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.
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
88 bogus warnings about undefined functions for magic functions like
89 SB!C::%%DEFUN and SB!C::%DEFCONSTANT when cross-compiling files
90 like src/code/float.lisp. Fixing this will probably require
91 straightening out enough bootstrap consistency issues that
92 the cross-compiler can run with *TYPE-SYSTEM-INITIALIZED*.
93 Instead, the cross-compiler runs in a slightly flaky state
94 which is sane enough to compile SBCL itself, but which is
95 also unstable in several ways, including its inability
96 to really grok function declarations.
98 As of sbcl-0.7.5, sbcl's cross-compiler does run with
99 *TYPE-SYSTEM-INITIALIZED*; however, this bug remains.
102 The "compiling top-level form:" output ought to be condensed.
103 Perhaps any number of such consecutive lines ought to turn into a
104 single "compiling top-level forms:" line.
107 The way that the compiler munges types with arguments together
108 with types with no arguments (in e.g. TYPE-EXPAND) leads to
109 weirdness visible to the user:
110 (DEFTYPE FOO () 'FIXNUM)
112 (TYPEP 11 '(FOO)) => T, which seems weird
113 (TYPEP 11 'FIXNUM) => T
114 (TYPEP 11 '(FIXNUM)) signals an error, as it should
115 The situation is complicated by the presence of Common Lisp types
116 like UNSIGNED-BYTE (which can either be used in list form or alone)
117 so I'm not 100% sure that the behavior above is actually illegal.
118 But I'm 90+% sure, and the following related behavior,
120 treating the bare symbol AND as equivalent to '(AND), is specifically
121 forbidden (by the ANSI specification of the AND type).
124 It would be nice if the
126 (during macroexpansion)
127 said what macroexpansion was at fault, e.g.
129 (during macroexpansion of IN-PACKAGE,
130 during macroexpansion of DEFFOO)
133 (SUBTYPEP '(FUNCTION (T BOOLEAN) NIL)
134 '(FUNCTION (FIXNUM FIXNUM) NIL)) => T, T
135 (Also, when this is fixed, we can enable the code in PROCLAIM which
136 checks for incompatible FTYPE redeclarations.)
139 (I *think* this is a bug. It certainly seems like strange behavior. But
140 the ANSI spec is scary, dark, and deep.. -- WHN)
141 (FORMAT NIL "~,1G" 1.4) => "1. "
142 (FORMAT NIL "~3,1G" 1.4) => "1. "
145 from Marco Antoniotti on cmucl-imp mailing list 1 Mar 2000:
147 (setf (find-class 'ccc1) (find-class 'ccc))
148 (defmethod zut ((c ccc1)) 123)
149 In sbcl-0.7.1.13, this gives an error,
150 There is no class named CCC1.
151 DTC's recommended workaround from the mailing list 3 Mar 2000:
152 (setf (pcl::find-class 'ccc1) (pcl::find-class 'ccc))
155 Sometimes (SB-EXT:QUIT) fails with
156 Argh! maximum interrupt nesting depth (4096) exceeded, exiting
157 Process inferior-lisp exited abnormally with code 1
158 I haven't noticed a repeatable case of this yet.
161 The printer doesn't report closures very well. This is true in
165 #<Closure Over Function "DEFUN STRUCTURE-SLOT-ACCESSOR" {134D1A1}>
166 It would be nice to make closures have a settable name slot,
167 and make things like DEFSTRUCT and FLET, which create closures,
168 set helpful values into this slot.
171 And as long as we're wishing, it would be awfully nice if INSPECT could
172 also report on closures, telling about the values of the bound variables.
175 The compiler assumes that any time a function of declared FTYPE
176 doesn't signal an error, its arguments were of the declared type.
177 E.g. compiling and loading
178 (DECLAIM (OPTIMIZE (SAFETY 3)))
179 (DEFUN FACTORIAL (X) (GAMMA (1+ X)))
181 (DECLAIM (FTYPE (FUNCTION (UNSIGNED-BYTE)) FACTORIAL))
183 (COND ((> (FACTORIAL X) 1.0E6)
184 (FORMAT T "too big~%"))
186 (FORMAT T "exactly ~S~%" (FACTORIAL X)))
188 (FORMAT T "approximately ~S~%" (FACTORIAL X)))))
191 will cause the INTEGERP case to be selected, giving bogus output a la
193 This violates the "declarations are assertions" principle.
194 According to the ANSI spec, in the section "System Class FUNCTION",
195 this is a case of "lying to the compiler", but the lying is done
196 by the code which calls FACTORIAL with non-UNSIGNED-BYTE arguments,
197 not by the unexpectedly general definition of FACTORIAL. In any case,
198 "declarations are assertions" means that lying to the compiler should
199 cause an error to be signalled, and should not cause a bogus
200 result to be returned. Thus, the compiler should not assume
201 that arbitrary functions check their argument types. (It might
202 make sense to add another flag (CHECKED?) to DEFKNOWN to
203 identify functions which *do* check their argument types.)
204 (Also, verify that the compiler handles declared function
205 return types as assertions.)
208 TYPEP of VALUES types is sometimes implemented very inefficiently, e.g. in
209 (DEFTYPE INDEXOID () '(INTEGER 0 1000))
211 (DECLARE (TYPE INDEXOID X))
212 (THE (VALUES INDEXOID)
214 where the implementation of the type check in function FOO
215 includes a full call to %TYPEP. There are also some fundamental problems
216 with the interpretation of VALUES types (inherited from CMU CL, and
217 from the ANSI CL standard) as discussed on the cmucl-imp@cons.org
218 mailing list, e.g. in Robert Maclachlan's post of 21 Jun 2000.
