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
87 c: Reading of not initialized slot sometimes causes SEGV (for inline
88 accessors it is fixed, but out-of-line still do not perform type
92 (declaim (optimize (safety 3) (speed 1) (space 1)))
95 (defstruct (stringwise-foo (:include foo
96 (x "x" :type simple-string)
97 (y "y" :type simple-string))))
98 (defparameter *stringwise-foo*
99 (make-stringwise-foo))
100 (setf (foo-x *stringwise-foo*) 0)
101 (defun frob-stringwise-foo (sf)
102 (aref (stringwise-foo-x sf) 0))
103 (frob-stringwise-foo *stringwise-foo*)
107 The "compiling top-level form:" output ought to be condensed.
108 Perhaps any number of such consecutive lines ought to turn into a
109 single "compiling top-level forms:" line.
112 The way that the compiler munges types with arguments together
113 with types with no arguments (in e.g. TYPE-EXPAND) leads to
114 weirdness visible to the user:
115 (DEFTYPE FOO () 'FIXNUM)
117 (TYPEP 11 '(FOO)) => T, which seems weird
118 (TYPEP 11 'FIXNUM) => T
119 (TYPEP 11 '(FIXNUM)) signals an error, as it should
120 The situation is complicated by the presence of Common Lisp types
121 like UNSIGNED-BYTE (which can either be used in list form or alone)
122 so I'm not 100% sure that the behavior above is actually illegal.
123 But I'm 90+% sure, and the following related behavior,
125 treating the bare symbol AND as equivalent to '(AND), is specifically
126 forbidden (by the ANSI specification of the AND type).
129 It would be nice if the
131 (during macroexpansion)
132 said what macroexpansion was at fault, e.g.
134 (during macroexpansion of IN-PACKAGE,
135 during macroexpansion of DEFFOO)
138 (SUBTYPEP '(FUNCTION (T BOOLEAN) NIL)
139 '(FUNCTION (FIXNUM FIXNUM) NIL)) => T, T
140 (Also, when this is fixed, we can enable the code in PROCLAIM which
141 checks for incompatible FTYPE redeclarations.)
144 (I *think* this is a bug. It certainly seems like strange behavior. But
145 the ANSI spec is scary, dark, and deep.. -- WHN)
146 (FORMAT NIL "~,1G" 1.4) => "1. "
147 (FORMAT NIL "~3,1G" 1.4) => "1. "
150 Sometimes (SB-EXT:QUIT) fails with
151 Argh! maximum interrupt nesting depth (4096) exceeded, exiting
152 Process inferior-lisp exited abnormally with code 1
153 I haven't noticed a repeatable case of this yet.
156 The printer doesn't report closures very well. This is true in
160 #<Closure Over Function "DEFUN STRUCTURE-SLOT-ACCESSOR" {134D1A1}>
161 It would be nice to make closures have a settable name slot,
162 and make things like DEFSTRUCT and FLET, which create closures,
163 set helpful values into this slot.
166 And as long as we're wishing, it would be awfully nice if INSPECT could
167 also report on closures, telling about the values of the bound variables.
170 The compiler assumes that any time a function of declared FTYPE
171 doesn't signal an error, its arguments were of the declared type.
172 E.g. compiling and loading
173 (DECLAIM (OPTIMIZE (SAFETY 3)))
174 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
176 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
178 (COND ((> (FACTORIAL X) 1.0E6)
179 (FORMAT T "too big~%"))
181 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
183 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
186 will cause the INTEGERP case to be selected, giving bogus output a la
188 This violates the "declarations are assertions" principle.
189 According to the ANSI spec, in the section "System Class FUNCTION",
190 this is a case of "lying to the compiler", but the lying is done
191 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
192 not by the unexpectedly general definition of FACTORIAL. In any case,
193 "declarations are assertions" means that lying to the compiler should
194 cause an error to be signalled, and should not cause a bogus
195 result to be returned. Thus, the compiler should not assume
196 that arbitrary functions check their argument types. (It might
197 make sense to add another flag (CHECKED?) to DEFKNOWN to
198 identify functions which *do* check their argument types.)
199 (Also, verify that the compiler handles declared function
200 return types as assertions.)
203 TYPEP of VALUES types is sometimes implemented very inefficiently, e.g. in
204 (DEFTYPE INDEXOID () '(INTEGER 0 1000))
206 (DECLARE (TYPE INDEXOID X))
207 (THE (VALUES INDEXOID)
209 where the implementation of the type check in function FOO
210 includes a full call to %TYPEP. There are also some fundamental problems
211 with the interpretation of VALUES types (inherited from CMU CL, and
212 from the ANSI CL standard) as discussed on the cmucl-imp@cons.org
213 mailing list, e.g. in Robert Maclachlan's post of 21 Jun 2000.
216 The definitions of SIGCONTEXT-FLOAT-REGISTER and
217 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
218 supported on FreeBSD because the floating point state is not saved,
219 but at least as of FreeBSD 4.0, the floating point state *is* saved,
220 so they could be supported after all. Very likely
221 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
224 (as discussed by Douglas Crosher on the cmucl-imp mailing list ca.
