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).
35 KNOWN BUGS OF NO SPECIAL CLASS:
38 DEFSTRUCT almost certainly should overwrite the old LAYOUT information
39 instead of just punting when a contradictory structure definition
40 is loaded. As it is, if you redefine DEFSTRUCTs in a way which
41 changes their layout, you probably have to rebuild your entire
42 program, even if you know or guess enough about the internals of
43 SBCL to wager that this (undefined in ANSI) operation would be safe.
45 3: "type checking of structure slots"
47 ANSI specifies that a type mismatch in a structure slot
48 initialization value should not cause a warning.
50 This one might not be fixed for a while because while we're big
51 believers in ANSI compatibility and all, (1) there's no obvious
52 simple way to do it (short of disabling all warnings for type
53 mismatches everywhere), and (2) there's a good portable
54 workaround, and (3) by their own reasoning, it looks as though
55 ANSI may have gotten it wrong. ANSI justifies this specification
57 The restriction against issuing a warning for type mismatches
58 between a slot-initform and the corresponding slot's :TYPE
59 option is necessary because a slot-initform must be specified
60 in order to specify slot options; in some cases, no suitable
62 However, in SBCL (as in CMU CL or, for that matter, any compiler
63 which really understands Common Lisp types) a suitable default
64 does exist, in all cases, because the compiler understands the
65 concept of functions which never return (i.e. has return type NIL).
66 Thus, as a portable workaround, you can use a call to some
67 known-never-to-return function as the default. E.g.
69 (BAR (ERROR "missing :BAR argument")
70 :TYPE SOME-TYPE-TOO-HAIRY-TO-CONSTRUCT-AN-INSTANCE-OF))
72 (DECLAIM (FTYPE (FUNCTION () NIL) MISSING-ARG))
73 (DEFUN REQUIRED-ARG () ; workaround for SBCL non-ANSI slot init typing
74 (ERROR "missing required argument"))
76 (BAR (REQUIRED-ARG) :TYPE TRICKY-TYPE-OF-SOME-SORT)
77 (BLETCH (REQUIRED-ARG) :TYPE TRICKY-TYPE-OF-SOME-SORT)
78 (N-REFS-SO-FAR 0 :TYPE (INTEGER 0)))
79 Such code should compile without complaint and work correctly either
80 on SBCL or on any other completely compliant Common Lisp system.
82 b: &AUX argument in a boa-constructor without a default value means
83 "do not initilize this slot" and does not cause type error. But
84 an error may be signalled at read time and it would be good if
90 The "compiling top-level form:" output ought to be condensed.
91 Perhaps any number of such consecutive lines ought to turn into a
92 single "compiling top-level forms:" line.
95 It would be nice if the
97 (during macroexpansion)
98 said what macroexpansion was at fault, e.g.
100 (during macroexpansion of IN-PACKAGE,
101 during macroexpansion of DEFFOO)
104 (I *think* this is a bug. It certainly seems like strange behavior. But
105 the ANSI spec is scary, dark, and deep.. -- WHN)
106 (FORMAT NIL "~,1G" 1.4) => "1. "
107 (FORMAT NIL "~3,1G" 1.4) => "1. "
110 Sometimes (SB-EXT:QUIT) fails with
111 Argh! maximum interrupt nesting depth (4096) exceeded, exiting
112 Process inferior-lisp exited abnormally with code 1
113 I haven't noticed a repeatable case of this yet.
116 The printer doesn't report closures very well. This is true in
120 #<Closure Over Function "DEFUN STRUCTURE-SLOT-ACCESSOR" {134D1A1}>
121 It would be nice to make closures have a settable name slot,
122 and make things like DEFSTRUCT and FLET, which create closures,
123 set helpful values into this slot.
126 And as long as we're wishing, it would be awfully nice if INSPECT could
127 also report on closures, telling about the values of the bound variables.
130 The compiler assumes that any time a function of declared FTYPE
131 doesn't signal an error, its arguments were of the declared type.
