Fix hash tables, #'member*, #'assoc*, #'eql compiler macros if bignums lisp/ChangeLog addition: 2010-01-24 Aidan Kehoe <kehoea@parhasard.net> Correct the semantics of #'member*, #'eql, #'assoc* in the presence of bignums; change the integerp byte code to fixnump semantics. * bytecomp.el (fixnump, integerp, byte-compile-integerp): Change the integerp byte code to fixnump; add a byte-compile method to integerp using fixnump and numberp and avoiding a funcall most of the time, since in the non-core contexts where integerp is used, it's mostly distinguishing between fixnums and things that are not numbers at all. * byte-optimize.el (side-effect-free-fns, byte-after-unbind-ops) (byte-compile-side-effect-and-error-free-ops): Replace the integerp bytecode with fixnump; add fixnump to the side-effect-free-fns. Add the other extended number type predicates to the list in passing. * obsolete.el (floatp-safe): Mark this as obsolete. * cl.el (eql): Go into more detail in the docstring here. Don't bother checking whether both arguments are numbers; one is enough, #'equal will fail correctly if they have distinct types. (subst): Replace a call to #'integerp (deciding whether to use #'memq or not) with one to #'fixnump. Delete most-positive-fixnum, most-negative-fixnum from this file; they're now always in C, so they can't be modified from Lisp. * cl-seq.el (member*, assoc*, rassoc*): Correct these functions in the presence of bignums. * cl-macs.el (cl-make-type-test): The type test for a fixnum is now fixnump. Ditch floatp-safe, use floatp instead. (eql): Correct this compiler macro in the presence of bignums. (assoc*): Correct this compiler macro in the presence of bignums. * simple.el (undo): Change #'integerp to #'fixnump here, since we use #'delq with the same value as ELT a few lines down. src/ChangeLog addition: 2010-01-24 Aidan Kehoe <kehoea@parhasard.net> Fix problems with #'eql, extended number types, and the hash table implementation; change the Bintegerp bytecode to fixnump semantics even on bignum builds, since #'integerp can have a fast implementation in terms of #'fixnump for most of its extant uses, but not vice-versa. * lisp.h: Always #include number.h; we want the macros provided in it, even if the various number types are not available. * number.h (NON_FIXNUM_NUMBER_P): New macro, giving 1 when its argument is of non-immediate number type. Equivalent to FLOATP if WITH_NUMBER_TYPES is not defined. * elhash.c (lisp_object_eql_equal, lisp_object_eql_hash): Use NON_FIXNUM_NUMBER_P in these functions, instead of FLOATP, giving more correct behaviour in the presence of the extended number types. * bytecode.c (Bfixnump, execute_optimized_program): Rename Bintegerp to Bfixnump; change its semantics to reflect the new name on builds with bignum support. * data.c (Ffixnump, Fintegerp, syms_of_data, vars_of_data): Always make #'fixnump available, even on non-BIGNUM builds; always implement #'integerp in this file, even on BIGNUM builds. Move most-positive-fixnum, most-negative-fixnum here from number.c, so they are Lisp constants even on builds without number types, and attempts to change or bind them error. Use the NUMBERP and INTEGERP macros even on builds without extended number types. * data.c (fixnum_char_or_marker_to_int): Rename this function from integer_char_or_marker_to_int, to better reflect the arguments it accepts. * number.c (Fevenp, Foddp, syms_of_number): Never provide #'integerp in this file. Remove #'oddp, #'evenp; their implementations are overridden by those in cl.el. * number.c (vars_of_number): most-positive-fixnum, most-negative-fixnum are no longer here. man/ChangeLog addition: 2010-01-23 Aidan Kehoe <kehoea@parhasard.net> Generally: be careful to say fixnum, not integer, when talking about fixed-precision integral types. I'm sure I've missed instances, both here and in the docstrings, but this is a decent start. * lispref/text.texi (Columns): Document where only fixnums, not integers generally, are accepted. (Registers): Remove some ancient char-int confoundance here. * lispref/strings.texi (Creating Strings, Creating Strings): Be more exact in describing where fixnums but not integers in general are accepted. (Creating Strings): Use a more contemporary example to illustrate how concat deals with lists including integers about #xFF. Delete some obsolete documentation on same. (Char Table Types): Document that only fixnums are accepted as values in syntax tables. * lispref/searching.texi (String Search, Search and Replace): Be exact in describing where fixnums but not integers in general are accepted. * lispref/range-tables.texi (Range Tables): Be exact in describing them; only fixnums are accepted to describe ranges. * lispref/os.texi (Killing XEmacs, User Identification) (Time of Day, Time Conversion): Be more exact about using fixnum where only fixed-precision integers are accepted. * lispref/objects.texi (Integer Type): Be more exact (and up-to-date) about the possible values for integers. Cross-reference to documentation of the bignum extension. (Equality Predicates): (Range Table Type): (Array Type): Use fixnum, not integer, to describe a fixed-precision integer. (Syntax Table Type): Correct some English syntax here. * lispref/numbers.texi (Numbers): Change the phrasing here to use fixnum to mean the fixed-precision integers normal in emacs. Document that our terminology deviates from that of Common Lisp, and that we're working on it. (Compatibility Issues): Reiterate the Common Lisp versus Emacs Lisp compatibility issues. (Comparison of Numbers, Arithmetic Operations): * lispref/commands.texi (Command Loop Info, Working With Events): * lispref/buffers.texi (Modification Time): Be more exact in describing where fixnums but not integers in general are accepted.