225 Aug. 10, 2000): CMUCL currently interprets 'member as '(member); same
226 issue with 'union, 'and, 'or etc. So even though according to the
227 ANSI spec, bare 'MEMBER, 'AND, and 'OR are not legal types, CMUCL
228 (and now SBCL) interpret them as legal types.
231 a slew of floating-point-related errors reported by Peter Van Eynde
233 b: SBCL's value for LEAST-POSITIVE-SHORT-FLOAT on the x86 is
234 bogus, and should probably be 1.4012985e-45. In SBCL,
235 (/ LEAST-POSITIVE-SHORT-FLOAT 2) returns a number smaller
236 than LEAST-POSITIVE-SHORT-FLOAT. Similar problems
237 exist for LEAST-NEGATIVE-SHORT-FLOAT, LEAST-POSITIVE-LONG-FLOAT,
238 and LEAST-NEGATIVE-LONG-FLOAT.
239 c: Many expressions generate floating infinity on x86/Linux:
244 PVE's regression tests want them to raise errors. sbcl-0.7.0.5
245 on x86/Linux generates the infinities instead. That might or
246 might not be conforming behavior, but it's also inconsistent,
247 which is almost certainly wrong. (Inconsistency: (/ 1 0.0)
248 should give the same result as (/ 1.0 0.0), but instead (/ 1 0.0)
249 generates SINGLE-FLOAT-POSITIVE-INFINITY and (/ 1.0 0.0)
251 d: (in section12.erg) various forms a la
252 (FLOAT 1 DOUBLE-FLOAT-EPSILON)
253 don't give the right behavior.
256 type safety errors reported by Peter Van Eynde July 25, 2000:
257 c: (COERCE 'AND 'FUNCTION) returns something related to
258 (MACRO-FUNCTION 'AND), but ANSI says it should raise an error.
259 k: READ-BYTE is supposed to signal TYPE-ERROR when its argument is
260 not a binary input stream, but instead cheerfully reads from
261 character streams, e.g. (MAKE-STRING-INPUT-STREAM "abc").
264 The debugger LIST-LOCATIONS command doesn't work properly.
267 Compiling and loading
268 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
270 then requesting a BACKTRACE at the debugger prompt gives no information
271 about where in the user program the problem occurred.
274 Paul Werkowski wrote on cmucl-imp@cons.org 2000-11-15
275 I am looking into this problem that showed up on the cmucl-help
276 list. It seems to me that the "implementation specific environment
277 hacking functions" found in pcl/walker.lisp are completely messed
278 up. The good thing is that they appear to be barely used within
279 PCL and the munged environment object is passed to cmucl only
280 in calls to macroexpand-1, which is probably why this case fails.
281 SBCL uses essentially the same code, so if the environment hacking
282 is screwed up, it affects us too.
285 Using the pretty-printer from the command prompt gives funny
286 results, apparently because the pretty-printer doesn't know
287 about user's command input, including the user's carriage return
288 that the user, and therefore the pretty-printer thinks that
289 the new output block should start indented 2 or more characters
290 rightward of the correct location.
293 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
294 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
295 crashes SBCL. In general tracing anything which is used in the
296 implementation of TRACE is likely to have the same problem.
299 As reported by Martin Atzmueller on sbcl-devel 26 Dec 2000,
300 ANSI says that WITH-OUTPUT-TO-STRING should have a keyword
301 :ELEMENT-TYPE, but in sbcl-0.6.9 this is not defined for
302 WITH-OUTPUT-TO-STRING.
305 ANSI says in one place that type declarations can be abbreviated even
306 when the type name is not a symbol, e.g.
307 (DECLAIM ((VECTOR T) *FOOVECTOR*))
308 SBCL doesn't support this. But ANSI says in another place that this
309 isn't allowed. So it's not clear this is a bug after all. (See the
310 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
314 as pointed out by Dan Barlow on sbcl-devel 2000-07-02:
315 The PICK-TEMPORARY-FILE-NAME utility used by LOAD-FOREIGN uses
316 an easily guessable temporary filename in a way which might open
317 applications using LOAD-FOREIGN to hijacking by malicious users
318 on the same machine. Incantations for doing this safely are
319 floating around the net in various "how to write secure programs
320 despite Unix" documents, and it would be good to (1) fix this in
321 LOAD-FOREIGN, and (2) hunt for any other code which uses temporary
322 files and make it share the same new safe logic.
324 (partially alleviated in sbcl-0.7.9.32 by a fix by Matthew Danish to
325 make the temporary filename less easily guessable)
328 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
329 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
330 the mean of the distribution can be systematically O(0.1%) wrong.
331 Just increasing R-I-E-B is probably not a good solution, since
332 it would decrease efficiency more than is probably necessary. Perhaps
333 using some sort of accept/reject method would be better.
336 Internally the compiler sometimes evaluates
337 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
338 (I stumbled across this when I added an
339 (assert (not (eq type1 *wild-type*)))
340 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
341 in a type context should probably be translated to T, and so it's
342 probably wrong to ask whether it's equal to the T type and then (using
343 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
344 (I haven't tried to investigate this bug enough to guess whether
345 there might be any user-level symptoms.)