132 E.g. compiling and loading
133 (DECLAIM (OPTIMIZE (SAFETY 3)))
134 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
136 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
138 (COND ((> (FACTORIAL X) 1.0E6)
139 (FORMAT T "too big~%"))
141 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
143 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
146 will cause the INTEGERP case to be selected, giving bogus output a la
148 This violates the "declarations are assertions" principle.
149 According to the ANSI spec, in the section "System Class FUNCTION",
150 this is a case of "lying to the compiler", but the lying is done
151 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
152 not by the unexpectedly general definition of FACTORIAL. In any case,
153 "declarations are assertions" means that lying to the compiler should
154 cause an error to be signalled, and should not cause a bogus
155 result to be returned. Thus, the compiler should not assume
156 that arbitrary functions check their argument types. (It might
157 make sense to add another flag (CHECKED?) to DEFKNOWN to
158 identify functions which *do* check their argument types.)
159 (Also, verify that the compiler handles declared function
160 return types as assertions.)
163 The definitions of SIGCONTEXT-FLOAT-REGISTER and
164 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
165 supported on FreeBSD because the floating point state is not saved,
166 but at least as of FreeBSD 4.0, the floating point state *is* saved,
167 so they could be supported after all. Very likely
168 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
171 a slew of floating-point-related errors reported by Peter Van Eynde
173 b: SBCL's value for LEAST-POSITIVE-SHORT-FLOAT on the x86 is
174 bogus, and should probably be 1.4012985e-45. In SBCL,
175 (/ LEAST-POSITIVE-SHORT-FLOAT 2) returns a number smaller
176 than LEAST-POSITIVE-SHORT-FLOAT. Similar problems
177 exist for LEAST-NEGATIVE-SHORT-FLOAT, LEAST-POSITIVE-LONG-FLOAT,
178 and LEAST-NEGATIVE-LONG-FLOAT.
179 c: Many expressions generate floating infinity on x86/Linux:
184 PVE's regression tests want them to raise errors. sbcl-0.7.0.5
185 on x86/Linux generates the infinities instead. That might or
186 might not be conforming behavior, but it's also inconsistent,
187 which is almost certainly wrong. (Inconsistency: (/ 1 0.0)
188 should give the same result as (/ 1.0 0.0), but instead (/ 1 0.0)
189 generates SINGLE-FLOAT-POSITIVE-INFINITY and (/ 1.0 0.0)
191 d: (in section12.erg) various forms a la
192 (FLOAT 1 DOUBLE-FLOAT-EPSILON)
193 don't give the right behavior.
196 type safety errors reported by Peter Van Eynde July 25, 2000:
197 k: READ-BYTE is supposed to signal TYPE-ERROR when its argument is
198 not a binary input stream, but instead cheerfully reads from
199 character streams, e.g. (MAKE-STRING-INPUT-STREAM "abc").
202 The debugger LIST-LOCATIONS command doesn't work properly.
203 (How should it work properly?)
206 Compiling and loading
207 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
209 then requesting a BACKTRACE at the debugger prompt gives no information
210 about where in the user program the problem occurred.
213 Using the pretty-printer from the command prompt gives funny
214 results, apparently because the pretty-printer doesn't know
215 about user's command input, including the user's carriage return
216 that the user, and therefore the pretty-printer thinks that
217 the new output block should start indented 2 or more characters
218 rightward of the correct location.
221 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
222 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
223 crashes SBCL. In general tracing anything which is used in the
224 implementation of TRACE is likely to have the same problem.
227 As reported by Martin Atzmueller on sbcl-devel 26 Dec 2000,
228 ANSI says that WITH-OUTPUT-TO-STRING should have a keyword
229 :ELEMENT-TYPE, but in sbcl-0.6.9 this is not defined for
230 WITH-OUTPUT-TO-STRING.
233 ANSI says in one place that type declarations can be abbreviated even
234 when the type name is not a symbol, e.g.
235 (DECLAIM ((VECTOR T) *FOOVECTOR*))
236 SBCL doesn't support this. But ANSI says in another place that this
237 isn't allowed. So it's not clear this is a bug after all. (See the
238 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
242 as pointed out by Dan Barlow on sbcl-devel 2000-07-02:
243 The PICK-TEMPORARY-FILE-NAME utility used by LOAD-FOREIGN uses
244 an easily guessable temporary filename in a way which might open
245 applications using LOAD-FOREIGN to hijacking by malicious users
246 on the same machine. Incantations for doing this safely are
247 floating around the net in various "how to write secure programs
248 despite Unix" documents, and it would be good to (1) fix this in
249 LOAD-FOREIGN, and (2) hunt for any other code which uses temporary
250 files and make it share the same new safe logic.
