the user, and therefore the pretty-printer thinks that
212 the new output block should start indented 2 or more characters
213 rightward of the correct location.
216 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
217 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
218 crashes SBCL. In general tracing anything which is used in the
219 implementation of TRACE is likely to have the same problem.
222 ANSI says in one place that type declarations can be abbreviated even
223 when the type name is not a symbol, e.g.
224 (DECLAIM ((VECTOR T) *FOOVECTOR*))
225 SBCL doesn't support this. But ANSI says in another place that this
226 isn't allowed. So it's not clear this is a bug after all. (See the
227 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
231 as pointed out by Dan Barlow on sbcl-devel 2000-07-02:
232 The PICK-TEMPORARY-FILE-NAME utility used by LOAD-FOREIGN uses
233 an easily guessable temporary filename in a way which might open
234 applications using LOAD-FOREIGN to hijacking by malicious users
235 on the same machine. Incantations for doing this safely are
236 floating around the net in various "how to write secure programs
237 despite Unix" documents, and it would be good to (1) fix this in
238 LOAD-FOREIGN, and (2) hunt for any other code which uses temporary
239 files and make it share the same new safe logic.
241 (partially alleviated in sbcl-0.7.9.32 by a fix by Matthew Danish to
242 make the temporary filename less easily guessable)
245 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
246 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
247 the mean of the distribution can be systematically O(0.1%) wrong.
248 Just increasing R-I-E-B is probably not a good solution, since
249 it would decrease efficiency more than is probably necessary. Perhaps
250 using some sort of accept/reject method would be better.
253 Internally the compiler sometimes evaluates
254 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
255 (I stumbled across this when I added an
256 (assert (not (eq type1 *wild-type*)))
257 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
258 in a type context should probably be translated to T, and so it's
259 probably wrong to ask whether it's equal to the T type and then (using
260 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
261 (I haven't tried to investigate this bug enough to guess whether
262 there might be any user-level symptoms.)
264 In fact, the type system is likely to depend on this inequality not
265 holding... * is not equivalent to T in many cases, such as
266 (VECTOR *) /= (VECTOR T).
269 The facility for dumping a running Lisp image to disk gets confused
270 when run without the PURIFY option, and creates an unnecessarily large
271 core file (apparently representing memory usage up to the previous
272 high-water mark). Moreover, when the file is loaded, it confuses the
273 GC, so that thereafter memory usage can never be reduced below that
277 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
278 CL), out-of-line structure slot setters are horribly inefficient
279 whenever the type of the slot is declared, because out-of-line
280 structure slot setters are implemented as closures to save space,
281 so the compiler doesn't compile the type test into code, but
282 instead just saves the type in a lexical closure and interprets it
284 To exercise the problem, compile and load
285 (cl:in-package :cl-user)
287 (bar (error "missing") :type bar))
290 (loop (setf (foo-bar *foo*) x)))
292 (defvar *bar* (make-bar))
293 (defvar *foo* (make-foo :bar *bar*))
294 (defvar *setf-foo-bar* #'(setf foo-bar))
296 (loop (funcall *setf-foo-bar* x *foo*)))
297 then run (WASTREL1 *BAR*) or (WASTREL2 *BAR*), hit Ctrl-C, and
298 use BACKTRACE, to see it's spending all essentially all its time
299 in %TYPEP and VALUES-SPECIFIER-TYPE and so forth.
300 One possible solution would be simply to give up on
301 representing structure slot accessors as functions, and represent
302 them as macroexpansions instead. This can be inconvenient for users,
303 but it's not clear that it's worse than trying to help by expanding
304 into a horribly inefficient implementation.
305 As a workaround for the problem, #'(SETF FOO) expressions
306 can be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
307 (defmacro efficient-setf-function (place-function-name)
308 (or #+sbcl (and (sb-int:info :function :accessor-for place-function-name)
309 ;; a workaround for the problem, encouraging the
310 ;; inline expansion of the structure accessor, so
311 ;; that the compiler can optimize its type test
312 (let ((new-value (gensym "NEW-VALUE-"))
313 (structure-value (gensym "STRUCTURE-VALUE-")))
314 `(lambda (,new-value ,structure-value)
315 (setf (,place-function-name ,structure-value)
317 ;; no problem, can just use the ordinary expansion
318 `(function (setf ,place-function-name))))
321 There's apparently a bug in CEILING optimization which caused
322 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
323 applied the patches to SBCL and they didn't seem to cause problems
324 (as reported sbcl-devel 2001-05-04). However, since the patches
325 modify nontrivial code which was apparently written incorrectly
326 the first time around, until regression tests are written I'm not
327 comfortable merging the patches in the CVS version of SBCL.