221 The definitions of SIGCONTEXT-FLOAT-REGISTER and
222 %SET-SIGCONTEXT-FLOAT-REGISTER in x86-vm.lisp say they're not
223 supported on FreeBSD because the floating point state is not saved,
224 but at least as of FreeBSD 4.0, the floating point state *is* saved,
225 so they could be supported after all. Very likely
226 SIGCONTEXT-FLOATING-POINT-MODES could now be supported, too.
229 (as discussed by Douglas Crosher on the cmucl-imp mailing list ca.
230 Aug. 10, 2000): CMUCL currently interprets 'member as '(member); same
231 issue with 'union, 'and, 'or etc. So even though according to the
232 ANSI spec, bare 'MEMBER, 'AND, and 'OR are not legal types, CMUCL
233 (and now SBCL) interpret them as legal types.
236 a slew of floating-point-related errors reported by Peter Van Eynde
238 b: SBCL's value for LEAST-POSITIVE-SHORT-FLOAT is bogus, and
239 should probably be 1.4012985e-45. In SBCL,
240 (/ LEAST-POSITIVE-SHORT-FLOAT 2) returns a number smaller
241 than LEAST-POSITIVE-SHORT-FLOAT. Similar problems
242 exist for LEAST-NEGATIVE-SHORT-FLOAT, LEAST-POSITIVE-LONG-FLOAT,
243 and LEAST-NEGATIVE-LONG-FLOAT.
244 c: Many expressions generate floating infinity on x86/Linux:
249 PVE's regression tests want them to raise errors. sbcl-0.7.0.5
250 on x86/Linux generates the infinities instead. That might or
251 might not be conforming behavior, but it's also inconsistent,
252 which is almost certainly wrong. (Inconsistency: (/ 1 0.0)
253 should give the same result as (/ 1.0 0.0), but instead (/ 1 0.0)
254 generates SINGLE-FLOAT-POSITIVE-INFINITY and (/ 1.0 0.0)
256 d: (in section12.erg) various forms a la
257 (FLOAT 1 DOUBLE-FLOAT-EPSILON)
258 don't give the right behavior.
261 type safety errors reported by Peter Van Eynde July 25, 2000:
262 c: (COERCE 'AND 'FUNCTION) returns something related to
263 (MACRO-FUNCTION 'AND), but ANSI says it should raise an error.
264 k: READ-BYTE is supposed to signal TYPE-ERROR when its argument is
265 not a binary input stream, but instead cheerfully reads from
266 character streams, e.g. (MAKE-STRING-INPUT-STREAM "abc").
269 DEFCLASS bugs reported by Peter Van Eynde July 25, 2000:
270 d: (DEFGENERIC IF (X)) should signal a PROGRAM-ERROR, but instead
271 causes a COMPILER-ERROR.
274 miscellaneous errors reported by Peter Van Eynde July 25, 2000:
276 (DEFGENERIC FOO02 (X))
277 (DEFMETHOD FOO02 ((X NUMBER)) T)
278 (LET ((M (FIND-METHOD (FUNCTION FOO02)
280 (LIST (FIND-CLASS (QUOTE NUMBER))))))
281 (REMOVE-METHOD (FUNCTION FOO02) M)
282 (DEFGENERIC FOO03 (X))
283 (ADD-METHOD (FUNCTION FOO03) M)))
284 should give an error, but SBCL allows it.
287 It has been reported (e.g. by Peter Van Eynde) that there are
288 several metaobject protocol "errors". (In order to fix them, we might
289 need to document exactly what metaobject protocol specification
290 we're following -- the current code is just inherited from PCL.)
293 The debugger LIST-LOCATIONS command doesn't work properly.
296 Compiling and loading
297 (DEFUN FAIL (X) (THROW 'FAIL-TAG X))
299 then requesting a BACKTRACE at the debugger prompt gives no information
300 about where in the user program the problem occurred.
303 Paul Werkowski wrote on cmucl-imp@cons.org 2000-11-15
304 I am looking into this problem that showed up on the cmucl-help
305 list. It seems to me that the "implementation specific environment
306 hacking functions" found in pcl/walker.lisp are completely messed
307 up. The good thing is that they appear to be barely used within
308 PCL and the munged environment object is passed to cmucl only
309 in calls to macroexpand-1, which is probably why this case fails.
310 SBCL uses essentially the same code, so if the environment hacking
311 is screwed up, it affects us too.
314 Using the pretty-printer from the command prompt gives funny
315 results, apparently because the pretty-printer doesn't know
316 about user's command input, including the user's carriage return
317 that the user, and therefore the pretty-printer thinks that
318 the new output block should start indented 2 or more characters
319 rightward of the correct location.
322 As reported by Winton Davies on a CMU CL mailing list 2000-01-10,
323 and reported for SBCL by Martin Atzmueller 2000-10-20: (TRACE GETHASH)
324 crashes SBCL. In general tracing anything which is used in the
325 implementation of TRACE is likely to have the same problem.
328 As reported by Martin Atzmueller on sbcl-devel 26 Dec 2000,
329 ANSI says that WITH-OUTPUT-TO-STRING should have a keyword
330 :ELEMENT-TYPE, but in sbcl-0.6.9 this is not defined for
331 WITH-OUTPUT-TO-STRING.
334 ANSI says in one place that type declarations can be abbreviated even
335 when the type name is not a symbol, e.g.
336 (DECLAIM ((VECTOR T) *FOOVECTOR*))
337 SBCL doesn't support this. But ANSI says in another place that this
338 isn't allowed. So it's not clear this is a bug after all. (See the
339 e-mail on cmucl-help@cons.org on 2001-01-16 and 2001-01-17 from WHN
343 as pointed out by Dan Barlow on sbcl-devel 2000-07-02:
344 The PICK-TEMPORARY-FILE-NAME utility used by LOAD-FOREIGN uses
345 an easily guessable temporary filename in a way which might open
346 applications using LOAD-FOREIGN to hijacking by malicious users
347 on the same machine. Incantations for doing this safely are
348 floating around the net in various "how to write secure programs
349 despite Unix" documents, and it would be good to (1) fix this in
350 LOAD-FOREIGN, and (2) hunt for any other code which uses temporary
351 files and make it share the same new safe logic.