252 (partially alleviated in sbcl-0.7.9.32 by a fix by Matthew Danish to
253 make the temporary filename less easily guessable)
256 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
257 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
258 the mean of the distribution can be systematically O(0.1%) wrong.
259 Just increasing R-I-E-B is probably not a good solution, since
260 it would decrease efficiency more than is probably necessary. Perhaps
261 using some sort of accept/reject method would be better.
264 Internally the compiler sometimes evaluates
265 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
266 (I stumbled across this when I added an
267 (assert (not (eq type1 *wild-type*)))
268 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
269 in a type context should probably be translated to T, and so it's
270 probably wrong to ask whether it's equal to the T type and then (using
271 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
272 (I haven't tried to investigate this bug enough to guess whether
273 there might be any user-level symptoms.)
275 In fact, the type system is likely to depend on this inequality not
276 holding... * is not equivalent to T in many cases, such as
277 (VECTOR *) /= (VECTOR T).
280 The facility for dumping a running Lisp image to disk gets confused
281 when run without the PURIFY option, and creates an unnecessarily large
282 core file (apparently representing memory usage up to the previous
283 high-water mark). Moreover, when the file is loaded, it confuses the
284 GC, so that thereafter memory usage can never be reduced below that
288 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
289 CL), out-of-line structure slot setters are horribly inefficient
290 whenever the type of the slot is declared, because out-of-line
291 structure slot setters are implemented as closures to save space,
292 so the compiler doesn't compile the type test into code, but
293 instead just saves the type in a lexical closure and interprets it
295 To exercise the problem, compile and load
296 (cl:in-package :cl-user)
298 (bar (error "missing") :type bar))
301 (loop (setf (foo-bar *foo*) x)))
303 (defvar *bar* (make-bar))
304 (defvar *foo* (make-foo :bar *bar*))
305 (defvar *setf-foo-bar* #'(setf foo-bar))
307 (loop (funcall *setf-foo-bar* x *foo*)))
308 then run (WASTREL1 *BAR*) or (WASTREL2 *BAR*), hit Ctrl-C, and
309 use BACKTRACE, to see it's spending all essentially all its time
310 in %TYPEP and VALUES-SPECIFIER-TYPE and so forth.
311 One possible solution would be simply to give up on
312 representing structure slot accessors as functions, and represent
313 them as macroexpansions instead. This can be inconvenient for users,
314 but it's not clear that it's worse than trying to help by expanding
315 into a horribly inefficient implementation.
316 As a workaround for the problem, #'(SETF FOO) expressions
317 can be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
318 (defmacro efficient-setf-function (place-function-name)
319 (or #+sbcl (and (sb-int:info :function :accessor-for place-function-name)
320 ;; a workaround for the problem, encouraging the
321 ;; inline expansion of the structure accessor, so
322 ;; that the compiler can optimize its type test
323 (let ((new-value (gensym "NEW-VALUE-"))
324 (structure-value (gensym "STRUCTURE-VALUE-")))
325 `(lambda (,new-value ,structure-value)
326 (setf (,place-function-name ,structure-value)
328 ;; no problem, can just use the ordinary expansion
329 `(function (setf ,place-function-name))))
332 There's apparently a bug in CEILING optimization which caused
333 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
334 applied the patches to SBCL and they didn't seem to cause problems
335 (as reported sbcl-devel 2001-05-04). However, since the patches
336 modify nontrivial code which was apparently written incorrectly
337 the first time around, until regression tests are written I'm not
338 comfortable merging the patches in the CVS version of SBCL.
341 (TIME (ROOM T)) reports more than 200 Mbytes consed even for
342 a clean, just-started SBCL system. And it seems to be right:
343 (ROOM T) can bring a small computer to its knees for a *long*
344 time trying to GC afterwards. Surely there's some more economical
345 way to implement (ROOM T).
348 When the compiler inline expands functions, it may be that different
349 kinds of return values are generated from different code branches.
350 E.g. an inline expansion of POSITION generates integer results
351 from one branch, and NIL results from another. When that inline
352 expansion is used in a context where only one of those results
355 (aref *a1* (position x *a2*)))
356 and the compiler can't prove that the unacceptable branch is
357 never taken, then bogus type mismatch warnings can be generated.