330 (TIME (ROOM T)) reports more than 200 Mbytes consed even for
331 a clean, just-started SBCL system. And it seems to be right:
332 (ROOM T) can bring a small computer to its knees for a *long*
333 time trying to GC afterwards. Surely there's some more economical
334 way to implement (ROOM T).
337 When the compiler inline expands functions, it may be that different
338 kinds of return values are generated from different code branches.
339 E.g. an inline expansion of POSITION generates integer results
340 from one branch, and NIL results from another. When that inline
341 expansion is used in a context where only one of those results
344 (aref *a1* (position x *a2*)))
345 and the compiler can't prove that the unacceptable branch is
346 never taken, then bogus type mismatch warnings can be generated.
347 If you need to suppress the type mismatch warnings, you can
348 suppress the inline expansion,
350 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
351 (aref *a1* (position x *a2*)))
352 or, sometimes, suppress them by declaring the result to be of an
355 (aref *a1* (the integer (position x *a2*))))
357 This is not a new compiler problem in 0.7.0, but the new compiler
358 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
359 more conspicuous. If you don't need performance from these functions,
360 and the bogus warnings are a nuisance for you, you can return to
361 your pre-0.7.0 state of grace with
362 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
367 as reported by Eric Marsden on cmucl-imp@cons.org 2001-08-14:
368 (= (FLOAT 1 DOUBLE-FLOAT-EPSILON)
369 (+ (FLOAT 1 DOUBLE-FLOAT-EPSILON) DOUBLE-FLOAT-EPSILON)) => T
370 when of course it should be NIL. (He says it only fails for X86,
371 not SPARC; dunno about Alpha.)
373 Also, "the same problem exists for LONG-FLOAT-EPSILON,
374 DOUBLE-FLOAT-NEGATIVE-EPSILON, LONG-FLOAT-NEGATIVE-EPSILON (though
375 for the -negative- the + is replaced by a - in the test)."
377 Raymond Toy comments that this is tricky on the X86 since its FPU
378 uses 80-bit precision internally.
381 Even in sbcl-0.pre7.x, which is supposed to be free of the old
382 non-ANSI behavior of treating the function return type inferred
383 from the current function definition as a declaration of the
384 return type from any function of that name, the return type of NIL
385 is attached to FOO in 120a above, and used to optimize code which
389 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
390 the entire lexical environment at the point of MACROLET available
391 in the bodies of the macroexpander functions. In particular, it
392 allows the function bodies (which run at compile time) to try to
393 access lexical variables (which are only defined at runtime).
394 It doesn't even issue a warning, which is bad.
396 The SBCL behavior arguably conforms to the ANSI spec (since the
397 spec says that the behavior is undefined, ergo anything conforms).
398 However, it would be better to issue a compile-time error.
399 Unfortunately I (WHN) don't see any simple way to detect this
400 condition in order to issue such an error, so for the meantime
401 SBCL just does this weird broken "conforming" thing.
403 The ANSI standard says, in the definition of the special operator
405 The macro-expansion functions defined by MACROLET are defined
406 in the lexical environment in which the MACROLET form appears.
407 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
408 the local macro definitions in a MACROLET, but the consequences
409 are undefined if the local macro definitions reference any
410 local variable or function bindings that are visible in that
412 Then it seems to contradict itself by giving the example
414 (macrolet ((fudge (z)
415 ;The parameters x and flag are not accessible
416 ; at this point; a reference to flag would be to
417 ; the global variable of that name.
418 ` (if flag (* ,z ,z) ,z)))
419 ;The parameters x and flag are accessible here.
423 The comment "a reference to flag would be to the global variable
424 of the same name" sounds like good behavior for the system to have.
