This file contains suggestions for further work.
See also <https://sourceforge.net/tracker/?group_id=1355&atid=351355>

URGENT PROBLEMS
===============
Should handle these soon.


bind-eval.tst failure on Linux/amd64 [bruno]


regression: [bruno]
=============== script.lisp ==================
(print (list 'script-args *args*))
(print (list '*standard-input* *standard-input*))
(print (list '*standard-output* *standard-output*))
(print (list '*error-output* *error-output*))
(print (list '*terminal-io* *terminal-io*))
(print (list '*debug-io* *debug-io*))
(print "tell me something...")
(print (read-line *standard-input*))
"first"
(with-open-stream (in (make-stream :input))
  (print in)
  (print "tell me something else...")
  (print (read-line in)))
"second"
(setf (stream-element-type *standard-input*) '(unsigned-byte 8))
(print (list '*standard-input* *standard-input*))
=============== script.lisp ==================
$ cat script.lisp | clisp
*** - (SETF STREAM-ELEMENT-TYPE) on #<IO TERMINAL-STREAM> is illegal
$ clisp < script.lisp
same + hangs instead of terminating.
$ clisp script.lisp
works fine.


Many pretty-printer ansi test failures.
(sds - but bruno has to fix the risky-test in iofkts.tst first)


<http://article.gmane.org/gmane.lisp.clisp.general:9893>:
DIRECTORY, *PATHNAME-ENCODING*, and Denial-Of-Service attack
I would call this an error in clisp: if it lists a directory
over which a user has no control (e.g., /tmp), then it is trivial for
an adversary to create a denial-of-service attack.
Unix pathnames are not strictly in any particular character encoding:
they are merely byte strings in which only slash and NUL have special
significance.
Should *PATHNAME-ENCODING* be 1:1?


Inefficiency of class redefinition when old or new class is unfinalizable


support perseverances also on write side
change wr_by_array_iau8_buffered to use persev instead of persev_full


binary stdio
<http://article.gmane.org/gmane.lisp.clisp.devel:12268>
one of the common issues with CLISP is
*** - (SETF STREAM-ELEMENT-TYPE) on #<IO TERMINAL-STREAM> is illegal
i.e., inability to use binary IO in stdio.
it all boils down to regular_handle_p(FD):
when the handle is found to be "regular", the appropriate
*STANDARD-*PUT* stream becomes a handle stream whose element type can be
changed, otherwise a TERMINAL-STREAM is used.
Right now, `regular' means "file on disk", as in
[unix] (S_ISREG(statbuf.st_mode) || S_ISBLK(statbuf.st_mode))
[win32] (filetype == FILE_TYPE_DISK)
what about pipes?
while MAKE-STREAM should provide a universal solution to these issues,
it still might make sense to change regular_handle_p().
Note that this runs into the FRESH-LINE/same_handle_p issues.
Maybe (SETF STREAM-ELEMENT-TYPE) should work on TERMINAL-STREAMs?
Maybe all streams should be bivalent?
Related issue: LISTEN works on binary streams in all Lisps except CLISP.
LISTEN should work on all streams and READ-CHAR-WILL-HANG-P should be
merged with READ-BYTE-WILL-HANG-P into WILL-HANG-P.


Changes to Unicode standard
(Bruno and Jörg discussing issues Nov. 2006)
Eliminate surrogates D800-DFFF from characters
"Surrogates do not represent characters"
"there are not and will never be *surrogate characters* (i.e. encoded
characters represented with a single surrogate code point)"
Change: (code-char #xD800) -> NIL, and #\uD800 becomes illegal
This is to be consistent and avoid problems with embedded string constants
 in .fas files, e.g. (defvar foo #.(string (code-char #xD800))) ; error
Goal: CLISP shall never generate an illegal UTF sequence (unlike
 versions <= 2.41). Similarly, CLISP shall not construct a
 weird string e.g. by reading an illegal sequence.
Conversion to and from binary signals error.
Affects: UTF-8, UTF-7, UCS2, UCS4 and likely others.
Unpaired surrogates shall signal error when reading (like iconv does).
CODE-CHAR-LIMIT is preserved, but CODE-CHAR yields NIL for all
surrogates because they don't stand on their own (they are considered
an artefact of the days when 16 bit where hoped sufficient).
Of course, valid sequences are not affected.  Code #x10400 is represented as
F0 90 90 80 in UTF-8 and as 01 D8 00 DC or D8 01 DC 00 in UTF-16.