358 If you need to suppress the type mismatch warnings, you can
359 suppress the inline expansion,
361 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
362 (aref *a1* (position x *a2*)))
363 or, sometimes, suppress them by declaring the result to be of an
366 (aref *a1* (the integer (position x *a2*))))
368 This is not a new compiler problem in 0.7.0, but the new compiler
369 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
370 more conspicuous. If you don't need performance from these functions,
371 and the bogus warnings are a nuisance for you, you can return to
372 your pre-0.7.0 state of grace with
373 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
376 as reported by Eric Marsden on cmucl-imp@cons.org 2001-08-14:
377 (= (FLOAT 1 DOUBLE-FLOAT-EPSILON)
378 (+ (FLOAT 1 DOUBLE-FLOAT-EPSILON) DOUBLE-FLOAT-EPSILON)) => T
379 when of course it should be NIL. (He says it only fails for X86,
380 not SPARC; dunno about Alpha.)
382 Also, "the same problem exists for LONG-FLOAT-EPSILON,
383 DOUBLE-FLOAT-NEGATIVE-EPSILON, LONG-FLOAT-NEGATIVE-EPSILON (though
384 for the -negative- the + is replaced by a - in the test)."
386 Raymond Toy comments that this is tricky on the X86 since its FPU
387 uses 80-bit precision internally.
390 Even in sbcl-0.pre7.x, which is supposed to be free of the old
391 non-ANSI behavior of treating the function return type inferred
392 from the current function definition as a declaration of the
393 return type from any function of that name, the return type of NIL
394 is attached to FOO in 120a above, and used to optimize code which
398 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
399 the entire lexical environment at the point of MACROLET available
400 in the bodies of the macroexpander functions. In particular, it
401 allows the function bodies (which run at compile time) to try to
402 access lexical variables (which are only defined at runtime).
403 It doesn't even issue a warning, which is bad.
405 The SBCL behavior arguably conforms to the ANSI spec (since the
406 spec says that the behavior is undefined, ergo anything conforms).
407 However, it would be better to issue a compile-time error.
408 Unfortunately I (WHN) don't see any simple way to detect this
409 condition in order to issue such an error, so for the meantime
410 SBCL just does this weird broken "conforming" thing.
412 The ANSI standard says, in the definition of the special operator
414 The macro-expansion functions defined by MACROLET are defined
415 in the lexical environment in which the MACROLET form appears.
416 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
417 the local macro definitions in a MACROLET, but the consequences
418 are undefined if the local macro definitions reference any
419 local variable or function bindings that are visible in that
421 Then it seems to contradict itself by giving the example
423 (macrolet ((fudge (z)
424 ;The parameters x and flag are not accessible
425 ; at this point; a reference to flag would be to
426 ; the global variable of that name.
427 ` (if flag (* ,z ,z) ,z)))
428 ;The parameters x and flag are accessible here.
432 The comment "a reference to flag would be to the global variable
433 of the same name" sounds like good behavior for the system to have.
434 but actual specification quoted above says that the actual behavior
437 (Since 0.7.8.23 macroexpanders are defined in a restricted version
438 of the lexical environment, containing no lexical variables and
439 functions, which seems to conform to ANSI and CLtL2, but signalling
440 a STYLE-WARNING for references to variables similar to locals might
444 (as reported by Gabe Garza on cmucl-help 2001-09-21)
446 (defun test-pred (x y)
450 (func (lambda () x)))
451 (print (eq func func))
452 (print (test-pred func func))
453 (delete func (list func))))
454 Now calling (TEST-CASE) gives output
457 (#<FUNCTION {500A9EF9}>)
458 Evidently Python thinks of the lambda as a code transformation so
459 much that it forgets that it's also an object.
462 Ideally, uninterning a symbol would allow it, and its associated
463 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
464 at least as of sbcl-0.7.0, this isn't the case. Information about
465 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
466 essentially in ordinary (non-weak) hash tables keyed by symbols.
467 Thus, once a system has an entry in this system, it tends to live
468 forever, even when it is uninterned and all other references to it
471 141: "pretty printing and backquote"
474 ``(FOO SB-IMPL::BACKQ-COMMA-AT S)
477 * (write '`(, .ala.) :readably t :pretty t)
480 (note the space between the comma and the point)
483 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
484 prominent SourceForge web/db bug tracking system, which is
485 unfortunately not a reliable way to get a timely response from
486 the SBCL maintainers)
487 In the course of trying to build a test case for an
488 application error, I encountered this behavior:
489 If you start up sbcl, and then lay on CTRL-C for a
490 minute or two, the lisp process will eventually say:
491 %PRIMITIVE HALT called; the party is over.