353 (partially alleviated in sbcl-0.7.9.32 by a fix by Matthew Danish to
354 make the temporary filename less easily guessable)
357 Functions are assigned names based on the context in which they're
358 defined. This is less than ideal for the functions which are
359 used to implement CLOS methods. E.g. the output of
360 (DESCRIBE 'PRINT-OBJECT) lists functions like
361 #<FUNCTION "DEF!STRUCT (TRACE-INFO (:MAKE-LOAD-FORM-FUN SB-KERNEL:JUST-DUMP-IT-NORMALLY) (:PRINT-OBJECT #))" {1020E49}>
363 #<FUNCTION "MACROLET ((FORCE-DELAYED-DEF!METHODS NIL #))" {1242871}>
364 It would be better if these functions' names always identified
365 them as methods, and identified their generic functions and
369 RANDOM-INTEGER-EXTRA-BITS=10 may not be large enough for the RANDOM
370 RNG to be high quality near RANDOM-FIXNUM-MAX; it looks as though
371 the mean of the distribution can be systematically O(0.1%) wrong.
372 Just increasing R-I-E-B is probably not a good solution, since
373 it would decrease efficiency more than is probably necessary. Perhaps
374 using some sort of accept/reject method would be better.
377 Internally the compiler sometimes evaluates
378 (sb-kernel:type/= (specifier-type '*) (specifier-type t))
379 (I stumbled across this when I added an
380 (assert (not (eq type1 *wild-type*)))
381 in the NAMED :SIMPLE-= type method.) '* isn't really a type, and
382 in a type context should probably be translated to T, and so it's
383 probably wrong to ask whether it's equal to the T type and then (using
384 the EQ type comparison in the NAMED :SIMPLE-= type method) return NIL.
385 (I haven't tried to investigate this bug enough to guess whether
386 there might be any user-level symptoms.)
388 In fact, the type system is likely to depend on this inequality not
389 holding... * is not equivalent to T in many cases, such as
390 (VECTOR *) /= (VECTOR T).
393 Inconsistencies between derived and declared VALUES return types for
394 DEFUN aren't checked very well. E.g. the logic which successfully
395 catches problems like
396 (declaim (ftype (function (fixnum) float) foo))
398 (declare (type integer x))
399 (values x)) ; wrong return type, detected, gives warning, good!
401 (declaim (ftype (function (t) (values t t)) bar))
403 (values x)) ; wrong number of return values, no warning, bad!
404 The cause of this is seems to be that (1) the internal function
405 VALUES-TYPES-EQUAL-OR-INTERSECT used to make the check handles its
406 arguments symmetrically, and (2) when the type checking code was
407 written back when when SBCL's code was still CMU CL, the intent
409 (declaim (ftype (function (t) t) bar))
411 (values x x)) ; wrong number of return values; should give warning?
412 not be warned for, because a two-valued return value is considered
413 to be compatible with callers who expects a single value to be
414 returned. That intent is probably not appropriate for modern ANSI
415 Common Lisp, but fixing this might be complicated because of other
416 divergences between auld-style and new-style handling of
417 multiple-VALUES types. (Some issues related to this were discussed
418 on cmucl-imp at some length sometime in 2000.)
421 The facility for dumping a running Lisp image to disk gets confused
422 when run without the PURIFY option, and creates an unnecessarily large
423 core file (apparently representing memory usage up to the previous
424 high-water mark). Moreover, when the file is loaded, it confuses the
425 GC, so that thereafter memory usage can never be reduced below that
429 In sbcl-0.6.11.41 (and in all earlier SBCL, and in CMU
430 CL), out-of-line structure slot setters are horribly inefficient
431 whenever the type of the slot is declared, because out-of-line
432 structure slot setters are implemented as closures to save space,
433 so the compiler doesn't compile the type test into code, but
434 instead just saves the type in a lexical closure and interprets it
436 A proper solution involves deciding whether it's really worth
437 saving space by implementing structure slot accessors as closures.
438 (If it's not worth it, the problem vanishes automatically. If it
439 is worth it, there are hacks we could use to force type tests to
440 be compiled anyway, and even shared. E.g. we could implement
441 an EQUAL hash table mapping from types to compiled type tests,
442 and save the appropriate compiled type test as part of each lexical
443 closure; or we could make the lexical closures be placeholders
444 which overwrite their old definition as a lexical closure with
445 a new compiled definition the first time that they're called.)
446 As a workaround for the problem, #'(SETF FOO) expressions can
447 be replaced with (EFFICIENT-SETF-FUNCTION FOO), where
448 (defmacro efficient-setf-function (place-function-name)
449 (or #+sbcl (and (sb-impl::info :function :accessor-for place-function-name)
450 ;; a workaround for the problem, encouraging the
451 ;; inline expansion of the structure accessor, so
452 ;; that the compiler can optimize its type test
453 (let ((new-value (gensym "NEW-VALUE-"))
454 (structure-value (gensym "STRUCTURE-VALUE-")))
455 `(lambda (,new-value ,structure-value)
456 (setf (,place-function-name ,structure-value)
458 ;; no problem, can just use the ordinary expansion
459 `(function (setf ,place-function-name))))
462 There's apparently a bug in CEILING optimization which caused
463 Douglas Crosher to patch the CMU CL version. Martin Atzmueller
464 applied the patches to SBCL and they didn't seem to cause problems
465 (as reported sbcl-devel 2001-05-04). However, since the patches
466 modify nontrivial code which was apparently written incorrectly
467 the first time around, until regression tests are written I'm not
468 comfortable merging the patches in the CVS version of SBCL.