347 In fact, the type system is likely to depend on this inequality not
348 holding... * is not equivalent to T in many cases, such as
349 (VECTOR *) /= (VECTOR T).
352 Inconsistencies between derived and declared VALUES return types for
353 DEFUN aren't checked very well. E.g. the logic which successfully
354 catches problems like
355 (declaim (ftype (function (fixnum) float) foo))
357 (declare (type integer x))
358 (values x)) ; wrong return type, detected, gives warning, good!
360 (declaim (ftype (function (t) (values t t)) bar))
362 (values x)) ; wrong number of return values, no warning, bad!
363 The cause of this is seems to be that (1) the internal function
364 VALUES-TYPES-EQUAL-OR-INTERSECT used to make the check handles its
365 arguments symmetrically, and (2) when the type checking code was
366 written back when when SBCL's code was still CMU CL, the intent
368 (declaim (ftype (function (t) t) bar))
370 (values x x)) ; wrong number of return values; should give warning?
371 not be warned for, because a two-valued return value is considered
372 to be compatible with callers who expects a single value to be
373 returned. That intent is probably not appropriate for modern ANSI
374 Common Lisp, but fixing this might be complicated because of other
375 divergences between auld-style and new-style handling of
376 multiple-VALUES types. (Some issues related to this were discussed
377 on cmucl-imp at some length sometime in 2000.)
380 The facility for dumping a running Lisp image to disk gets confused
381 when run without the PURIFY option, and creates an unnecessarily large
382 core file (apparently representing memory usage up to the previous
383 high-water mark). Moreover, when the file is loaded, it confuses the
384 GC, so that thereafter memory usage can never be reduced below that
388 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
389 CL), out-of-line structure slot setters are horribly inefficient
390 whenever the type of the slot is declared, because out-of-line
391 structure slot setters are implemented as closures to save space,
392 so the compiler doesn't compile the type test into code, but
393 instead just saves the type in a lexical closure and interprets it
395 A proper solution involves deciding whether it's really worth
396 saving space by implementing structure slot accessors as closures.
397 (If it's not worth it, the problem vanishes automatically. If it
398 is worth it, there are hacks we could use to force type tests to
399 be compiled anyway, and even shared. E.g. we could implement
400 an EQUAL hash table mapping from types to compiled type tests,
401 and save the appropriate compiled type test as part of each lexical
402 closure; or we could make the lexical closures be placeholders
403 which overwrite their old definition as a lexical closure with
404 a new compiled definition the first time that they're called.)
405 As a workaround for the problem, #'(SETF FOO) expressions can
406 be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
407 (defmacro efficient-setf-function (place-function-name)
408 (or #+sbcl (and (sb-impl::info :function :accessor-for place-function-name)
409 ;; a workaround for the problem, encouraging the
410 ;; inline expansion of the structure accessor, so
411 ;; that the compiler can optimize its type test
412 (let ((new-value (gensym "NEW-VALUE-"))
413 (structure-value (gensym "STRUCTURE-VALUE-")))
414 `(lambda (,new-value ,structure-value)
415 (setf (,place-function-name ,structure-value)
417 ;; no problem, can just use the ordinary expansion
418 `(function (setf ,place-function-name))))
421 There's apparently a bug in CEILING optimization which caused
422 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
423 applied the patches to SBCL and they didn't seem to cause problems
424 (as reported sbcl-devel 2001-05-04). However, since the patches
425 modify nontrivial code which was apparently written incorrectly
426 the first time around, until regression tests are written I'm not
427 comfortable merging the patches in the CVS version of SBCL.
430 (TIME (ROOM T)) reports more than 200 Mbytes consed even for
431 a clean, just-started SBCL system. And it seems to be right:
432 (ROOM T) can bring a small computer to its knees for a *long*
433 time trying to GC afterwards. Surely there's some more economical
434 way to implement (ROOM T).
437 When the compiler inline expands functions, it may be that different
438 kinds of return values are generated from different code branches.
439 E.g. an inline expansion of POSITION generates integer results
440 from one branch, and NIL results from another. When that inline
441 expansion is used in a context where only one of those results
444 (aref *a1* (position x *a2*)))
445 and the compiler can't prove that the unacceptable branch is
446 never taken, then bogus type mismatch warnings can be generated.
447 If you need to suppress the type mismatch warnings, you can
448 suppress the inline expansion,
450 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
451 (aref *a1* (position x *a2*)))
452 or, sometimes, suppress them by declaring the result to be of an
455 (aref *a1* (the integer (position x *a2*))))
457 This is not a new compiler problem in 0.7.0, but the new compiler
458 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
459 more conspicuous. If you don't need performance from these functions,
460 and the bogus warnings are a nuisance for you, you can return to
461 your pre-0.7.0 state of grace with
462 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
465 as reported by Eric Marsden on cmucl-imp@cons.org 2001-08-14:
466 (= (FLOAT 1 DOUBLE-FLOAT-EPSILON)
467 (+ (FLOAT 1 DOUBLE-FLOAT-EPSILON) DOUBLE-FLOAT-EPSILON)) => T
468 when of course it should be NIL. (He says it only fails for X86,
469 not SPARC; dunno about Alpha.)