492 and throw you into the monitor. If I start up lisp,
493 attach to the process with strace, and then do the same
494 (abusive) thing, I get instead:
495 access failure in heap page not marked as write-protected
496 and the monitor again. I don't know enough to have the
497 faintest idea of what is going on here.
498 This is with sbcl 6.12, uname -a reports:
499 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
500 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
501 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
502 it took more than two minutes (but less than five) for me.
505 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
506 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
507 COMPLEX implementation didn't deal with this, and hasn't been
508 upgraded to do so. (This doesn't seem to be a high priority
509 conformance problem, since seems hard to construct useful code
513 Floating point errors are reported poorly. E.g. on x86 OpenBSD
516 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
517 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
518 No traps are enabled? How can this be?
519 It should be possible to be much more specific (overflow, division
520 by zero, etc.) and of course the "How can this be?" should be fixable.
522 See also bugs #45.c and #183
525 (reported by Robert E. Brown 2002-04-16)
526 When a function is called with too few arguments, causing the
527 debugger to be entered, the uninitialized slots in the bad call frame
528 seem to cause GCish problems, being interpreted as tagged data even
529 though they're not. In particular, executing ROOM in the
530 debugger at that point causes AVER failures:
533 * (lisp-implementation-version)
539 failed AVER: "(SAP= CURRENT END)"
540 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
541 isn't too surprising since there are many differences in stack
542 implementation and GC conservatism between the X86 and other ports.)
545 In sbcl-0.7.3.11, compiling the (illegal) code
546 (in-package :cl-user)
547 (defmethod prove ((uustk uustk))
550 gives the (not terribly clear) error message
552 ; (during macroexpansion of (DEFMETHOD PROVE ...))
553 ; can't get template for (FROB NIL NIL)
554 The problem seems to be that the code walker used by the DEFMETHOD
555 macro is unhappy with the illegal syntax in the method body, and
556 is giving an unclear error message.
559 The compiler sometimes tries to constant-fold expressions before
560 it checks to see whether they can be reached. This can lead to
561 bogus warnings about errors in the constant folding, e.g. in code
564 (WRITE-STRING (> X 0) "+" "0"))
565 compiled in a context where the compiler can prove that X is NIL,
566 and the compiler complains that (> X 0) causes a type error because
567 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
568 caused a full WARNING, which made the bug really annoying because then
569 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
570 code. Since then the warning has been downgraded to STYLE-WARNING,
571 so it's still a bug but at least it's a little less annoying.
573 183: "IEEE floating point issues"
574 Even where floating point handling is being dealt with relatively
575 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
576 accrued-exceptions and current-exceptions part of the fp control
577 word don't seem to bear much relation to reality. E.g. on
581 debugger invoked on condition of type DIVISION-BY-ZERO:
582 arithmetic error DIVISION-BY-ZERO signalled
583 0] (sb-vm::get-floating-point-modes)
585 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
586 :ROUNDING-MODE :NEAREST
587 :CURRENT-EXCEPTIONS NIL
588 :ACCRUED-EXCEPTIONS (:INEXACT)
591 * (sb-vm::get-floating-point-modes)
592 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
593 :ROUNDING-MODE :NEAREST
594 :CURRENT-EXCEPTIONS (:INEXACT)
595 :ACCRUED-EXCEPTIONS (:INEXACT)
598 188: "compiler performance fiasco involving type inference and UNION-TYPE"
602 (declare (optimize (safety 3)))
603 (declare (optimize (compilation-speed 2)))
604 (declare (optimize (speed 1) (debug 1) (space 1)))
606 (declare (type (integer 0) start))
607 (print (incf start 22))
608 (print (incf start 26))
609 (print (incf start 28)))
611 (declare (type (integer 0) start))
612 (print (incf start 22))
613 (print (incf start 26)))
615 (declare (type (integer 0) start))
616 (print (incf start 22))
617 (print (incf start 26))))))
619 This example could be solved with clever enough constraint
620 propagation or with SSA, but consider
625 The careful type of X is {2k} :-(. Is it really important to be
626 able to work with unions of many intervals?
628 190: "PPC/Linux pipe? buffer? bug"
629 In sbcl-0.7.6, the run-program.test.sh test script sometimes hangs
630 on the PPC/Linux platform, waiting for a zombie env process. This
631 is a classic symptom of buffer filling and deadlock, but it seems
632 only sporadically reproducible.