471 (TIME (ROOM T)) reports more than 200 Mbytes consed even for
472 a clean, just-started SBCL system. And it seems to be right:
473 (ROOM T) can bring a small computer to its knees for a *long*
474 time trying to GC afterwards. Surely there's some more economical
475 way to implement (ROOM T).
478 When the compiler inline expands functions, it may be that different
479 kinds of return values are generated from different code branches.
480 E.g. an inline expansion of POSITION generates integer results
481 from one branch, and NIL results from another. When that inline
482 expansion is used in a context where only one of those results
485 (aref *a1* (position x *a2*)))
486 and the compiler can't prove that the unacceptable branch is
487 never taken, then bogus type mismatch warnings can be generated.
488 If you need to suppress the type mismatch warnings, you can
489 suppress the inline expansion,
491 #+sbcl (declare (notinline position)) ; to suppress bug 117 bogowarnings
492 (aref *a1* (position x *a2*)))
493 or, sometimes, suppress them by declaring the result to be of an
496 (aref *a1* (the integer (position x *a2*))))
498 This is not a new compiler problem in 0.7.0, but the new compiler
499 transforms for FIND, POSITION, FIND-IF, and POSITION-IF make it
500 more conspicuous. If you don't need performance from these functions,
501 and the bogus warnings are a nuisance for you, you can return to
502 your pre-0.7.0 state of grace with
503 #+sbcl (declaim (notinline find position find-if position-if)) ; bug 117..
506 as reported by Eric Marsden on cmucl-imp@cons.org 2001-08-14:
507 (= (FLOAT 1 DOUBLE-FLOAT-EPSILON)
508 (+ (FLOAT 1 DOUBLE-FLOAT-EPSILON) DOUBLE-FLOAT-EPSILON)) => T
509 when of course it should be NIL. (He says it only fails for X86,
510 not SPARC; dunno about Alpha.)
512 Also, "the same problem exists for LONG-FLOAT-EPSILON,
513 DOUBLE-FLOAT-NEGATIVE-EPSILON, LONG-FLOAT-NEGATIVE-EPSILON (though
514 for the -negative- the + is replaced by a - in the test)."
516 Raymond Toy comments that this is tricky on the X86 since its FPU
517 uses 80-bit precision internally.
520 Even in sbcl-0.pre7.x, which is supposed to be free of the old
521 non-ANSI behavior of treating the function return type inferred
522 from the current function definition as a declaration of the
523 return type from any function of that name, the return type of NIL
524 is attached to FOO in 120a above, and used to optimize code which
528 As of version 0.pre7.14, SBCL's implementation of MACROLET makes
529 the entire lexical environment at the point of MACROLET available
530 in the bodies of the macroexpander functions. In particular, it
531 allows the function bodies (which run at compile time) to try to
532 access lexical variables (which are only defined at runtime).
533 It doesn't even issue a warning, which is bad.
535 The SBCL behavior arguably conforms to the ANSI spec (since the
536 spec says that the behavior is undefined, ergo anything conforms).
537 However, it would be better to issue a compile-time error.
538 Unfortunately I (WHN) don't see any simple way to detect this
539 condition in order to issue such an error, so for the meantime
540 SBCL just does this weird broken "conforming" thing.
542 The ANSI standard says, in the definition of the special operator
544 The macro-expansion functions defined by MACROLET are defined
545 in the lexical environment in which the MACROLET form appears.
546 Declarations and MACROLET and SYMBOL-MACROLET definitions affect
547 the local macro definitions in a MACROLET, but the consequences
548 are undefined if the local macro definitions reference any
549 local variable or function bindings that are visible in that
551 Then it seems to contradict itself by giving the example
553 (macrolet ((fudge (z)
554 ;The parameters x and flag are not accessible
555 ; at this point; a reference to flag would be to
556 ; the global variable of that name.
557 ` (if flag (* ,z ,z) ,z)))
558 ;The parameters x and flag are accessible here.
562 The comment "a reference to flag would be to the global variable
563 of the same name" sounds like good behavior for the system to have.
564 but actual specification quoted above says that the actual behavior
567 (Since 0.7.8.23 macroexpanders are defined in a restricted version
568 of the lexical environment, containing no lexical variables and
569 functions, which seems to conform to ANSI and CLtL2, but signalling
570 a STYLE-WARNING for references to variables similar to locals might
574 (as reported by Gabe Garza on cmucl-help 2001-09-21)
576 (defun test-pred (x y)
580 (func (lambda () x)))
581 (print (eq func func))
582 (print (test-pred func func))
583 (delete func (list func))))
584 Now calling (TEST-CASE) gives output
587 (#<FUNCTION {500A9EF9}>)
588 Evidently Python thinks of the lambda as a code transformation so
589 much that it forgets that it's also an object.
592 Ideally, uninterning a symbol would allow it, and its associated
593 FDEFINITION and PROCLAIM data, to be reclaimed by the GC. However,
594 at least as of sbcl-0.7.0, this isn't the case. Information about
595 FDEFINITIONs and PROCLAIMed properties is stored in globaldb.lisp
596 essentially in ordinary (non-weak) hash tables keyed by symbols.
597 Thus, once a system has an entry in this system, it tends to live
598 forever, even when it is uninterned and all other references to it
601 141: "pretty printing and backquote"
604 ``(FOO SB-IMPL::BACKQ-COMMA-AT S)
607 * (write '`(, .ala.) :readably t :pretty t)
610 (note the space between the comma and the point)
614 (reported by Jesse Bouwman 2001-10-24 through the unfortunately
615 prominent SourceForge web/db bug tracking system, which is
616 unfortunately not a reliable way to get a timely response from
617 the SBCL maintainers)
618 In the course of trying to build a test case for an
619 application error, I encountered this behavior:
620 If you start up sbcl, and then lay on CTRL-C for a
621 minute or two, the lisp process will eventually say:
622 %PRIMITIVE HALT called; the party is over.