471 Also, "the same problem exists for LONG-FLOAT-EPSILON,
472 DOUBLE-FLOAT-NEGATIVE-EPSILON, LONG-FLOAT-NEGATIVE-EPSILON (though
473 for the -negative- the + is replaced by a - in the test)."
475 Raymond Toy comments that this is tricky on the X86 since its FPU
476 uses 80-bit precision internally.
479 Even in sbcl-0.pre7.x, which is supposed to be free of the old
480 non-ANSI behavior of treating the function return type inferred
481 from the current function definition as a declaration of the
482 return type from any function of that name, the return type of NIL
483 is attached to FOO in 120a above, and used to optimize code which
487 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
488 the entire lexical environment at the point of MACROLET available
489 in the bodies of the macroexpander functions. In particular, it
490 allows the function bodies (which run at compile time) to try to
491 access lexical variables (which are only defined at runtime).
492 It doesn't even issue a warning, which is bad.
494 The SBCL behavior arguably conforms to the ANSI spec (since the
495 spec says that the behavior is undefined, ergo anything conforms).
496 However, it would be better to issue a compile-time error.
497 Unfortunately I (WHN) don't see any simple way to detect this
498 condition in order to issue such an error, so for the meantime
499 SBCL just does this weird broken "conforming" thing.
501 The ANSI standard says, in the definition of the special operator
503 The macro-expansion functions defined by MACROLET are defined
504 in the lexical environment in which the MACROLET form appears.
505 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
506 the local macro definitions in a MACROLET, but the consequences
507 are undefined if the local macro definitions reference any
508 local variable or function bindings that are visible in that
510 Then it seems to contradict itself by giving the example
512 (macrolet ((fudge (z)
513 ;The parameters x and flag are not accessible
514 ; at this point; a reference to flag would be to
515 ; the global variable of that name.
516 ` (if flag (* ,z ,z) ,z)))
517 ;The parameters x and flag are accessible here.
521 The comment "a reference to flag would be to the global variable
522 of the same name" sounds like good behavior for the system to have.
523 but actual specification quoted above says that the actual behavior
526 (Since 0.7.8.23 macroexpanders are defined in a restricted version
527 of the lexical environment, containing no lexical variables and
528 functions, which seems to conform to ANSI and CLtL2, but signalling
529 a STYLE-WARNING for references to variables similar to locals might
533 (as reported by Gabe Garza on cmucl-help 2001-09-21)
535 (defun test-pred (x y)
539 (func (lambda () x)))
540 (print (eq func func))
541 (print (test-pred func func))
542 (delete func (list func))))
543 Now calling (TEST-CASE) gives output
546 (#<FUNCTION {500A9EF9}>)
547 Evidently Python thinks of the lambda as a code transformation so
548 much that it forgets that it's also an object.
551 Ideally, uninterning a symbol would allow it, and its associated
552 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
553 at least as of sbcl-0.7.0, this isn't the case. Information about
554 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
555 essentially in ordinary (non-weak) hash tables keyed by symbols.
556 Thus, once a system has an entry in this system, it tends to live
557 forever, even when it is uninterned and all other references to it
560 141: "pretty printing and backquote"
563 ``(FOO SB-IMPL::BACKQ-COMMA-AT S)
566 * (write '`(, .ala.) :readably t :pretty t)
569 (note the space between the comma and the point)
572 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
573 prominent SourceForge web/db bug tracking system, which is
574 unfortunately not a reliable way to get a timely response from
575 the SBCL maintainers)
576 In the course of trying to build a test case for an
577 application error, I encountered this behavior:
578 If you start up sbcl, and then lay on CTRL-C for a
579 minute or two, the lisp process will eventually say:
580 %PRIMITIVE HALT called; the party is over.
581 and throw you into the monitor. If I start up lisp,
582 attach to the process with strace, and then do the same
583 (abusive) thing, I get instead:
584 access failure in heap page not marked as write-protected
585 and the monitor again. I don't know enough to have the
586 faintest idea of what is going on here.
587 This is with sbcl 6.12, uname -a reports:
588 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
589 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
590 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
591 it took more than two minutes (but less than five) for me.
594 (This was once known as IR1-4, but it lived on even after the
595 IR1 interpreter went to the big bit bucket in the sky.)
596 The system accepts DECLAIM in most places where DECLARE would be
597 accepted, without even issuing a warning. ANSI allows this, but since
598 it's fairly easy to mistype DECLAIM instead of DECLARE, and the
599 meaning is rather different, and it's unlikely that the user
600 has a good reason for doing DECLAIM not at top level, it would be
601 good to issue a STYLE-WARNING when this happens. A possible
602 fix would be to issue STYLE-WARNINGs for DECLAIMs not at top level,
603 or perhaps to issue STYLE-WARNINGs for any EVAL-WHEN not at top level.
604 [This is considered an IR1-interpreter-related bug because until
605 EVAL-WHEN is rewritten, which won't happen until after the IR1
606 interpreter is gone, the system's notion of what's a top-level form
607 and what's not will remain too confused to fix this problem.]