425 but actual specification quoted above says that the actual behavior
428 (Since 0.7.8.23 macroexpanders are defined in a restricted version
429 of the lexical environment, containing no lexical variables and
430 functions, which seems to conform to ANSI and CLtL2, but signalling
431 a STYLE-WARNING for references to variables similar to locals might
435 (as reported by Gabe Garza on cmucl-help 2001-09-21)
437 (defun test-pred (x y)
441 (func (lambda () x)))
442 (print (eq func func))
443 (print (test-pred func func))
444 (delete func (list func))))
445 Now calling (TEST-CASE) gives output
448 (#<FUNCTION {500A9EF9}>)
449 Evidently Python thinks of the lambda as a code transformation so
450 much that it forgets that it's also an object.
453 Ideally, uninterning a symbol would allow it, and its associated
454 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
455 at least as of sbcl-0.7.0, this isn't the case. Information about
456 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
457 essentially in ordinary (non-weak) hash tables keyed by symbols.
458 Thus, once a system has an entry in this system, it tends to live
459 forever, even when it is uninterned and all other references to it
462 141: "pretty printing and backquote"
465 ``(FOO SB-IMPL::BACKQ-COMMA-AT S)
468 * (write '`(, .ala.) :readably t :pretty t)
471 (note the space between the comma and the point)
474 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
475 prominent SourceForge web/db bug tracking system, which is
476 unfortunately not a reliable way to get a timely response from
477 the SBCL maintainers)
478 In the course of trying to build a test case for an
479 application error, I encountered this behavior:
480 If you start up sbcl, and then lay on CTRL-C for a
481 minute or two, the lisp process will eventually say:
482 %PRIMITIVE HALT called; the party is over.
483 and throw you into the monitor. If I start up lisp,
484 attach to the process with strace, and then do the same
485 (abusive) thing, I get instead:
486 access failure in heap page not marked as write-protected
487 and the monitor again. I don't know enough to have the
488 faintest idea of what is going on here.
489 This is with sbcl 6.12, uname -a reports:
490 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
491 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
492 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
493 it took more than two minutes (but less than five) for me.
497 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
498 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
499 COMPLEX implementation didn't deal with this, and hasn't been
500 upgraded to do so. (This doesn't seem to be a high priority
501 conformance problem, since seems hard to construct useful code
506 (declare (type (double-float -0d0) x))
507 (declare (optimize speed))
508 (+ x (sqrt (log (random 1d0)))))
509 debugger invoked on condition of type SIMPLE-ERROR:
510 bad thing to be a type specifier: ((COMPLEX
512 #.SB-EXT:DOUBLE-FLOAT-POSITIVE-INFINITY))
513 #C(0.0d0 #.SB-EXT:DOUBLE-FLOAT-POSITIVE-INFINITY)
514 #C(0.0d0 #.SB-EXT:DOUBLE-FLOAT-POSITIVE-INFINITY))
517 Floating point errors are reported poorly. E.g. on x86 OpenBSD
520 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
521 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
522 No traps are enabled? How can this be?
523 It should be possible to be much more specific (overflow, division
524 by zero, etc.) and of course the "How can this be?" should be fixable.
526 See also bugs #45.c and #183
529 (reported by Robert E. Brown 2002-04-16)
530 When a function is called with too few arguments, causing the
531 debugger to be entered, the uninitialized slots in the bad call frame
532 seem to cause GCish problems, being interpreted as tagged data even
533 though they're not. In particular, executing ROOM in the
534 debugger at that point causes AVER failures:
537 * (lisp-implementation-version)
543 failed AVER: "(SAP= CURRENT END)"
544 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
545 isn't too surprising since there are many differences in stack
546 implementation and GC conservatism between the X86 and other ports.)
548 This is probably the same bug as 216
551 In sbcl-0.7.3.11, compiling the (illegal) code
552 (in-package :cl-user)
553 (defmethod prove ((uustk uustk))
556 gives the (not terribly clear) error message
558 ; (during macroexpansion of (DEFMETHOD PROVE ...))
559 ; can't get template for (FROB NIL NIL)
560 The problem seems to be that the code walker used by the DEFMETHOD
561 macro is unhappy with the illegal syntax in the method body, and
562 is giving an unclear error message.
565 The compiler sometimes tries to constant-fold expressions before
566 it checks to see whether they can be reached. This can lead to
567 bogus warnings about errors in the constant folding, e.g. in code
570 (WRITE-STRING (> X 0) "+" "0"))
571 compiled in a context where the compiler can prove that X is NIL,
572 and the compiler complains that (> X 0) causes a type error because
573 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
574 caused a full WARNING, which made the bug really annoying because then
575 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
576 code. Since then the warning has been downgraded to STYLE-WARNING,
577 so it's still a bug but at least it's a little less annoying.