623 and throw you into the monitor. If I start up lisp,
624 attach to the process with strace, and then do the same
625 (abusive) thing, I get instead:
626 access failure in heap page not marked as write-protected
627 and the monitor again. I don't know enough to have the
628 faintest idea of what is going on here.
629 This is with sbcl 6.12, uname -a reports:
630 Linux prep 2.2.19 #4 SMP Tue Apr 24 13:59:52 CDT 2001 i686 unknown
631 I (WHN) have verified that the same thing occurs on sbcl-0.pre7.141
632 under OpenBSD 2.9 on my X86 laptop. Do be patient when you try it:
633 it took more than two minutes (but less than five) for me.
636 (This was once known as IR1-4, but it lived on even after the
637 IR1 interpreter went to the big bit bucket in the sky.)
638 The system accepts DECLAIM in most places where DECLARE would be
639 accepted, without even issuing a warning. ANSI allows this, but since
640 it's fairly easy to mistype DECLAIM instead of DECLARE, and the
641 meaning is rather different, and it's unlikely that the user
642 has a good reason for doing DECLAIM not at top level, it would be
643 good to issue a STYLE-WARNING when this happens. A possible
644 fix would be to issue STYLE-WARNINGs for DECLAIMs not at top level,
645 or perhaps to issue STYLE-WARNINGs for any EVAL-WHEN not at top level.
646 [This is considered an IR1-interpreter-related bug because until
647 EVAL-WHEN is rewritten, which won't happen until after the IR1
648 interpreter is gone, the system's notion of what's a top-level form
649 and what's not will remain too confused to fix this problem.]
652 ANSI allows types `(COMPLEX ,FOO) to use very hairy values for
653 FOO, e.g. (COMPLEX (AND REAL (SATISFIES ODDP))). The old CMU CL
654 COMPLEX implementation didn't deal with this, and hasn't been
655 upgraded to do so. (This doesn't seem to be a high priority
656 conformance problem, since seems hard to construct useful code
660 Floating point errors are reported poorly. E.g. on x86 OpenBSD
663 debugger invoked on condition of type SB-KERNEL:FLOATING-POINT-EXCEPTION:
664 An arithmetic error SB-KERNEL:FLOATING-POINT-EXCEPTION was signalled.
665 No traps are enabled? How can this be?
666 It should be possible to be much more specific (overflow, division
667 by zero, etc.) and of course the "How can this be?" should be fixable.
669 See also bugs #45.c and #183
672 In sbcl-0.7.1.3 on x86, COMPILE-FILE on the file
673 (in-package :cl-user)
676 (defstruct foo bar bletch)
678 (labels ((kidify1 (kid)
686 (declare (inline kid-frob))
689 (the simple-vector (foo-bar perd)))))
691 debugger invoked on condition of type TYPE-ERROR:
692 The value NIL is not of type SB-C::NODE.
693 The location of this failure has moved around as various related
694 issues were cleaned up. As of sbcl-0.7.1.9, it occurs in
695 NODE-BLOCK called by LAMBDA-COMPONENT called by IR2-CONVERT-CLOSURE.
697 (Python LET-converts KIDIFY1 into KID-FROB, then tries to inline
698 expand KID-FROB into %ZEEP. Having partially done it, it sees a call
699 of KIDIFY1, which already does not exist. So it gives up on
700 expansion, leaving garbage consisting of infinished blocks of the
701 partially converted function.)
704 Functions SUBTYPEP, TYPEP, UPGRADED-ARRAY-ELEMENT-TYPE, and
705 UPGRADED-COMPLEX-PART-TYPE should have an optional environment argument.
706 (reported by Alexey Dejneka sbcl-devel 2002-04-12)
709 (reported by Robert E. Brown 2002-04-16)
710 When a function is called with too few arguments, causing the
711 debugger to be entered, the uninitialized slots in the bad call frame
712 seem to cause GCish problems, being interpreted as tagged data even
713 though they're not. In particular, executing ROOM in the
714 debugger at that point causes AVER failures:
717 * (lisp-implementation-version)
723 failed AVER: "(SAP= CURRENT END)"
724 (Christophe Rhodes reports that this doesn't occur on the SPARC, which
725 isn't too surprising since there are many differences in stack
726 implementation and GC conservatism between the X86 and other ports.)
729 In sbcl-0.7.3.11, compiling the (illegal) code
730 (in-package :cl-user)
731 (defmethod prove ((uustk uustk))
734 gives the (not terribly clear) error message
736 ; (during macroexpansion of (DEFMETHOD PROVE ...))
737 ; can't get template for (FROB NIL NIL)
738 The problem seems to be that the code walker used by the DEFMETHOD
739 macro is unhappy with the illegal syntax in the method body, and
740 is giving an unclear error message.
743 The compiler sometimes tries to constant-fold expressions before
744 it checks to see whether they can be reached. This can lead to
745 bogus warnings about errors in the constant folding, e.g. in code
748 (WRITE-STRING (> X 0) "+" "0"))
749 compiled in a context where the compiler can prove that X is NIL,
750 and the compiler complains that (> X 0) causes a type error because
751 NIL isn't a valid argument to #'>. Until sbcl-0.7.4.10 or so this
752 caused a full WARNING, which made the bug really annoying because then
753 COMPILE and COMPILE-FILE returned FAILURE-P=T for perfectly legal
754 code. Since then the warning has been downgraded to STYLE-WARNING,
755 so it's still a bug but at least it's a little less annoying.