610 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
611 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
612 COMPLEX implementation didn't deal with this, and hasn't been
613 upgraded to do so. (This doesn't seem to be a high priority
614 conformance problem, since seems hard to construct useful code
618 Floating point errors are reported poorly. E.g. on x86 OpenBSD
621 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
622 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
623 No traps are enabled? How can this be?
624 It should be possible to be much more specific (overflow, division
625 by zero, etc.) and of course the "How can this be?" should be fixable.
627 See also bugs #45.c and #183
630 In sbcl-0.7.1.3 on x86, COMPILE-FILE on the file
631 (in-package :cl-user)
634 (defstruct foo bar bletch)
636 (labels ((kidify1 (kid)
644 (declare (inline kid-frob))
647 (the simple-vector (foo-bar perd)))))
649 debugger invoked on condition of type TYPE-ERROR:
650 The value NIL is not of type SB-C::NODE.
651 The location of this failure has moved around as various related
652 issues were cleaned up. As of sbcl-0.7.1.9, it occurs in
653 NODE-BLOCK called by LAMBDA-COMPONENT called by IR2-CONVERT-CLOSURE.
655 (Python LET-converts KIDIFY1 into KID-FROB, then tries to inline
656 expand KID-FROB into %ZEEP. Having partially done it, it sees a call
657 of KIDIFY1, which already does not exist. So it gives up on
658 expansion, leaving garbage consisting of infinished blocks of the
659 partially converted function.)
661 (due to reordering of the compiler this example is compiled
662 successfully by 0.7.14, but the bug probably remains)
665 (reported by Robert E. Brown 2002-04-16)
666 When a function is called with too few arguments, causing the
667 debugger to be entered, the uninitialized slots in the bad call frame
668 seem to cause GCish problems, being interpreted as tagged data even
669 though they're not. In particular, executing ROOM in the
670 debugger at that point causes AVER failures:
673 * (lisp-implementation-version)
679 failed AVER: "(SAP= CURRENT END)"
680 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
681 isn't too surprising since there are many differences in stack
682 implementation and GC conservatism between the X86 and other ports.)
685 In sbcl-0.7.3.11, compiling the (illegal) code
686 (in-package :cl-user)
687 (defmethod prove ((uustk uustk))
690 gives the (not terribly clear) error message
692 ; (during macroexpansion of (DEFMETHOD PROVE ...))
693 ; can't get template for (FROB NIL NIL)
694 The problem seems to be that the code walker used by the DEFMETHOD
695 macro is unhappy with the illegal syntax in the method body, and
696 is giving an unclear error message.
699 The compiler sometimes tries to constant-fold expressions before
700 it checks to see whether they can be reached. This can lead to
701 bogus warnings about errors in the constant folding, e.g. in code
704 (WRITE-STRING (> X 0) "+" "0"))
705 compiled in a context where the compiler can prove that X is NIL,
706 and the compiler complains that (> X 0) causes a type error because
707 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
708 caused a full WARNING, which made the bug really annoying because then
709 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
710 code. Since then the warning has been downgraded to STYLE-WARNING,
711 so it's still a bug but at least it's a little less annoying.
713 183: "IEEE floating point issues"
714 Even where floating point handling is being dealt with relatively
715 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
716 accrued-exceptions and current-exceptions part of the fp control
717 word don't seem to bear much relation to reality. E.g. on
721 debugger invoked on condition of type DIVISION-BY-ZERO:
722 arithmetic error DIVISION-BY-ZERO signalled
723 0] (sb-vm::get-floating-point-modes)
725 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
726 :ROUNDING-MODE :NEAREST
727 :CURRENT-EXCEPTIONS NIL
728 :ACCRUED-EXCEPTIONS (:INEXACT)
731 * (sb-vm::get-floating-point-modes)
732 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
733 :ROUNDING-MODE :NEAREST
734 :CURRENT-EXCEPTIONS (:INEXACT)
735 :ACCRUED-EXCEPTIONS (:INEXACT)
738 187: "type inference confusion around DEFTRANSFORM time"
739 (reported even more verbosely on sbcl-devel 2002-06-28 as "strange
740 bug in DEFTRANSFORM")
741 After the file below is compiled and loaded in sbcl-0.7.5, executing
742 (TCX (MAKE-ARRAY 4 :FILL-POINTER 2) 0)
743 at the REPL returns an adjustable vector, which is wrong. Presumably
744 somehow the DERIVE-TYPE information for the output values of %WAD is
745 being mispropagated as a type constraint on the input values of %WAD,
746 and so causing the type test to be optimized away. It's unclear how
747 hand-expanding the DEFTRANSFORM would change this, but it suggests
748 the DEFTRANSFORM machinery (or at least the way DEFTRANSFORMs are
749 invoked at a particular phase) is involved.
750 (cl:in-package :sb-c)
751 (eval-when (:compile-toplevel)
752 ;;; standin for %DATA-VECTOR-AND-INDEX
753 (defknown %dvai (array index)
755 (foldable flushable))
756 (deftransform %dvai ((array index)
760 (let* ((atype (continuation-type array))
761 (eltype (array-type-specialized-element-type atype)))
762 (when (eq eltype *wild-type*)
763 (give-up-ir1-transform
764 "specialized array element type not known at compile-time"))
765 (when (not (array-type-complexp atype))
766 (give-up-ir1-transform "SIMPLE array!"))