579 183: "IEEE floating point issues"
580 Even where floating point handling is being dealt with relatively
581 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
582 accrued-exceptions and current-exceptions part of the fp control
583 word don't seem to bear much relation to reality. E.g. on
587 debugger invoked on condition of type DIVISION-BY-ZERO:
588 arithmetic error DIVISION-BY-ZERO signalled
589 0] (sb-vm::get-floating-point-modes)
591 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
592 :ROUNDING-MODE :NEAREST
593 :CURRENT-EXCEPTIONS NIL
594 :ACCRUED-EXCEPTIONS (:INEXACT)
597 * (sb-vm::get-floating-point-modes)
598 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
599 :ROUNDING-MODE :NEAREST
600 :CURRENT-EXCEPTIONS (:INEXACT)
601 :ACCRUED-EXCEPTIONS (:INEXACT)
604 188: "compiler performance fiasco involving type inference and UNION-TYPE"
608 (declare (optimize (safety 3)))
609 (declare (optimize (compilation-speed 2)))
610 (declare (optimize (speed 1) (debug 1) (space 1)))
612 (declare (type (integer 0) start))
613 (print (incf start 22))
614 (print (incf start 26))
615 (print (incf start 28)))
617 (declare (type (integer 0) start))
618 (print (incf start 22))
619 (print (incf start 26)))
621 (declare (type (integer 0) start))
622 (print (incf start 22))
623 (print (incf start 26))))))
625 This example could be solved with clever enough constraint
626 propagation or with SSA, but consider
631 The careful type of X is {2k} :-(. Is it really important to be
632 able to work with unions of many intervals?
634 190: "PPC/Linux pipe? buffer? bug"
635 In sbcl-0.7.6, the run-program.test.sh test script sometimes hangs
636 on the PPC/Linux platform, waiting for a zombie env process. This
637 is a classic symptom of buffer filling and deadlock, but it seems
638 only sporadically reproducible.
640 191: "Miscellaneous PCL deficiencies"
641 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
642 a. DEFCLASS does not inform the compiler about generated
643 functions. Compiling a file with
647 (WITH-SLOTS (A-CLASS-X) A
649 results in a STYLE-WARNING:
651 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
653 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
654 Mai points out that the declamation of functions is in fact
655 incorrect in some cases (most notably for structure
656 classes). This means that at present erroneous attempts to use
657 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
658 won't get the corresponding STYLE-WARNING.
659 c. the examples in CLHS 7.6.5.1 (regarding generic function lambda
660 lists and &KEY arguments) do not signal errors when they should.
662 201: "Incautious type inference from compound types"
663 a. (reported by APD sbcl-devel 2002-09-17)
665 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
667 (FORMAT NIL "~S IS ~S, Y = ~S"
674 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
678 (declare (type (array * (4 4)) x))
680 (setq x (make-array '(4 4)))
681 (adjust-array y '(3 5))
682 (= (array-dimension y 0) (eval `(array-dimension ,y 0)))))
684 * (foo (make-array '(4 4) :adjustable t))
687 205: "environment issues in cross compiler"
688 (These bugs have no impact on user code, but should be fixed or
690 a. Macroexpanders introduced with MACROLET are defined in the null
692 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
693 the null lexical environment.
694 c. The cross-compiler cannot inline functions defined in a non-null
697 206: ":SB-FLUID feature broken"
698 (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
699 Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
702 207: "poorly distributed SXHASH results for compound data"
703 SBCL's SXHASH could probably try a little harder. ANSI: "the
704 intent is that an implementation should make a good-faith
705 effort to produce hash-codes that are well distributed
706 within the range of non-negative fixnums". But
707 (let ((hits (make-hash-table)))
710 (let* ((ij (cons i j))
711 (newlist (push ij (gethash (sxhash ij) hits))))
713 (format t "~&collision: ~S~%" newlist))))))
714 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
715 would be an obvious way of fix. Maybe it would be acceptably efficient
716 to redo MIX using a lookup into a 256-entry s-box containing
717 29-bit pseudorandom numbers?
719 211: "keywords processing"
720 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
721 number of keyword arguments.