757 183: "IEEE floating point issues"
758 Even where floating point handling is being dealt with relatively
759 well (as of sbcl-0.7.5, on sparc/sunos and alpha; see bug #146), the
760 accrued-exceptions and current-exceptions part of the fp control
761 word don't seem to bear much relation to reality. E.g. on
765 debugger invoked on condition of type DIVISION-BY-ZERO:
766 arithmetic error DIVISION-BY-ZERO signalled
767 0] (sb-vm::get-floating-point-modes)
769 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
770 :ROUNDING-MODE :NEAREST
771 :CURRENT-EXCEPTIONS NIL
772 :ACCRUED-EXCEPTIONS (:INEXACT)
775 * (sb-vm::get-floating-point-modes)
776 (:TRAPS (:OVERFLOW :INVALID :DIVIDE-BY-ZERO)
777 :ROUNDING-MODE :NEAREST
778 :CURRENT-EXCEPTIONS (:INEXACT)
779 :ACCRUED-EXCEPTIONS (:INEXACT)
782 187: "type inference confusion around DEFTRANSFORM time"
783 (reported even more verbosely on sbcl-devel 2002-06-28 as "strange
784 bug in DEFTRANSFORM")
785 After the file below is compiled and loaded in sbcl-0.7.5, executing
786 (TCX (MAKE-ARRAY 4 :FILL-POINTER 2) 0)
787 at the REPL returns an adjustable vector, which is wrong. Presumably
788 somehow the DERIVE-TYPE information for the output values of %WAD is
789 being mispropagated as a type constraint on the input values of %WAD,
790 and so causing the type test to be optimized away. It's unclear how
791 hand-expanding the DEFTRANSFORM would change this, but it suggests
792 the DEFTRANSFORM machinery (or at least the way DEFTRANSFORMs are
793 invoked at a particular phase) is involved.
794 (cl:in-package :sb-c)
795 (eval-when (:compile-toplevel)
796 ;;; standin for %DATA-VECTOR-AND-INDEX
797 (defknown %dvai (array index)
799 (foldable flushable))
800 (deftransform %dvai ((array index)
804 (let* ((atype (continuation-type array))
805 (eltype (array-type-specialized-element-type atype)))
806 (when (eq eltype *wild-type*)
807 (give-up-ir1-transform
808 "specialized array element type not known at compile-time"))
809 (when (not (array-type-complexp atype))
810 (give-up-ir1-transform "SIMPLE array!"))
811 `(if (array-header-p array)
812 (%wad array index nil)
813 (values array index))))
814 ;;; standin for %WITH-ARRAY-DATA
815 (defknown %wad (array index (or index null))
816 (values (simple-array * (*)) index index index)
817 (foldable flushable))
818 ;;; (Commenting out this optimizer causes the bug to go away.)
819 (defoptimizer (%wad derive-type) ((array start end))
820 (let ((atype (continuation-type array)))
821 (when (array-type-p atype)
822 (values-specifier-type
823 `(values (simple-array ,(type-specifier
824 (array-type-specialized-element-type atype))
826 index index index)))))
828 (defun %wad (array start end)
829 (format t "~&in %WAD~%")
830 (%with-array-data array start end))
831 (cl:in-package :cl-user)
833 (declare (type (vector t) v))
834 (declare (notinline sb-kernel::%with-array-data))
835 ;; (Hand-expending DEFTRANSFORM %DVAI here also causes the bug to
839 188: "compiler performance fiasco involving type inference and UNION-TYPE"
840 (In sbcl-0.7.6.10, DEFTRANSFORM CONCATENATE was commented out until this
841 bug could be fixed properly, so you won't see the bug unless you restore
842 the DEFTRANSFORM by hand.) In sbcl-0.7.5.11 on a 700 MHz Pentium III,
846 (declare (optimize (safety 3)))
847 (declare (optimize (compilation-speed 2)))
848 (declare (optimize (speed 1) (debug 1) (space 1)))
849 (let ((fn "if-this-file-exists-the-universe-is-strange"))
850 (load fn :if-does-not-exist nil)
851 (load (concatenate 'string fn ".lisp") :if-does-not-exist nil)
852 (load (concatenate 'string fn ".fasl") :if-does-not-exist nil)
853 (load (concatenate 'string fn ".misc-garbage")
854 :if-does-not-exist nil)))))
856 134.552 seconds of real time
857 133.35156 seconds of user run time
858 0.03125 seconds of system run time
859 [Run times include 2.787 seconds GC run time.]
861 246883368 bytes consed.
862 BACKTRACE from Ctrl-C in the compilation shows that the compiler is
863 thinking about type relationships involving types like
865 (OR (INTEGER 576 576)
876 190: "PPC/Linux pipe? buffer? bug"
877 In sbcl-0.7.6, the run-program.test.sh test script sometimes hangs
878 on the PPC/Linux platform, waiting for a zombie env process. This
879 is a classic symptom of buffer filling and deadlock, but it seems
880 only sporadically reproducible.
882 191: "Miscellaneous PCL deficiencies"
883 (reported by Alexey Dejneka sbcl-devel 2002-08-04)
884 a. DEFCLASS does not inform the compiler about generated
885 functions. Compiling a file with
889 (WITH-SLOTS (A-CLASS-X) A
891 results in a STYLE-WARNING:
893 SB-SLOT-ACCESSOR-NAME::|COMMON-LISP-USER A-CLASS-X slot READER|
895 APD's fix for this was checked in to sbcl-0.7.6.20, but Pierre
896 Mai points out that the declamation of functions is in fact
897 incorrect in some cases (most notably for structure
898 classes). This means that at present erroneous attempts to use
899 WITH-SLOTS and the like on classes with metaclass STRUCTURE-CLASS
900 won't get the corresponding STYLE-WARNING.
901 c. the examples in CLHS 7.6.5.1 (regarding generic function lambda
902 lists and &KEY arguments) do not signal errors when they should.
904 192: "Python treats free type declarations as promises."