634 191: "Miscellaneous PCL deficiencies"
635 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
636 a. DEFCLASS does not inform the compiler about generated
637 functions. Compiling a file with
641 (WITH-SLOTS (A-CLASS-X) A
643 results in a STYLE-WARNING:
645 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
647 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
648 Mai points out that the declamation of functions is in fact
649 incorrect in some cases (most notably for structure
650 classes). This means that at present erroneous attempts to use
651 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
652 won't get the corresponding STYLE-WARNING.
653 c. the examples in CLHS 7.6.5.1 (regarding generic function lambda
654 lists and &KEY arguments) do not signal errors when they should.
656 201: "Incautious type inference from compound types"
657 a. (reported by APD sbcl-devel 2002-09-17)
659 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
661 (FORMAT NIL "~S IS ~S, Y = ~S"
668 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
672 (declare (type (array * (4 4)) x))
674 (setq x (make-array '(4 4)))
675 (adjust-array y '(3 5))
676 (= (array-dimension y 0) (eval `(array-dimension ,y 0)))))
678 * (foo (make-array '(4 4) :adjustable t))
681 205: "environment issues in cross compiler"
682 (These bugs have no impact on user code, but should be fixed or
684 a. Macroexpanders introduced with MACROLET are defined in the null
686 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
687 the null lexical environment.
688 c. The cross-compiler cannot inline functions defined in a non-null
691 206: ":SB-FLUID feature broken"
692 (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
693 Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
696 207: "poorly distributed SXHASH results for compound data"
697 SBCL's SXHASH could probably try a little harder. ANSI: "the
698 intent is that an implementation should make a good-faith
699 effort to produce hash-codes that are well distributed
700 within the range of non-negative fixnums". But
701 (let ((hits (make-hash-table)))
704 (let* ((ij (cons i j))
705 (newlist (push ij (gethash (sxhash ij) hits))))
707 (format t "~&collision: ~S~%" newlist))))))
708 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
709 would be an obvious way of fix. Maybe it would be acceptably efficient
710 to redo MIX using a lookup into a 256-entry s-box containing
711 29-bit pseudorandom numbers?
713 211: "keywords processing"
714 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
715 number of keyword arguments.
718 (flet ((foo (&key y) (list y)))
719 (list (foo :y 1 :y 2)))
721 issues confusing message
726 ; caught STYLE-WARNING:
727 ; The variable #:G15 is defined but never used.
729 212: "Sequence functions and circular arguments"
730 COERCE, MERGE and CONCATENATE go into an infinite loop when given
731 circular arguments; it would be good for the user if they could be
732 given an error instead (ANSI 17.1.1 allows this behaviour on the part
733 of the implementation, as conforming code cannot give non-proper
734 sequences to these functions. MAP also has this problem (and
735 solution), though arguably the convenience of being able to do
736 (MAP 'LIST '+ FOO '#1=(1 . #1#))
737 might be classed as more important (though signalling an error when
738 all of the arguments are circular is probably desireable).
740 213: "Sequence functions and type checking"
741 a. MAKE-SEQUENCE, COERCE, MERGE and CONCATENATE cannot deal with
742 various complicated, though recognizeable, CONS types [e.g.
743 (CONS * (CONS * NULL))
744 which according to ANSI should be recognized] (and, in SAFETY 3
745 code, should return a list of LENGTH 2 or signal an error)
746 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
747 check that it will return a sequence of the given type. Fixing
748 it along the same lines as the others (cf. work done around
749 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
750 entirely straightforward.
751 c. All of these functions will silently accept a type of the form
753 whether or not the return value is of this type. This is
754 probably permitted by ANSI (see "Exceptional Situations" under
755 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
756 know about this escape clause, so code of the form
757 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
758 can erroneously return T.
761 SBCL 0.6.12.43 fails to compile
764 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
765 (flet ((foo (&key (x :vx x-p)) (list x x-p)))
768 or a more simple example:
771 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
772 (lambda (x) (declare (fixnum x)) (if (< x 0) 0 (1- x))))
774 215: ":TEST-NOT handling by functions"
775 a. FIND and POSITION currently signal errors when given non-NIL for
776 both their :TEST and (deprecated) :TEST-NOT arguments, but by
777 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
778 means that the effect is "unpredictable but harmless". It's not
779 clear what that actually means; it may preclude conforming
780 implementations from signalling errors.