724 (flet ((foo (&key y) (list y)))
725 (list (foo :y 1 :y 2)))
727 issues confusing message
732 ; caught STYLE-WARNING:
733 ; The variable #:G15 is defined but never used.
735 212: "Sequence functions and circular arguments"
736 COERCE, MERGE and CONCATENATE go into an infinite loop when given
737 circular arguments; it would be good for the user if they could be
738 given an error instead (ANSI 17.1.1 allows this behaviour on the part
739 of the implementation, as conforming code cannot give non-proper
740 sequences to these functions. MAP also has this problem (and
741 solution), though arguably the convenience of being able to do
742 (MAP 'LIST '+ FOO '#1=(1 . #1#))
743 might be classed as more important (though signalling an error when
744 all of the arguments are circular is probably desireable).
746 213: "Sequence functions and type checking"
747 a. MAKE-SEQUENCE, COERCE, MERGE and CONCATENATE cannot deal with
748 various complicated, though recognizeable, CONS types [e.g.
749 (CONS * (CONS * NULL))
750 which according to ANSI should be recognized] (and, in SAFETY 3
751 code, should return a list of LENGTH 2 or signal an error)
752 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
753 check that it will return a sequence of the given type. Fixing
754 it along the same lines as the others (cf. work done around
755 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
756 entirely straightforward.
757 c. All of these functions will silently accept a type of the form
759 whether or not the return value is of this type. This is
760 probably permitted by ANSI (see "Exceptional Situations" under
761 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
762 know about this escape clause, so code of the form
763 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
764 can erroneously return T.
767 SBCL 0.6.12.43 fails to compile
770 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
771 (flet ((foo (&key (x :vx x-p)) (list x x-p)))
774 or a more simple example:
777 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
778 (lambda (x) (declare (fixnum x)) (if (< x 0) 0 (1- x))))
780 215: ":TEST-NOT handling by functions"
781 a. FIND and POSITION currently signal errors when given non-NIL for
782 both their :TEST and (deprecated) :TEST-NOT arguments, but by
783 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
784 means that the effect is "unpredictable but harmless". It's not
785 clear what that actually means; it may preclude conforming
786 implementations from signalling errors.
787 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
788 when conflict occurs. As a quality of implementation issue, it
789 might be preferable to treat :TEST and :TEST-NOT as being in some
790 sense the same &KEY, and effectively take the first test function in
792 c. Again, a quality of implementation issue: it would be good to issue a
793 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
794 WARNING for calls with both :TEST and :TEST-NOT; possibly this
795 latter should be WARNed about at execute-time too.
797 216: "debugger confused by frames with invalid number of arguments"
798 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
799 leaves the system confused, enough so that (QUIT) no longer works.
800 It's as though the process of working with the uninitialized slot in
801 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
802 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
804 This is probably the same bug as 162
806 217: "Bad type operations with FUNCTION types"
809 * (values-type-union (specifier-type '(function (base-char)))
810 (specifier-type '(function (integer))))
812 #<FUN-TYPE (FUNCTION (BASE-CHAR) *)>
814 It causes insertion of wrong type assertions into generated
818 (let ((f (etypecase x
819 (character #'write-char)
820 (integer #'write-byte))))
823 (character (write-char x s))
824 (integer (write-byte x s)))))
826 Then (FOO #\1 *STANDARD-OUTPUT*) signals type error.
828 (In 0.7.9.1 the result type is (FUNCTION * *), so Python does not
829 produce invalid code, but type checking is not accurate.)
831 233: bugs in constraint propagation
833 (declaim (optimize (speed 2) (safety 3)))
835 (if (typep (prog1 x (setq x y)) 'double-float)
838 (foo 1d0 5) => segmentation violation
840 235: "type system and inline expansion"
842 (declaim (ftype (function (cons) number) acc))
843 (declaim (inline acc))
845 (the number (car c)))
848 (values (locally (declare (optimize (safety 0)))
850 (locally (declare (optimize (safety 3)))
853 (foo '(nil) '(t)) => NIL, T.
855 237: "Environment arguments to type functions"
856 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
857 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
858 argument, but they ignore it completely. This is almost
859 certainly not correct.
860 b. Also, the compiler's optimizers for TYPEP have not been informed
861 about the new argument; consequently, they will not transform
862 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
863 just as optimizeable as (TYPEP 1 'INTEGER).