905 b. What seemed like the same fundamental problem as bug 192a, but
906 was not fixed by the same (APD "more strict type checking
907 sbcl-devel 2002-08-97) patch:
908 (DOTIMES (I ...) (DOTIMES (J ...) (DECLARE ...) ...)):
909 (declaim (optimize (speed 1) (safety 3)))
910 (defun trust-assertion (i)
912 (declare (type (mod 4) i)) ; when commented out, behavior changes!
915 (trust-assertion 6) ; prints nothing unless DECLARE is commented out
920 (locally (declare (type fixnum x y))
924 194: "no error from (THE REAL '(1 2 3)) in some cases"
927 (multiple-value-prog1 (progn (the real '(1 2 3))))
928 returns (1 2 3) instead of signalling an error. This was fixed by
929 APD's "more strict type checking patch", but although the fixed
930 code (in sbcl-0.7.7.19) works (signals TYPE-ERROR) interactively,
931 it's difficult to write a regression test for it, because
932 (IGNORE-ERRORS (MULTIPLE-VALUE-PROG1 (PROGN (THE REAL '(1 2 3)))))
933 still returns (1 2 3).
935 b. (IGNORE-ERRORS (MULTIPLE-VALUE-PROG1 (PROGN (THE REAL '(1 2 3)))))
936 returns (1 2 3). (As above, this shows up when writing regression
937 tests for fixed-ness of part a.)
938 c. Also in sbcl-0.7.7.9, (IGNORE-ERRORS (THE REAL '(1 2 3))) => (1 2 3).
942 (arg2 (identity (the real #(1 2 3)))))
943 (if (< arg1 arg2) arg1 arg2))))
945 but putting the same expression inside (DEFUN FOO () ...),
948 * Actually this entry is probably multiple bugs, as
949 Alexey Dejneka commented on sbcl-devel 2002-09-03:)
950 I don't think that placing these two bugs in one entry is
951 a good idea: they have different explanations. The second
952 (min 1 nil) is caused by flushing of unused code--IDENTITY
953 can do nothing with it. So it is really bug 122. The first
954 (min nil) is due to M-V-PROG1: substituting a continuation
955 for the result, it forgets about type assertion. The purpose
956 of IDENTITY is to save the restricted continuation from
957 inaccurate transformations.
958 * Alexey Dejneka pointed out that
959 (IGNORE-ERRORS (IDENTITY (THE REAL '(1 2 3))))
961 (IGNORE-ERRORS (VALUES (THE REAL '(1 2 3))))
964 201: "Incautious type inference from compound CONS types"
965 (reported by APD sbcl-devel 2002-09-17)
967 (LET ((Y (CAR (THE (CONS INTEGER *) X))))
969 (FORMAT NIL "~S IS ~S, Y = ~S"
976 (FOO ' (1 . 2)) => "NIL IS INTEGER, Y = 1"
979 Compiler does not check THEs on unused values, e.g. in
981 (progn (the real (list 1)) t)
983 This situation may appear during optimizing away degenerate cases of
984 certain functions: see bug 192b.
986 205: "environment issues in cross compiler"
987 (These bugs have no impact on user code, but should be fixed or
989 a. Macroexpanders introduced with MACROLET are defined in the null
991 b. The body of (EVAL-WHEN (:COMPILE-TOPLEVEL) ...) is evaluated in
992 the null lexical environment.
993 c. The cross-compiler cannot inline functions defined in a non-null
996 206: ":SB-FLUID feature broken"
997 (reported by Antonio Martinez-Shotton sbcl-devel 2002-10-07)
998 Enabling :SB-FLUID in the target-features list in sbcl-0.7.8 breaks
1001 207: "poorly distributed SXHASH results for compound data"
1002 SBCL's SXHASH could probably try a little harder. ANSI: "the
1003 intent is that an implementation should make a good-faith
1004 effort to produce hash-codes that are well distributed
1005 within the range of non-negative fixnums". But
1006 (let ((hits (make-hash-table)))
1009 (let* ((ij (cons i j))
1010 (newlist (push ij (gethash (sxhash ij) hits))))
1012 (format t "~&collision: ~S~%" newlist))))))
1013 reports lots of collisions in sbcl-0.7.8. A stronger MIX function
1014 would be an obvious way of fix. Maybe it would be acceptably efficient
1015 to redo MIX using a lookup into a 256-entry s-box containing
1016 29-bit pseudorandom numbers?
1018 208: "package confusion in PCL handling of structure slot handlers"
1019 In sbcl-0.7.8 compiling and loading
1021 (defstruct foo (slot (error "missing")) :type list :read-only t)
1022 (defmethod print-object ((foo foo) stream) (print nil stream))
1023 causes CERROR "attempting to modify a symbol in the COMMON-LISP
1024 package: FOO-SLOT". (This is fairly bad code, but still it's hard
1025 to see that it should cause symbols to be interned in the CL package.)
1027 211: "keywords processing"
1028 a. :ALLOW-OTHER-KEYS T should allow a function to receive an odd
1029 number of keyword arguments.
1032 (flet ((foo (&key y) (list y)))
1033 (list (foo :y 1 :y 2)))
1035 issues confusing message
1040 ; caught STYLE-WARNING:
1041 ; The variable #:G15 is defined but never used.
1043 212: "Sequence functions and circular arguments"
1044 COERCE, MERGE and CONCATENATE go into an infinite loop when given
1045 circular arguments; it would be good for the user if they could be
1046 given an error instead (ANSI 17.1.1 allows this behaviour on the part
1047 of the implementation, as conforming code cannot give non-proper
1048 sequences to these functions. MAP also has this problem (and
1049 solution), though arguably the convenience of being able to do
1050 (MAP 'LIST '+ FOO '#1=(1 . #1#))
1051 might be classed as more important (though signalling an error when
1052 all of the arguments are circular is probably desireable).