Not adressed here: "The two values FFFE and FFFF as well as the 32
values from FDD0 to FDEF represent noncharacters."


Reading illegal UTF byte sequences
Markus Kuhn has some recommendations
http://www.cl.cam.ac.uk/~mgk25/ucs/examples/UTF-8-test.txt
CLISP currently differs from these, c.f. bug #[ 1575569 ]
Should we follow these more closely?
http://sourceforge.net/tracker/index.php?func=detail&aid=1575569&group_id=1355&atid=101355
(ext:convert-string-from-bytes #(65 #xB5 66)
 (ext:make-encoding :charset "utf-8" :input-error-action #\Z))
"AB" (clisp src/encoding.d)
(ext:convert-string-from-bytes #(65 #xB5 66)
 (ext:make-encoding :charset "\\utf-8" :input-error-action #\Z))
"AZB" (iconv)


Embeddability: additional API in spvw.d; example.
<https://sourceforge.net/tracker/index.php?func=detail&aid=423264&group_id=1355&atid=351355>


Update PARI module: there is no global precision variable anymore.
Many functions have been added since the module has been written.


2000-12-04 did some changes to not goto outside `with_string_0', however
pathname.d:open_file() still has such a case.
Likewise, pathname.d:directory_search_hashcode `return's while in scope.
Important only for those compilers that neither support alloca() nor
{ var char x[foo(y)]; ...} variable-sized arrays?


nested REQUIRE is broken
<http://article.gmane.org/gmane.lisp.clisp.devel:13691>:
suppose I have a library which consists of 10 files - which create
packages, define macros &c.
I create a special file foo.lisp:
--- foo ---
(load "foo0")
(load "foo1")
....
(load "foo9")
(provide "foo")
--- foo ---
then (require "foo") will load the whole library,
but since foo.lib is _empty_, this won't work:
--- bar ---
(require "foo")
(defun bar () (foo:with-foo (...) ...))
--- bar ---
when bar.lisp is compiled, the empty foo.lib is loaded and there is no
package FOO, no macro WITH-FOO &c.
to fix this we need to, first, replace LOAD with REQUIRE in foo.lisp,
and, second, change c-REQUIRE to write the REQUIRE form into *.lib.
unfortunately, this means that each foo?.lisp must PROVIDE "foo?" which
is quite ugly.


Review all macros and tighten the runtime interface of the macros.
  check-redefinition


contagion warnings: engineering vs portability
<http://sourceforge.net/mailarchive/message.php?msg_id=10935288>
<http://article.gmane.org/gmane.lisp.clisp.devel:13758>
*warn-on-floating-point-contagion*              engineering warning
      the user is doing something mathematically unsound
*warn-on-floating-point-rational-contagion*     "portability" warning
      we don't like the standard requirement
since our ANSI mode makes these portability concerns non-existent,
logically, either this variable should be removed,
or every *...-ansi* variable should have a corresponding *warn-...* variable
   (e.g., warn on (pathname "foo:bar") and (print #\space))
or there should be a single variable *enable-portability-warnings*
the last option is best: simple and inobtrusive.
(setq *ansi* t) should also disable *enable-portability-warnings*.
initially, *warn-on-floating-point-contagion* and *enable-portability-warnings*
should be T.


Cannot remove symbolic links to directories
with either DELETE-FILE or DELETE-DIR.
<http://article.gmane.org/gmane.lisp.clisp.devel:13452>
the case of symlinks to directories, there is absolutely no way out.
you cannot get the name of the link,
you cannot delete it (with either delete-file or delete-dir),
and you cannot even probe it! (with either probe-file or probe-directory)
(note BTW a naming inconsistency between probe-directory and delete-dir).