865 238: "REPL compiler overenthusiasm for CLOS code"
867 * (defclass foo () ())
868 * (defmethod bar ((x foo) (foo foo)) (call-next-method))
869 causes approximately 100 lines of code deletion notes. Some
870 discussion on this issue happened under the title 'Three "interesting"
871 bugs in PCL', resulting in a fix for this oververbosity from the
872 compiler proper; however, the problem persists in the interactor
873 because the notion of original source is not preserved: for the
874 compiler, the original source of the above expression is (DEFMETHOD
875 BAR ((X FOO) (FOO FOO)) (CALL-NEXT-METHOD)), while by the time the
876 compiler gets its hands on the code needing compilation from the REPL,
877 it has been macroexpanded several times.
879 A symptom of the same underlying problem, reported by Tony Martinez:
881 (with-input-from-string (*query-io* " no")
883 (simple-type-error () 'error))
885 ; (SB-KERNEL:FLOAT-WAIT)
887 ; note: deleting unreachable code
888 ; compilation unit finished
891 241: "DEFCLASS mysteriously remembers uninterned accessor names."
892 (from tonyms on #lisp IRC 2003-02-25)
893 In sbcl-0.7.12.55, typing
894 (defclass foo () ((bar :accessor foo-bar)))
897 (defclass foo () ((bar :accessor foo-bar)))
898 gives the error message
899 "#:FOO-BAR already names an ordinary function or a macro."
900 So it's somehow checking the uninterned old accessor name instead
901 of the new requested accessor name, which seems broken to me (WHN).
903 242: "WRITE-SEQUENCE suboptimality"
904 (observed from clx performance)
905 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
906 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
907 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
908 rather than writing the sequence in one go, leading to severe
909 performance degradation.
911 243: "STYLE-WARNING overenthusiasm for unused variables"
912 (observed from clx compilation)
913 In sbcl-0.7.14, in the presence of the macros
914 (DEFMACRO FOO (X) `(BAR ,X))
915 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
916 somewhat surprising style warnings are emitted for
917 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
919 ; (LAMBDA (Y) (FOO Y))
921 ; caught STYLE-WARNING:
922 ; The variable Y is defined but never used.
924 245: bugs in disassembler
925 a. On X86 an immediate operand for IMUL is printed incorrectly.
926 b. On X86 operand size prefix is not recognized.
928 248: "reporting errors in type specifier syntax"
929 (TYPEP 1 '(SYMBOL NIL)) says something about "unknown type
933 (defun foo (&key (a :x))
937 does not cause a warning. (BTW: old SBCL issued a warning, but for a
938 function, which was never called!)
941 Compiler does not emit warnings for
943 a. (lambda () (svref (make-array 8 :adjustable t) 1))
946 (list (let ((y (the real x)))
947 (unless (floatp y) (error ""))
952 (declare (optimize (debug 0)))
953 (declare (type vector x))
954 (list (fill-pointer x)
958 Complex array type does not have corresponding type specifier.
960 This is a problem because the compiler emits optimization notes when
961 you use a non-simple array, and without a type specifier for hairy
962 array types, there's no good way to tell it you're doing it
963 intentionally so that it should shut up and just compile the code.
965 Another problem is confusing error message "asserted type ARRAY
966 conflicts with derived type (VALUES SIMPLE-VECTOR &OPTIONAL)" during
967 compiling (LAMBDA (V) (VALUES (SVREF V 0) (VECTOR-POP V))).
969 The last problem is that when type assertions are converted to type
970 checks, types are represented with type specifiers, so we could lose
971 complex attribute. (Now this is probably not important, because
972 currently checks for complex arrays seem to be performed by
976 (compile nil '(lambda () (aref (make-array 0) 0))) compiles without
977 warning. Analogous cases with the index and length being equal and
978 greater than 0 are warned for; the problem here seems to be that the
979 type required for an array reference of this type is (INTEGER 0 (0))
980 which is canonicalized to NIL.
985 (t1 (specifier-type s)))
986 (eval `(defstruct ,s))
987 (type= t1 (specifier-type s)))
992 b. The same for CSUBTYPEP.