781 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
782 when conflict occurs. As a quality of implementation issue, it
783 might be preferable to treat :TEST and :TEST-NOT as being in some
784 sense the same &KEY, and effectively take the first test function in
786 c. Again, a quality of implementation issue: it would be good to issue a
787 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
788 WARNING for calls with both :TEST and :TEST-NOT; possibly this
789 latter should be WARNed about at execute-time too.
791 216: "debugger confused by frames with invalid number of arguments"
792 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
793 leaves the system confused, enough so that (QUIT) no longer works.
794 It's as though the process of working with the uninitialized slot in
795 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
796 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
798 217: "Bad type operations with FUNCTION types"
801 * (values-type-union (specifier-type '(function (base-char)))
802 (specifier-type '(function (integer))))
804 #<FUN-TYPE (FUNCTION (BASE-CHAR) *)>
806 It causes insertion of wrong type assertions into generated
810 (let ((f (etypecase x
811 (character #'write-char)
812 (integer #'write-byte))))
815 (character (write-char x s))
816 (integer (write-byte x s)))))
818 Then (FOO #\1 *STANDARD-OUTPUT*) signals type error.
820 (In 0.7.9.1 the result type is (FUNCTION * *), so Python does not
821 produce invalid code, but type checking is not accurate.)
823 233: bugs in constraint propagation
826 (declare (optimize (speed 2) (safety 3)))
832 (quux y (+ y 2d0) (* y 3d0)))))
833 (foo 4) => segmentation violation
835 (see usage of CONTINUATION-ASSERTED-TYPE in USE-RESULT-CONSTRAINTS)
839 (declaim (optimize (speed 2) (safety 3)))
841 (if (typep (prog1 x (setq x y)) 'double-float)
844 (foo 1d0 5) => segmentation violation
846 235: "type system and inline expansion"
848 (declaim (ftype (function (cons) number) acc))
849 (declaim (inline acc))
851 (the number (car c)))
854 (values (locally (declare (optimize (safety 0)))
856 (locally (declare (optimize (safety 3)))
859 (foo '(nil) '(t)) => NIL, T.
861 237: "Environment arguments to type functions"
862 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
863 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
864 argument, but they ignore it completely. This is almost
865 certainly not correct.
866 b. Also, the compiler's optimizers for TYPEP have not been informed
867 about the new argument; consequently, they will not transform
868 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
869 just as optimizeable as (TYPEP 1 'INTEGER).
871 238: "REPL compiler overenthusiasm for CLOS code"
873 * (defclass foo () ())
874 * (defmethod bar ((x foo) (foo foo)) (call-next-method))
875 causes approximately 100 lines of code deletion notes. Some
876 discussion on this issue happened under the title 'Three "interesting"
877 bugs in PCL', resulting in a fix for this oververbosity from the
878 compiler proper; however, the problem persists in the interactor
879 because the notion of original source is not preserved: for the
880 compiler, the original source of the above expression is (DEFMETHOD
881 BAR ((X FOO) (FOO FOO)) (CALL-NEXT-METHOD)), while by the time the
882 compiler gets its hands on the code needing compilation from the REPL,
883 it has been macroexpanded several times.
885 A symptom of the same underlying problem, reported by Tony Martinez:
887 (with-input-from-string (*query-io* " no")
889 (simple-type-error () 'error))
891 ; (SB-KERNEL:FLOAT-WAIT)
893 ; note: deleting unreachable code
894 ; compilation unit finished
897 241: "DEFCLASS mysteriously remembers uninterned accessor names."
898 (from tonyms on #lisp IRC 2003-02-25)
899 In sbcl-0.7.12.55, typing
900 (defclass foo () ((bar :accessor foo-bar)))
903 (defclass foo () ((bar :accessor foo-bar)))
904 gives the error message
905 "#:FOO-BAR already names an ordinary function or a macro."
906 So it's somehow checking the uninterned old accessor name instead
907 of the new requested accessor name, which seems broken to me (WHN).
909 242: "WRITE-SEQUENCE suboptimality"
910 (observed from clx performance)
911 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
912 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
913 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
914 rather than writing the sequence in one go, leading to severe
915 performance degradation.
