ViewVC logotype

Diff of /src/general-info/release-19d.txt

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.17 by rtoy, Wed May 31 13:07:48 2006 UTC revision 1.18 by rtoy, Fri Jun 30 18:41:32 2006 UTC
# Line 52  New in this release: Line 52  New in this release:
52        - When a core file is started, any shared libraries that were        - When a core file is started, any shared libraries that were
53          loaded when the core was made will be reloaded on startup.  If          loaded when the core was made will be reloaded on startup.  If
54          the library cannot be found, several restarts are available.          the library cannot be found, several restarts are available.
55          - Support for the new float type EXT:DOUBLE-DOUBLE-FLOAT.  This
56            includes support for complex numbers with this component type
57            and simple arrays for this type and complex numbers of this
58            type.
60            A DOUBLE-DOUBLE-FLOAT uses two DOUBLE-FLOAT's to represent a
61            number with >= 106 bits of precision (about 33 digits).  Known
62            issues:
64            * If you are expecting IEEE-style behavior, you don't get it:
65              - signed zeroes aren't really available.
66              - overflows don't return infinity but return NaN instead.
67              - rounding might not be quite the same as IEEE
68              - SQRT is not accurate to the last bit, as required by IEEE.
69            * Multiplying by a number very close to
70              most-positive-double-float will produce an error even if the
71              result does not overflow.  (This is an artifact of how
72              multiplication is done.  I don't have a solution to this.)
73            * Read/write consistency is not working.  (Because conversion
74              from a bignum to a double-double-float doesn't really
75              understand the internal double-double-float format.)
76            * INTEGER-DECODE-FLOAT and SCALE-FLOAT aren't "inverses".
77              That is, you can't take the result of integer-decode-float
78              and use scale-float to produce exactly the same number.
79              This is because of how bignums are converted to
80              double-doubles.
81            * FLOAT-DIGITS always returns 106 even though there could be
82              more bits.  (Consider the double-double (1d0,1d-200)).  This
83              will show up in PRINT where the printed result will have way
84              more than the normal 33 digits or so.  But reading such a
85              number back won't give the same value.
86            * There is probably more consing than is necessary in many of
87              the standard Common Lisp functions like floor, ffloor, etc.
88            * The special functions are not fully tested.  I did a few
89              random spot checks for each function and compared the
90              results with maxima to verify them.
91            * The branch cuts for the special functions very likely will
92              not match the double-float versions, mostly because we don't
93              have working signed zeroes.
94            * Type derivation for double-double-floats might not be
95              working quite right.
96            * PI is still a double-float.  If you want a double-double
97              version of pi, it's KERNEL:DD-PI.  (Soon to be EXT:DD-PI.)
98            * All double-double operations are inlined by default (unless
99              *INLINE-EXPANSION-LIMIT* is reached), so functions will tend
100              to be very large.  (Each basic double-double operation takes
101              some 20 FP instructions).
102            * There are probably still many bugs where double-double-float
103              support was overlooked.
106    * Numerous ANSI compliance fixes:    * Numerous ANSI compliance fixes:

Removed from v.1.17  
changed lines
  Added in v.1.18

  ViewVC Help
Powered by ViewVC 1.1.5