--with-dynamic-modules ==> link errors:
gcc -I. -fPIC -x none -Wl,-export-dynamic -shared -o dynmod/lib-i18n.dll gettext.o
gettext.o(.text+0x7f):i18n/gettext.c:2296: undefined reference to `_I_to_L'
gettext.o(.text+0xcb):i18n/gettext.c:2301: undefined reference to `_STACK'
gettext.o(.text+0xd2):i18n/gettext.c:2301: undefined reference to `_symbol_tab_data'
These errors mean that the lisp run time must be compiled into lisp.so,
not lisp.a, and dynmod/lib-i18n.dll should be linked against it.
So, this all goes in the same direction as embeddability
and elimination of lisp.run in favor of a driver launching the right dll.
Also, --with-dynamic-modules should _NOT_ make BASE_MODULES dynamic.


libopen should DWIM -- include libltdl and use lt_dlopenext().
Note: including libltdl is non-trivial and libtoolize requires configure.ac


CLISP_LANGUAGE=german LC_ALL=C clisp -q -norc
> (/ 0)
=> endless loop
*** - Character #\u00F6 cannot be represented in the character set
      CHARSET:ASCII


Fix unintuitive behaviour of REPL introduced on 2003-01-21:
[1]> (read-char)
abcdef
#\a
[2]>
*** - EVAL: variable BCDEF has no value
Mögliche Optionen:
USE-VALUE      :R1      Sie dürfen einen Ersatzwertfür BDEF eingeben.
STORE-VALUE    :R2      Sie dürfen einen neuen Wert für BDEF eingeben.
ABORT          :R3      ABORT

Break 1 [3]>


Make the *KEYBOARD-STREAM* recognize all kinds of function keys in Linux
console and xterm again.


Reentrancy: remove the static filestatus variable from pathname.d.


Better error checking in get-setf-expansion for long defsetf


Better error report when calling a foreign function with not enough
arguments. So far, it says it needs one more argument than it
was given, instead of telling the precise number the function wants.
E.g. > (linux:qsort)
*** - FFI::FOREIGN-CALL-OUT: Too few arguments (0 instead of at least 1) given
      to #<FOREIGN-FUNCTION "qsort" #x080511C8>
A solution might involve moving the count-inarguments code from *.lisp
to foreign.d, see
http://sourceforge.net/tracker/index.php?func=detail&aid=654718&group_id=1355&atid=101355


FFI: passing structs by value:
http://clisp.cons.org/impnotes/dffi.html#ffi-struct-arg
foreign.d should signal an error when a struct is passed by value
foreign1.lisp should emit C code for a wrapper function
 that passes the struct by pointer
http://article.gmane.org/gmane.lisp.clisp.general:11075
http://sourceforge.net/tracker/index.php?func=detail&aid=1486583&group_id=1355&atid=351355


for Emacs: GC root macro, YETANOTHER macro
           [sds: what does this mean?]


(directory "/proc/self/*" :full t )
((#P"/proc/self/status" #P"/proc/self/status" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/mem" #P"/proc/self/mem" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/environ" #P"/proc/self/environ" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/cmdline" #P"/proc/self/cmdline" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/stat" #P"/proc/self/stat" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/statm" #P"/proc/self/statm" (6 44 0 1 4 1999) 0)
  (#P"/proc/self/maps" #P"/proc/self/maps" (6 44 0 1 4 1999) 0)


(directory "/proc/self/fd/*")

stat("/proc", {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
stat("/proc/self", {st_mode=S_IFDIR|0555, st_size=0, ...}) = 0
stat("/proc/self/fd", {st_mode=S_IFDIR|0500, st_size=0, ...}) = 0
stat("/proc/self/fd/.", {st_mode=S_IFDIR|0500, st_size=0, ...}) = 0
open("/proc/self/fd/.", O_RDONLY)       = 6
fcntl(6, F_SETFD, FD_CLOEXEC)           = 0
getdents(6, /* 9 entries */, 1024)      = 112
stat("/proc/self/fd/0", {st_mode=S_IFCHR|0620, st_rdev=makedev(3, 2), ...}) = 0
lstat("/proc/self/fd/0", {st_mode=S_IFLNK|0700, st_size=64, ...}) = 0
readlink("/proc/self/fd/0", "[0303]:39982", 64) = 12
readlink("/proc/self/fd/0", "[0303]:39982", 12) = 12
readlink("/proc", 0xbfffd498, 4094)     = -1 EINVAL (Invalid argument)
readlink("/proc/self", "1108", 4094)    = 5
lstat("/proc/1108/fd/[0303]:39982", 0x8138260) = -1 ENOENT (No such file or directory)


Problem is the call to assure_dir_exists(TRUE,FALSE); where in fact not
all links are resolved in the directory!


print-readably for interpreted closures and #<specdecl> ?


parse_namestring: split_name_type(1), und bei make_pathname
Klein->Großschreibung umwandeln, so dass (make-pathname :type "fas") immer geht.