995 * (let () (list (the (values &optional fixnum) (eval '(values)))))
996 debugger invoked on condition of type TYPE-ERROR:
997 The value NIL is not of type FIXNUM.
999 262: "yet another bug in inline expansion of local functions"
1003 (declare (integer x y))
1006 (declare (integer u))
1007 (if (> (1+ (the unsigned-byte u)) 0)
1009 (return (+ 38 (cos (/ u 78)))))))
1010 (declare (inline xyz))
1012 (* (funcall (eval #'xyz) x)
1014 (funcall (if (> x 5) #'xyz #'identity)
1019 Urgh... It's time to write IR1-copier.
1022 SB-EXT:RUN-PROGRAM is currently non-functional on Linux/PPC;
1023 attempting to use it leads to segmentation violations. This is
1024 probably because of a bogus implementation of
1025 os_restore_fp_control().
1028 David Lichteblau provided (sbcl-devel 2003-06-01) a patch to fix
1029 behaviour of streams with element-type (SIGNED-BYTE 8). The patch
1030 looks reasonable, if not obviously correct; however, it caused the
1031 PPC/Linux port to segfault during warm-init while loading
1032 src/pcl/std-class.fasl. A workaround patch was made, but it would
1033 be nice to understand why the first patch caused problems, and to
1034 fix the cause if possible.
1036 268: "wrong free declaration scope"
1037 The following code must signal type error:
1039 (locally (declare (optimize (safety 3)))
1040 (flet ((foo (x &optional (y (car x)))
1041 (declare (optimize (safety 0)))
1043 (funcall (eval #'foo) 1)))
1046 SCALE-FLOAT should accept any integer for its second argument.
1049 In the following function constraint propagator optimizes nothing:
1052 (declare (integer x))
1053 (declare (optimize speed))
1061 Compilation of the following two forms causes "X is unbound" error:
1063 (symbol-macrolet ((x pi))
1064 (macrolet ((foo (y) (+ x y)))
1065 (declaim (inline bar))
1071 (See (COERCE (CDR X) 'FUNCTION) in IR1-CONVERT-INLINE-LAMBDA.)
1074 CLHS says that type declaration of a symbol macro should not affect
1075 its expansion, but in SBCL it does. (If you like magic and want to
1076 fix it, don't forget to change all uses of MACROEXPAND to
1080 The following code (taken from CLOCC) takes a lot of time to compile:
1083 (declare (type (integer 0 #.large-constant) n))
1086 (fixed in 0.8.2.51, but a test case would be good)
1089 (defmethod fee ((x fixnum))
1092 (fee 1) => type error
1099 (declare (optimize speed))
1100 (loop for i of-type (integer 0) from 0 by 2 below 10
1103 uses generic arithmetic.
1105 b. For the example above, the compiler does not issue a note.
1106 (fixed in 0.8.3.6, but a test case would be good)
1108 279: type propagation error -- correctly inferred type goes astray?
1109 In sbcl-0.8.3 and sbcl-0.8.1.47, the warning
1110 The binding of ABS-FOO is a (VALUES (INTEGER 0 0)
1111 &OPTIONAL), not a (INTEGER 1 536870911)
1112 is emitted when compiling this file:
1113 (declaim (ftype (function ((integer 0 #.most-positive-fixnum))
1114 (integer #.most-negative-fixnum 0))
1119 (let* (;; Uncomment this for a type mismatch warning indicating
1120 ;; that the type of (FOO X) is correctly understood.
1121 #+nil (fs-foo (float-sign (foo x)))
1122 ;; Uncomment this for a type mismatch warning
1123 ;; indicating that the type of (ABS (FOO X)) is
1124 ;; correctly understood.
1125 #+nil (fs-abs-foo (float-sign (abs (foo x))))
1126 ;; something wrong with this one though
1127 (abs-foo (abs (foo x))))
1128 (declare (type (integer 1 100) abs-foo))
1133 280: bogus WARNING about duplicate function definition
1134 In sbcl-0.8.3 and sbcl-0.8.1.47, if BS.MIN is defined inline,
1136 (declaim (inline bs.min))
1137 (defun bs.min (bases) nil)
1138 before compiling the file below, the compiler warns
1139 Duplicate definition for BS.MIN found in one static
1140 unit (usually a file).