917 243: "STYLE-WARNING overenthusiasm for unused variables"
918 (observed from clx compilation)
919 In sbcl-0.7.14, in the presence of the macros
920 (DEFMACRO FOO (X) `(BAR ,X))
921 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
922 somewhat surprising style warnings are emitted for
923 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
925 ; (LAMBDA (Y) (FOO Y))
927 ; caught STYLE-WARNING:
928 ; The variable Y is defined but never used.
930 245: bugs in disassembler
931 a. On X86 an immediate operand for IMUL is printed incorrectly.
932 b. On X86 operand size prefix is not recognized.
934 248: "reporting errors in type specifier syntax"
935 (TYPEP 1 '(SYMBOL NIL)) says something about "unknown type
939 (defun foo (&key (a :x))
943 does not cause a warning. (BTW: old SBCL issued a warning, but for a
944 function, which was never called!)
947 Compiler does not emit warnings for
949 a. (lambda () (svref (make-array 8 :adjustable t) 1))
952 (list (let ((y (the real x)))
953 (unless (floatp y) (error ""))
958 (declare (optimize (debug 0)))
959 (declare (type vector x))
960 (list (fill-pointer x)
964 Complex array type does not have corresponding type specifier.
966 This is a problem because the compiler emits optimization notes when
967 you use a non-simple array, and without a type specifier for hairy
968 array types, there's no good way to tell it you're doing it
969 intentionally so that it should shut up and just compile the code.
971 Another problem is confusing error message "asserted type ARRAY
972 conflicts with derived type (VALUES SIMPLE-VECTOR &OPTIONAL)" during
973 compiling (LAMBDA (V) (VALUES (SVREF V 0) (VECTOR-POP V))).
975 The last problem is that when type assertions are converted to type
976 checks, types are represented with type specifiers, so we could lose
977 complex attribute. (Now this is probably not important, because
978 currently checks for complex arrays seem to be performed by
982 (compile nil '(lambda () (aref (make-array 0) 0))) compiles without
983 warning. Analogous cases with the index and length being equal and
984 greater than 0 are warned for; the problem here seems to be that the
985 type required for an array reference of this type is (INTEGER 0 (0))
986 which is canonicalized to NIL.
991 (t1 (specifier-type s)))
992 (eval `(defstruct ,s))
993 (type= t1 (specifier-type s)))
998 b. The same for CSUBTYPEP.
1001 * (let () (list (the (values &optional fixnum) (eval '(values)))))
1002 debugger invoked on condition of type TYPE-ERROR:
1003 The value NIL is not of type FIXNUM.
1005 262: "yet another bug in inline expansion of local functions"
1009 (declare (integer x y))
1012 (declare (integer u))
1013 (if (> (1+ (the unsigned-byte u)) 0)
1015 (return (+ 38 (cos (/ u 78)))))))
1016 (declare (inline xyz))
1018 (* (funcall (eval #'xyz) x)
1020 (funcall (if (> x 5) #'xyz #'identity)
1025 Urgh... It's time to write IR1-copier.
1028 SB-EXT:RUN-PROGRAM is currently non-functional on Linux/PPC;
1029 attempting to use it leads to segmentation violations. This is
1030 probably because of a bogus implementation of
1031 os_restore_fp_control().
1034 David Lichteblau provided (sbcl-devel 2003-06-01) a patch to fix
1035 behaviour of streams with element-type (SIGNED-BYTE 8). The patch
1036 looks reasonable, if not obviously correct; however, it caused the
1037 PPC/Linux port to segfault during warm-init while loading
1038 src/pcl/std-class.fasl. A workaround patch was made, but it would
1039 be nice to understand why the first patch caused problems, and to
1040 fix the cause if possible.
1047 (fact (1- x) (* x i))))
1048 sbcl does not convert the self-recursive call to a jump, though it
1049 is allowed to by CLHS 3.2.2.3. CMUCL, however, does perform this
1052 268: "wrong free declaration scope"
1053 The following code must signal type error:
1055 (locally (declare (optimize (safety 3)))
1056 (flet ((foo (x &optional (y (car x)))
1057 (declare (optimize (safety 0)))
1059 (funcall (eval #'foo) 1)))
1062 SCALE-FLOAT should accept any integer for its second argument.
1064 DEFUNCT CATEGORIES OF BUGS
1066 These labels were used for bugs related to the old IR1 interpreter.
1067 The # values reached 6 before the category was closed down.