767 `(if (array-header-p array)
768 (%wad array index nil)
769 (values array index))))
770 ;;; standin for %WITH-ARRAY-DATA
771 (defknown %wad (array index (or index null))
772 (values (simple-array * (*)) index index index)
773 (foldable flushable))
774 ;;; (Commenting out this optimizer causes the bug to go away.)
775 (defoptimizer (%wad derive-type) ((array start end))
776 (let ((atype (continuation-type array)))
777 (when (array-type-p atype)
778 (values-specifier-type
779 `(values (simple-array ,(type-specifier
780 (array-type-specialized-element-type atype))
782 index index index)))))
784 (defun %wad (array start end)
785 (format t "~&in %WAD~%")
786 (%with-array-data array start end))
787 (cl:in-package :cl-user)
789 (declare (type (vector t) v))
790 (declare (notinline sb-kernel::%with-array-data))
791 ;; (Hand-expending DEFTRANSFORM %DVAI here also causes the bug to
795 188: "compiler performance fiasco involving type inference and UNION-TYPE"
796 (In sbcl-0.7.6.10, DEFTRANSFORM CONCATENATE was commented out until this
797 bug could be fixed properly, so you won't see the bug unless you restore
798 the DEFTRANSFORM by hand.) In sbcl-0.7.5.11 on a 700 MHz Pentium III,
802 (declare (optimize (safety 3)))
803 (declare (optimize (compilation-speed 2)))
804 (declare (optimize (speed 1) (debug 1) (space 1)))
805 (let ((fn "if-this-file-exists-the-universe-is-strange"))
806 (load fn :if-does-not-exist nil)
807 (load (concatenate 'string fn ".lisp") :if-does-not-exist nil)
808 (load (concatenate 'string fn ".fasl") :if-does-not-exist nil)
809 (load (concatenate 'string fn ".misc-garbage")
810 :if-does-not-exist nil)))))
812 134.552 seconds of real time
813 133.35156 seconds of user run time
814 0.03125 seconds of system run time
815 [Run times include 2.787 seconds GC run time.]
817 246883368 bytes consed.
818 BACKTRACE from Ctrl-C in the compilation shows that the compiler is
819 thinking about type relationships involving types like
821 (OR (INTEGER 576 576)
832 In recent SBCL the following example also illustrates this bug:
837 (declare (optimize (safety 3)))
838 (declare (optimize (compilation-speed 2)))
839 (declare (optimize (speed 1) (debug 1) (space 1)))
841 (declare (type (integer 0) start))
842 (print (incf start 22))
843 (print (incf start 26))
844 (print (incf start 28)))
846 (declare (type (integer 0) start))
847 (print (incf start 22))
848 (print (incf start 26)))
850 (declare (type (integer 0) start))
851 (print (incf start 22))
852 (print (incf start 26))))))
854 190: "PPC/Linux pipe? buffer? bug"
855 In sbcl-0.7.6, the run-program.test.sh test script sometimes hangs
856 on the PPC/Linux platform, waiting for a zombie env process. This
857 is a classic symptom of buffer filling and deadlock, but it seems
858 only sporadically reproducible.
860 191: "Miscellaneous PCL deficiencies"
861 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
862 a. DEFCLASS does not inform the compiler about generated
863 functions. Compiling a file with
867 (WITH-SLOTS (A-CLASS-X) A
869 results in a STYLE-WARNING:
871 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
873 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
874 Mai points out that the declamation of functions is in fact
875 incorrect in some cases (most notably for structure
876 classes). This means that at present erroneous attempts to use
877 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
878 won't get the corresponding STYLE-WARNING.
879 c. the examples in CLHS 7.6.5.1 (regarding generic function lambda
880 lists and &KEY arguments) do not signal errors when they should.
883 201: "Incautious type inference from compound CONS types"
884 (reported by APD sbcl-devel 2002-09-17)
886 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
888 (FORMAT NIL "~S IS ~S, Y = ~S"
895 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
897 205: "environment issues in cross compiler"
898 (These bugs have no impact on user code, but should be fixed or
900 a. Macroexpanders introduced with MACROLET are defined in the null
902 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
903 the null lexical environment.
904 c. The cross-compiler cannot inline functions defined in a non-null
907 206: ":SB-FLUID feature broken"
908 (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
909 Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
912 207: "poorly distributed SXHASH results for compound data"
913 SBCL's SXHASH could probably try a little harder. ANSI: "the
914 intent is that an implementation should make a good-faith
915 effort to produce hash-codes that are well distributed
916 within the range of non-negative fixnums". But
917 (let ((hits (make-hash-table)))
920 (let* ((ij (cons i j))
921 (newlist (push ij (gethash (sxhash ij) hits))))
923 (format t "~&collision: ~S~%" newlist))))))
924 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
925 would be an obvious way of fix. Maybe it would be acceptably efficient
926 to redo MIX using a lookup into a 256-entry s-box containing
927 29-bit pseudorandom numbers?