Should reject invalid format string:
> (format t "abc~,'*Sdef" '(a b c))
abc(A B C)def
> (format t (formatter "abc~,'*Sdef") '(a b c))
abc*(A B C)def
Similarly "abc~3W"
how about a compile-time format string check to warn about
 (format t "~A" a b) and (format t "~A ~B ~C" a b)


If you (TRACE generic-function) and then reload the file defining that
function, you got an error saying that #<CLOSURE TRACED-xxx> is not a generic
function. Now the function is silently untraced, which is just as bad. For
non-generic functions, DEFUN untraces the function, but at least gives a
warning. The desired behaviour is:
  1. no difference between non-generic and generic functions,
  2. a custom:*untrace-upon-redefinition* variable, default value nil,
  3. if custom:*untrace-upon-redefinition* is true, redefining a function
     removes the trace wrapper but gives a warning,
  4. if custom:*untrace-upon-redefinition* is false, redefining a function
     keeps the trace wrapper and replaces the "inner" function inside the
     wrapper.
A documented API should be provided for the trace wrapper, so that profilers
etc. can do the same as TRACE without needing to hack clisp's TRACE, COMPILE,
DISASSEMBLE, DEFUN, DEFGENERIC.
<http://article.gmane.org/gmane.lisp.clisp.devel:12837>
Such a def-fwrapper is what we need, so that profilers
etc. that are ported to ACL can be ported to clisp with minimal effort.
<http://www.franz.com/support/documentation/6.2/doc/fwrappers-and-advice.htm>


clean up and modularize spvw.d


Add fast unicode attribute accessors (like CHAR-NUMERIC-P and CHAR-DECOMPOSE)
[bruno] Agreed, this is a binding it my libunistring
and augment the reader to return 168 from
 (let ((*read-base* 16))
   (read-from-string
     (concatenate 'string
       (string #\MATHEMATICAL_BOLD_CAPITAL_A)
       (string #\MATHEMATICAL_BOLD_DIGIT_EIGHT))))
[bruno] better not. It was argued in comp.lang.lisp that the behaviour of
        the Lisp reader should remain based on the traditional notion of
        "digit". (ANSI CL is ambiguous in its definition of "digit". Unexpected
        behaviour in this area can lead to security issues.)


error recovery
<https://sourceforge.net/tracker/index.php?func=detail&aid=946463&group_id=1355&atid=351355>
The ability not to lose the computation on error is
crucial for any serious task.
Therefore, we need to be able to recover from ALL errors.
There are two independent issues here:
1. "retry" restart on i/o, see
<https://sourceforge.net/mailarchive/message.php?msg_id=7607066>
2. "return-from" restart that will return the specified
values from the given function on the stack. we
already have that in interpreted functions, but we must
also have it in compiled functions.
<http://thread.gmane.org/gmane.lisp.clisp.devel/11448>


GENERATIONAL_GC on Cygwin
<http://sourceforge.net/tracker/index.php?func=detail&aid=1104677&group_id=1355&atid=101355>
the mmap() autoconf test fails but maybe one can work around it?


GC/Memory Management:
1. Better post-mortem on segfault
   <http://article.gmane.org/gmane.lisp.clisp.devel:13211>
   <http://sourceforge.net/tracker/index.php?func=detail&aid=1565790&group_id=1355&atid=351355>
2. More GC generations as a build-time option
   <http://sourceforge.net/tracker/index.php?func=detail&aid=1565794&group_id=1355&atid=351355>
3. Better TIME/ROOM GC statistics reporting (per generation stats)
   <http://article.gmane.org/gmane.lisp.clisp.devel:13207>
   <http://article.gmane.org/gmane.lisp.clisp.devel:13209>
   <http://article.gmane.org/gmane.lisp.clisp.devel:13212>
   <http://sourceforge.net/tracker/index.php?func=detail&aid=1565795&group_id=1355&atid=351355>
4. Ability to "tenure" a specific object (with all its references!)
   or the whole current heap (also untenure the object).
   <http://sourceforge.net/tracker/index.php?func=detail&aid=1565801&group_id=1355&atid=351355>


local function (LABEL/FLET) calls waste heap (COPY-CLOSURE).
 use STACK instead?