1142 (declaim (special *minus* *plus* *stagnant*))
1143 (defun b.*.min (&optional (x () xp) (y () yp) &rest rest)
1145 (define-compiler-macro b.*.min (&rest rest)
1147 (defun afish-d-rbd (pd)
1149 (b.*.min (foo-d-rbd *stagnant*))
1150 (multiple-value-bind (reduce-fn initial-value)
1152 (list (values #'bs.min 0))
1153 (vector (values #'bs.min *plus*)))
1154 (let ((cv-ks (cv (kpd.ks pd))))
1155 (funcall reduce-fn d-rbds)))))
1156 (defun bfish-d-rbd (pd)
1158 (b.*.min (foo-d-rbd *stagnant*))
1159 (multiple-value-bind (reduce-fn initial-value)
1161 (list (values #'bs.min *minus*))
1162 (vector (values #'bs.min 0)))
1163 (let ((cv-ks (cv (kpd.ks pd))))
1164 (funcall reduce-fn d-rbds)))))
1166 281: COMPUTE-EFFECTIVE-METHOD error signalling.
1167 (slightly obscured by a non-0 default value for
1168 SB-PCL::*MAX-EMF-PRECOMPUTE-METHODS*)
1169 It would be natural for COMPUTE-EFFECTIVE-METHOD to signal errors
1170 when it finds a method with invalid qualifiers. However, it
1171 shouldn't signal errors when any such methods are not applicable to
1172 the particular call being evaluated, and certainly it shouldn't when
1173 simply precomputing effective methods that may never be called.
1174 (setf sb-pcl::*max-emf-precompute-methods* 0)
1176 (:method-combination +)
1177 (:method ((x symbol)) 1)
1178 (:method + ((x number)) x))
1179 (foo 1) -> ERROR, but should simply return 1
1181 The issue seems to be that construction of a discriminating function
1182 calls COMPUTE-EFFECTIVE-METHOD with methods that are not all applicable.
1184 282: "type checking in full calls"
1185 In current (0.8.3.6) implementation a CAST in a full call argument
1186 is not checked; but the continuation between the CAST and the
1187 combination has the "checked" type and CAST performs unsafe
1188 coercion; this may lead to errors: if FOO is declared to take a
1189 FIXNUM, this code will produce garbage on a machine with 30-bit
1192 (foo (aref (the (array (unsigned-byte 32)) x)))
1194 283: Thread safety: libc functions
1195 There are places that we call unsafe-for-threading libc functions
1196 that we should find alternatives for, or put locks around. Known or
1197 strongly suspected problems, as of 0.8.3.10: please update this
1198 bug instead of creating new ones
1200 localtime() - called for timezone calculations in code/time.lisp
1202 284: Thread safety: special variables
1203 There are lots of special variables in SBCL, and I feel sure that at
1204 least some of them are indicative of potentially thread-unsafe
1205 parts of the system. See doc/internals/notes/threading-specials
1208 In SBCL 0.8.3.1x on a powerpc running Linux (dunno if Darwin is
1209 similarly affected):
1210 * (dotimes (i 100) (random 1663553320000000))
1213 * (dotimes (i 100) (random 1663553340000000))
1216 * (dotimes (i 100) (random 1663553350000000))
1218 debugger invoked on condition of type TYPE-ERROR:
1219 The value -30653269094906
1221 (OR (SINGLE-FLOAT 0.0) (DOUBLE-FLOAT 0.0d0) (RATIONAL 0)).
1223 and, weirdly, the frame is:
1224 ("hairy arg processor for top level local call RANDOM"
1227 :STATE #(0 2567483615 188 1503590015 2333049409 322761517 ...)))
1229 (the type error doesn't seem to be terribly deterministic in when it
1230 occurs. Bigger numbers seem better able to trigger the error)
1232 286: "recursive known functions"
1233 Self-call recognition conflicts with known function
1234 recognition. Currently cross compiler and target COMPILE do not
1235 recognize recursion, and in target compiler it can be disabled. We
1236 can always disable it for known functions with RECURSIVE attribute,
1237 but there remains a possibility of a function with a
1238 (tail)-recursive simplification pass and transforms/VOPs for base
1241 287: PPC/Linux miscompilation or corruption in first GC
1242 When the runtime is compiled with -O3 on certain PPC/Linux machines, a
1243 segmentation fault is reported at the point of first triggered GC,
1244 during the compilation of DEFSTRUCT WRAPPER. As a temporary workaround,
1245 the runtime is no longer compiled with -O3 on PPC/Linux, but it is likely
1246 that this merely obscures, not solves, the underlying problem; as and when
1247 underlying problems are fixed, it would be worth trying again to provoke