929 208: "package confusion in PCL handling of structure slot handlers"
930 In sbcl-0.7.8 compiling and loading
932 (defstruct foo (slot (error "missing")) :type list :read-only t)
933 (defmethod print-object ((foo foo) stream) (print nil stream))
934 causes CERROR "attempting to modify a symbol in the COMMON-LISP
935 package: FOO-SLOT". (This is fairly bad code, but still it's hard
936 to see that it should cause symbols to be interned in the CL package.)
938 211: "keywords processing"
939 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
940 number of keyword arguments.
943 (flet ((foo (&key y) (list y)))
944 (list (foo :y 1 :y 2)))
946 issues confusing message
951 ; caught STYLE-WARNING:
952 ; The variable #:G15 is defined but never used.
954 212: "Sequence functions and circular arguments"
955 COERCE, MERGE and CONCATENATE go into an infinite loop when given
956 circular arguments; it would be good for the user if they could be
957 given an error instead (ANSI 17.1.1 allows this behaviour on the part
958 of the implementation, as conforming code cannot give non-proper
959 sequences to these functions. MAP also has this problem (and
960 solution), though arguably the convenience of being able to do
961 (MAP 'LIST '+ FOO '#1=(1 . #1#))
962 might be classed as more important (though signalling an error when
963 all of the arguments are circular is probably desireable).
965 213: "Sequence functions and type checking"
966 a. MAKE-SEQUENCE, COERCE, MERGE and CONCATENATE cannot deal with
967 various complicated, though recognizeable, CONS types [e.g.
968 (CONS * (CONS * NULL))
969 which according to ANSI should be recognized] (and, in SAFETY 3
970 code, should return a list of LENGTH 2 or signal an error)
971 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
972 check that it will return a sequence of the given type. Fixing
973 it along the same lines as the others (cf. work done around
974 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
975 entirely straightforward.
976 c. All of these functions will silently accept a type of the form
978 whether or not the return value is of this type. This is
979 probably permitted by ANSI (see "Exceptional Situations" under
980 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
981 know about this escape clause, so code of the form
982 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
983 can erroneously return T.
986 SBCL 0.6.12.43 fails to compile
989 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
990 (flet ((foo (&key (x :vx x-p)) (list x x-p)))
993 or a more simple example:
996 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
997 (lambda (x) (declare (fixnum x)) (if (< x 0) 0 (1- x))))
999 215: ":TEST-NOT handling by functions"
1000 a. FIND and POSITION currently signal errors when given non-NIL for
1001 both their :TEST and (deprecated) :TEST-NOT arguments, but by
1002 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
1003 means that the effect is "unpredictable but harmless". It's not
1004 clear what that actually means; it may preclude conforming
1005 implementations from signalling errors.
1006 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
1007 when conflict occurs. As a quality of implementation issue, it
1008 might be preferable to treat :TEST and :TEST-NOT as being in some
1009 sense the same &KEY, and effectively take the first test function in
1011 c. Again, a quality of implementation issue: it would be good to issue a
1012 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
1013 WARNING for calls with both :TEST and :TEST-NOT; possibly this
1014 latter should be WARNed about at execute-time too.
1016 216: "debugger confused by frames with invalid number of arguments"
1017 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
1018 leaves the system confused, enough so that (QUIT) no longer works.
1019 It's as though the process of working with the uninitialized slot in
1020 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
1021 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
1023 217: "Bad type operations with FUNCTION types"
1026 * (values-type-union (specifier-type '(function (base-char)))
1027 (specifier-type '(function (integer))))
1029 #<FUN-TYPE (FUNCTION (BASE-CHAR) *)>
1031 It causes insertion of wrong type assertions into generated
1035 (let ((f (etypecase x
1036 (character #'write-char)
1037 (integer #'write-byte))))
1040 (character (write-char x s))
1041 (integer (write-byte x s)))))
1043 Then (FOO #\1 *STANDARD-OUTPUT*) signals type error.
1045 (In 0.7.9.1 the result type is (FUNCTION * *), so Python does not
1046 produce invalid code, but type checking is not accurate. Similar
1047 problems exist with VALUES-TYPE-INTERSECTION.)
1049 218: "VALUES type specifier semantics"
1050 (THE (VALUES ...) ...) in safe code discards extra values.
1052 (defun test (x y) (the (values integer) (truncate x y)))
1056 Sbcl 0.7.9 fails to compile
1058 (multiple-value-call #'list
1059 (the integer (helper))
1062 Type check for INTEGER, the result of which serves as the first
1063 argument of M-V-C, is inserted after evaluation of NIL. So arguments
1064 of M-V-C are pushed in the wrong order. As a temporary workaround
1065 type checking was disabled for M-V-Cs in 0.7.9.13. A better solution
1066 would be to put the check between evaluation of arguments, but it
1067 could be tricky to check result types of PROG1, IF etc.
1070 (subtypep 'function '(function)) => nil, t.