LAUNCH/RUN-PROGRAM:
CLISP has now 6 built-ins for launching external executables:
SHELL, EXECUTE, MAKE-PIPE-*-STREAM, LAUNCH.
We need to have just _one_ such built-in, with many more options,
like CMUCL's RUN-PROGRAM
<http://common-lisp.net/project/cmucl/doc/cmu-user/extensions.html#toc46>
We now have LAUNCH which is a good start.  More work is needed.
[bruno]:
A possible outcome is just two facilities:
  1) A type PROCESS, which describes the result of an asynchronous launch,
     and a function which returns a process. (Similar to the Java API. The
     .NET API, where you set the input parameters through an object, is
     more confusing.) This is powerful but difficult to use.
  2) A simple to use facility which doesn't produce a PROCESS object, but
     rather returns stream(s) directly. Like RUN-PROGRAM with fewer options.
SHELL, EXECUTE, MAKE-PIPE-*-STREAM can then become shorthands for facility #2.
Of course, facility #2 should be based on facility #1, otherwise porting and
testing becomes twice as time consuming.
[sds]: (1) is an overkill, (2) is not needed if (1) is simplified:
 when :WAIT is T (synchronous launch), the return value is the exit code.
 when :WAIT is NIL (asynchronous), the return values are streams, and
      the number and type of the return values is determined by the
      :INPUT and :OUTPUT arguments.
   This is Lisp, not C or Java, and the return type can depend on arguments!
 (this, of course, can be implemented via several internal functions
 which even use a _C_ struct clisp_process, but the user does not need that)


Debugger:
- Think about the function call traces on the STACK and SP. Which stack
  should be used for which purpose?
- Turn the <n> entries of show-stack into normal entries, with equal rights
  as the other entries. Remove the <n>, as it's no information.
  * actually, these entries specify control flow (function calls)
    rather than local data (the other entries), they are "headers",
    so they should be printed differently and counted separately
- Devise a better variety of stack traces, maybe one without args and one with
  args?


MAYBE
=====

Still have to think about these, whether they are good ideas or not.


Try to increase the string size limit, to get rid of
"*** - string too long: desired length 6553600 exceeds the supported maximum
       length" errors
instead, the string size limit just went down!
http://sourceforge.net/tracker/index.php?func=detail&aid=1308440&group_id=1355&atid=101355


Add highlight/underline/etc. capability to the output side of *terminal-io*.


Optimize REPLACE so that (replace v v :start1 10 :end1 30 :start2 0 :end2 20)
doesn't cons. (vectors only, not lists)


CLOSE :ABORT T maybe do something on file streams.


Having two different handlings of Ctrl-C (once as a serious-condition, once
directly through tast_break(), bypassing ERROR and SIGNAL) is horrible.
Introduce a class of SERIOUS-CONDITION and then remove BATCHMODE-BREAK-DRIVER.


Distribute impnotes.html in the form of CLAHS, as an annotation against
Harlequin's HyperSpec. See http://clisp.cons.org/~haible/clahs/.


FFI: Actually support :language :cc/:gcc by passing appropriate flags
to the avcall/vacall macros. The default should be the same compiler
as the one used to compile clisp.


(read-char *keyboard-input*) is still not perfect. Under Linux, it
returns the following key sequences for some keystrokes:

Under xterm:
Shift F1 --> F11
Shift F2 --> F12
Shift F3 --> Escape [ 2 5 ~
Shift F4 --> Escape [ 2 6 ~
Shift F5 --> Escape [ 2 8 ~
Shift F6 --> Escape [ 2 9 ~
Shift F7 --> Escape [ 3 1 ~
Shift F8 --> Escape [ 3 2 ~
Shift F9 --> F9
Shift F10 --> F10
Shift F11 --> F11
Shift F12 --> F12

Under console:
Shift F1 --> F11
Shift F2 --> F12
Shift F3 --> Escape [ 2 5 ~
Shift F4 --> Escape [ 2 6 ~
Shift F5 --> Escape [ 2 8 ~
Shift F6 --> Escape [ 2 9 ~
Shift F7 --> Escape [ 3 1 ~
Shift F8 --> Escape [ 3 2 ~
Shift F9 --> Escape [ 3 3 ~
Shift F10 --> Escape [ 3 4 ~
Shift F11 --> F11
Shift F12 --> F12
Ctrl-Alt-Q --> nothing at all


Looks at GCL CLCS (debugger.lsp top-patches.lsp condition-definitions.lsp)
whether it contains something clisp could profit from.