--- a/man/lispref/objects.texi	Wed Jan 20 17:30:29 2010 +0000
+++ b/man/lispref/objects.texi	Sun Jan 24 15:21:27 2010 +0000
@@ -290,31 +290,25 @@
 @node Integer Type
 @subsection Integer Type
 
-  The range of values for integers in XEmacs Lisp is @minus{}134217728 to
-134217727 (28 bits; i.e.,
-@ifinfo
--2**27
-@end ifinfo
-@tex
-$-2^{27}$
-@end tex
-to
-@ifinfo
-2**27 - 1)
-@end ifinfo
-@tex
-$2^{28}-1$)
-@end tex
-on most machines.  (Some machines, in particular 64-bit machines such as
-the DEC Alpha, may provide a wider range.)  It is important to note that
-the XEmacs Lisp arithmetic functions do not check for overflow.  Thus
-@code{(1+ 134217727)} is @minus{}134217728 on most machines. (However,
-you @emph{will} get an error if you attempt to read an out-of-range
-number using the Lisp reader.)
-
-  The read syntax for integers is a sequence of (base ten) digits with
-an optional sign at the beginning. (The printed representation produced
-by the Lisp interpreter never has a leading @samp{+}.)
+  In XEmacs Lisp, integers can be fixnums (that is, fixed-precision
+integers) or bignums (arbitrary-precision integers), if compile-time
+configuration supports this.  The read syntax for the two types is the
+same, the type chosen depending on the numeric values involved.
+
+  The range of values for fixnums in XEmacs Lisp is given by the
+constants @code{most-positive-fixnum} and @code{most-negative-fixnum}.
+On 32-bit machines, these constants reflect 31 value bits, ranging from
+@minus{}1073741824 to 1073741823.
+
+  In the absence of @xref{The Bignum Extension}, XEmacs Lisp
+arithmetic functions do not check for overflow; so the code snippet
+@code{(= most-negative-fixnum (1+ most-positive-fixnum))} will give
+@code{t}.  However, you @emph{will} get an error if you attempt to read
+an out-of-range number using the Lisp reader.
+
+  The main read syntax for integers is a sequence of base ten digits
+with an optional sign at the beginning. (The printed representation
+produced by the Lisp interpreter never has a leading @samp{+}.)
 
 @example
 @group
@@ -1022,7 +1016,7 @@
 read syntax; see @ref{String Type}, @ref{Vector Type}, and @ref{Bit
 Vector Type}.
 
-  An array may have any length up to the largest integer; but once
+  An array may have any length up to the largest fixnum; but once
 created, it has a fixed size.  The first element of an array has index
 zero, the second element has index 1, and so on.  This is called
 @dfn{zero-origin} indexing.  For example, an array of four elements has
@@ -1331,7 +1325,7 @@
 @subsection Range Table Type
 @cindex range table type
 
-  A @dfn{range table} is a table that maps from ranges of integers to
+  A @dfn{range table} is a table that maps from ranges of fixnums to
 arbitrary Lisp objects.  Range tables automatically combine overlapping
 ranges that map to the same Lisp object, and operations are provided
 for mapping over all of the ranges in a range table.
@@ -1714,7 +1708,7 @@
 @subsection Syntax Table Type
 
   Under XEmacs 20, a @dfn{syntax table} is a particular type of char
-table.  Under XEmacs 19, a syntax table a vector of 256 integers.  In
+table.  Under XEmacs 19, a syntax table is a vector of 256 integers.  In
 both cases, each element defines how one character is interpreted when it
 appears in a buffer.  For example, in C mode (@pxref{Major Modes}), the
 @samp{+} character is punctuation, but in Lisp mode it is a valid
@@ -2245,7 +2239,7 @@
 change in one will be reflected by the same change in the other.
 
 @code{eq} returns @code{t} if @var{object1} and @var{object2} are
-integers with the same value.  It is preferable to use @code{=} or
+fixnums with the same value.  It is preferable to use @code{=} or
 @code{eql} in many contexts for numeric comparison, especially since
 bignums (integers with values that would have otherwise overflowed, only
 available on some builds) with the same value are not @code{eq};