1072 233: bugs in constraint propagation
1075 (declare (optimize (speed 2) (safety 3)))
1078 (the double-float x)
1081 (quux y (+ y 2d0) (* y 3d0)))))
1082 (foo 4) => segmentation violation
1084 (see usage of CONTINUATION-ASSERTED-TYPE in USE-RESULT-CONSTRAINTS)
1088 (declaim (optimize (speed 2) (safety 3)))
1090 (if (typep (prog1 x (setq x y)) 'double-float)
1093 (foo 1d0 5) => segmentation violation
1095 235: "type system and inline expansion"
1097 (declaim (ftype (function (cons) number) acc))
1098 (declaim (inline acc))
1100 (the number (car c)))
1103 (values (locally (declare (optimize (safety 0)))
1105 (locally (declare (optimize (safety 3)))
1108 (foo '(nil) '(t)) => NIL, T.
1110 b. (reported by brown on #lisp 2003-01-21)
1113 (declare (optimize (speed 3) (safety 0)))
1114 (declare (notinline mapcar))
1115 (let ((z (mapcar #'car x)))
1118 Without (DECLARE (NOTINLINE MAPCAR)), Python cannot derive that Z is
1121 237: "Environment arguments to type functions"
1122 a. Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
1123 UPGRADED-COMPLEX-PART-TYPE now have an optional environment
1124 argument, but they ignore it completely. This is almost
1125 certainly not correct.
1126 b. Also, the compiler's optimizers for TYPEP have not been informed
1127 about the new argument; consequently, they will not transform
1128 calls of the form (TYPEP 1 'INTEGER NIL), even though this is
1129 just as optimizeable as (TYPEP 1 'INTEGER).
1131 238: "REPL compiler overenthusiasm for CLOS code"
1133 * (defclass foo () ())
1134 * (defmethod bar ((x foo) (foo foo)) (call-next-method))
1135 causes approximately 100 lines of code deletion notes. Some
1136 discussion on this issue happened under the title 'Three "interesting"
1137 bugs in PCL', resulting in a fix for this oververbosity from the
1138 compiler proper; however, the problem persists in the interactor
1139 because the notion of original source is not preserved: for the
1140 compiler, the original source of the above expression is (DEFMETHOD
1141 BAR ((X FOO) (FOO FOO)) (CALL-NEXT-METHOD)), while by the time the
1142 compiler gets its hands on the code needing compilation from the REPL,
1143 it has been macroexpanded several times.
1145 A symptom of the same underlying problem, reported by Tony Martinez:
1147 (with-input-from-string (*query-io* " no")
1149 (simple-type-error () 'error))
1151 ; (SB-KERNEL:FLOAT-WAIT)
1153 ; note: deleting unreachable code
1154 ; compilation unit finished
1157 241: "DEFCLASS mysteriously remembers uninterned accessor names."
1158 (from tonyms on #lisp IRC 2003-02-25)
1159 In sbcl-0.7.12.55, typing
1160 (defclass foo () ((bar :accessor foo-bar)))
1163 (defclass foo () ((bar :accessor foo-bar)))
1164 gives the error message
1165 "#:FOO-BAR already names an ordinary function or a macro."
1166 So it's somehow checking the uninterned old accessor name instead
1167 of the new requested accessor name, which seems broken to me (WHN).
1169 242: "WRITE-SEQUENCE suboptimality"
1170 (observed from clx performance)
1171 In sbcl-0.7.13, WRITE-SEQUENCE of a sequence of type
1172 (SIMPLE-ARRAY (UNSIGNED-BYTE 8) (*)) on a stream with element-type
1173 (UNSIGNED-BYTE 8) will write to the stream one byte at a time,
1174 rather than writing the sequence in one go, leading to severe
1175 performance degradation.
1177 243: "STYLE-WARNING overenthusiasm for unused variables"
1178 (observed from clx compilation)
1179 In sbcl-0.7.14, in the presence of the macros
1180 (DEFMACRO FOO (X) `(BAR ,X))
1181 (DEFMACRO BAR (X) (DECLARE (IGNORABLE X)) 'NIL)
1182 somewhat surprising style warnings are emitted for
1183 (COMPILE NIL '(LAMBDA (Y) (FOO Y))):
1185 ; (LAMBDA (Y) (FOO Y))
1187 ; caught STYLE-WARNING:
1188 ; The variable Y is defined but never used.
1190 245: bugs in disassembler
1191 a. On X86 an immediate operand for IMUL is printed incorrectly.
1192 b. On X86 operand size prefix is not recognized.
1194 246: "NTH-VALUE scaling problem"
1195 NTH-VALUE's current implementation for constant integers scales in
1196 compile-time as O(n^4), as indeed must the optional dispatch
1197 mechanism on which it is implemented. While it is unlikely to
1198 matter in real user code, it's still unpleasant to observe that
1199 (NTH-VALUE 1000 (VALUES-LIST (MAKE-LIST 1001))) takes several hours
1202 248: "reporting errors in type specifier syntax"
1203 (TYPEP 1 '(SYMBOL NIL)) says something about "unknown type
1207 (defun foo (&key (a :x))
1208 (declare (fixnum a))
1211 does not cause a warning. (BTW: old SBCL issued a warning, but for a
1212 function, which was never called!)
1214 DEFUNCT CATEGORIES OF BUGS
1216 These labels were used for bugs related to the old IR1 interpreter.
1217 The # values reached 6 before the category was closed down.