Debugger (reploop.lisp, debug.d): Implement backtrace-? commands with a limit,
such that  BACKTRACE-1 20  will show the 20 newest frames. Good for people
without an xterm with logging. sys::debug-backtrace already takes a limit
as optional argument; the hard part is to make the debugger accept commands
with arguments, instead of just single-word commands.


More CLtL2 compliance:
; List of X3J13 votes that may be implemented in CLISP
;
; Number: from CLtL2, Index of X3J13 Votes.
; Status: indicates whether CLISP currently supports code that makes use of
; this vote.
;
; Number Title                               Status          Files affected
;
;  <24>  compiler diagnostics                no              compiler.lisp
;  <85>  FORMAT & pretty print               yes             format.lisp
;                                            no: ~E, ~F, ~G, ~$ also bind *PRINT-BASE* to 10 and *PRINT-RADIX* to NIL
; <110>  LOAD & objects                      no              loadform.lisp
; <127>  package function consistency        no              package.d
; <129>  pathname: component value           no              pathname.d
; <133>  pathname: subdirectory list         no              pathname.d
; <134>  pathname: symbol                    no              pathname.d, stream.d
; <136>  pathname: unspecific component      no              pathname.d
; <139>  pretty-print interface              no              xp.lisp
; <153>  REMF & destruction: unspecified     no for NRECONC  list.d
;                                            yes for anything else


The completion in user2.lisp does bad things with respect to upper case and
lower case of symbols. Assume the symbols |TMP|, |TMP structure class|
are interned. Then type a tab in the following situations:
> (symbol-plist '\Tmp
> 'tm
> '|


Optimize the compiler or the LOOP expander so that
  (macroexpand '(loop for (a b c) = (multiple-value-list (foo))))
contains
  (MULTIPLE-VALUE-SETQ (A B C) (FOO))


NOT TO DO
=========

These have been in my mind, but are probably bad ideas.


upgrade to newest CLN, but still big-endian and without gmp

intgcd

* Speed up the functions `cl_I gcd (const cl_I&, const cl_I&)' and
  `cl_I xgcd (const cl_I&, const cl_I&, cl_I*, cl_I*)' by a factor
  of 1.5. Simple trick.
Why is there no cl_gcd_double_threshold in xgcd, only in gcd??

lfloat

* Speed up multiplication and division of large long-floats by small integers.

realtran

* Speed up the internal computation of ln(2) and ln(10) by about 20%.

* Speed up the computation of exp (for 1000 digits or more) by about 10%.

* Speed up the computation of the elementary transcendental functions:
  exp, sinh, cosh, asinh, atanh (for 100 digits) by about 30%,
  log (for 100 digits) by about 25%,
  sin, cos (for 1000 or 10000 digits) by about 20%,
  asin, acos, atan (for 1000 digits) by about 25%.

* Speed up the computation of pi, e, exp, log, sin, cos, tan, sinh, cosh, tanh,
  asin, acos, atan, asinh, acosh, atanh.
  All of them are now computed with asymptotically fast algorithms, of
  bit complexity O(log(N)^2*M(N)) = O(N^(1+epsilon)) for N digits.

It's not worth the effort: Most people really interested in fast numerics are
already using CLN, NTL, or a similar library. CLISP is already the champion
among the Common Lisp implementations regarding bignum performance.


Block the signals SIGHUP SIGPIPE SIGTERM SIGXCPU SIGXFSZ during GC, and
when they occur outside of GC, do a quit(1); .

Programs that deal too much with signals tend to be unfriendly. I hate
programs which stay alive when I try to kill them.


CLtL2 issue
; <162>  sharp-comma confusion               no              io.d

Backward compatibility to CLtL2. Doesn't cost much.


CLtL2 issue
; <174>  syntactic environment access        no

It never made it into ANSI CL.


CLtL2 issue
; <176>  :TEST-NOT, -IF-NOT                  no              sequence.d, list.d

It's a bad idea because REMOVE-IF-NOT is a kind of "positive filter", which
is more confusing when expressed as REMOVE of COMPLEMENT. Also, COMPLEMENT
is harder to understand than XXX-IF-NOT because COMPLEMENT is a function
returning a function.