1054 213: "Sequence functions and type checking"
1055 a. MAKE-SEQUENCE, COERCE, MERGE and CONCATENATE cannot deal with
1056 various complicated, though recognizeable, CONS types [e.g.
1057 (CONS * (CONS * NULL))
1058 which according to ANSI should be recognized] (and, in SAFETY 3
1059 code, should return a list of LENGTH 2 or signal an error)
1060 b. MAP, when given a type argument that is SUBTYPEP LIST, does not
1061 check that it will return a sequence of the given type. Fixing
1062 it along the same lines as the others (cf. work done around
1063 sbcl-0.7.8.45) is possible, but doing so efficiently didn't look
1064 entirely straightforward.
1065 c. All of these functions will silently accept a type of the form
1067 whether or not the return value is of this type. This is
1068 probably permitted by ANSI (see "Exceptional Situations" under
1069 ANSI MAKE-SEQUENCE), but the DERIVE-TYPE mechanism does not
1070 know about this escape clause, so code of the form
1071 (INTEGERP (CAR (MAKE-SEQUENCE '(CONS INTEGER *) 2)))
1072 can erroneously return T.
1075 SBCL 0.6.12.43 fails to compile
1078 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
1079 (flet ((foo (&key (x :vx x-p)) (list x x-p)))
1082 or a more simple example:
1085 (declare (optimize (inhibit-warnings 0) (compilation-speed 2)))
1086 (lambda (x) (declare (fixnum x)) (if (< x 0) 0 (1- x))))
1088 215: ":TEST-NOT handling by functions"
1089 a. FIND and POSITION currently signal errors when given non-NIL for
1090 both their :TEST and (deprecated) :TEST-NOT arguments, but by
1091 ANSI 17.2 "the consequences are unspecified", which by ANSI 1.4.2
1092 means that the effect is "unpredictable but harmless". It's not
1093 clear what that actually means; it may preclude conforming
1094 implementations from signalling errors.
1095 b. COUNT, REMOVE and the like give priority to a :TEST-NOT argument
1096 when conflict occurs. As a quality of implementation issue, it
1097 might be preferable to treat :TEST and :TEST-NOT as being in some
1098 sense the same &KEY, and effectively take the first test function in
1100 c. Again, a quality of implementation issue: it would be good to issue a
1101 STYLE-WARNING at compile-time for calls with :TEST-NOT, and a
1102 WARNING for calls with both :TEST and :TEST-NOT; possibly this
1103 latter should be WARNed about at execute-time too.
1105 216: "debugger confused by frames with invalid number of arguments"
1106 In sbcl-0.7.8.51, executing e.g. (VECTOR-PUSH-EXTEND T), BACKTRACE, Q
1107 leaves the system confused, enough so that (QUIT) no longer works.
1108 It's as though the process of working with the uninitialized slot in
1109 the bad VECTOR-PUSH-EXTEND frame causes GC problems, though that may
1110 not be the actual problem. (CMU CL 18c doesn't have problems with this.)
1112 217: "Bad type operations with FUNCTION types"
1115 * (values-type-union (specifier-type '(function (base-char)))
1116 (specifier-type '(function (integer))))
1118 #<FUN-TYPE (FUNCTION (BASE-CHAR) *)>
1120 It causes insertion of wrong type assertions into generated
1124 (let ((f (etypecase x
1125 (character #'write-char)
1126 (integer #'write-byte))))
1129 (character (write-char x s))
1130 (integer (write-byte x s)))))
1132 Then (FOO #\1 *STANDARD-OUTPUT*) signals type error.
1134 (In 0.7.9.1 the result type is (FUNCTION * *), so Python does not
1135 produce invalid code, but type checking is not accurate. Similar
1136 problems exist with VALUES-TYPE-INTERSECTION.)
1138 218: "VALUES type specifier semantics"
1139 (THE (VALUES ...) ...) in safe code discards extra values.
1141 (defun test (x y) (the (values integer) (truncate x y)))
1145 Sbcl 0.7.9 fails to compile
1147 (multiple-value-call #'list
1148 (the integer (helper))
1151 Type check for INTEGER, the result of which serves as the first
1152 argument of M-V-C, is inserted after evaluation of NIL. So arguments
1153 of M-V-C are pushed in the wrong order. As a temporary workaround
1154 type checking was disabled for M-V-Cs in 0.7.9.13. A better solution
1155 would be to put the check between evaluation of arguments, but it
1156 could be tricky to check result types of PROG1, IF etc.
1158 228: "function-lambda-expression problems"
1159 in sbcl-0.7.9.6x, from the REPL:
1160 * (progn (declaim (inline foo)) (defun foo (x) x))
1162 * (function-lambda-expression #'foo)
1163 (SB-C:LAMBDA-WITH-LEXENV NIL NIL NIL (X) (BLOCK FOO X)), NIL, FOO
1164 but this first return value is not suitable for input to FUNCTION or
1165 COMPILE, as required by ANSI.
1168 (subtypep 'function '(function)) => nil, t.
1170 233: bugs in constraint propagation
1173 (declare (optimize (speed 2) (safety 3)))
1176 (the double-float x)
1179 (quux y (+ y 2d0) (* y 3d0)))))
1180 (foo 4) => segmentation violation
1182 (see usage of CONTINUATION-ASSERTED-TYPE in USE-RESULT-CONSTRAINTS)
1185 (declaim (optimize (speed 2) (safety 3)))
1187 (if (typep (prog1 x (setq x y)) 'double-float)
1190 (foo 1d0 5) => segmentation violation
1193 (fixed in sbcl-0.7.10.36)
1195 DEFUNCT CATEGORIES OF BUGS
1197 These labels were used for bugs related to the old IR1 interpreter.
1198 The # values reached 6 before the category was closed down.