/[cmucl]/src/README.1st
ViewVC logotype

Contents of /src/README.1st

Parent Directory Parent Directory | Revision Log Revision Log


Revision 1.1 - (show annotations)
Sat Jan 18 14:32:04 1997 UTC (17 years, 3 months ago) by ram
Branch: MAIN
Werkowskis source kit 1.03.7
1 January 5, 1997
2
3 This is release 1.03.7 of the X86 port of CMUCL.
4
5 The source kit should work on any port. The kit contains
6 port specific directories only for X86. The best way to
7 use the source kit is to shadow the original 17f tree
8 using the EXTENTIONS:SEARCH-LIST feature.
9
10 Changes since the 1.03.6 kit (23 November 96) include:
11
12 Generic CMUCL:
13
14 1. Ray Toy has provided support for logical-pathname
15 namestrings in more (all?) IO functions that take
16 logical-pathname arguments. Probe-file and load
17 with LP namestrings work.
18
19 2. A bug in TRANSLATE-PATHNAME which caused recursive
20 logical-pathname translations to fail is fixed.
21
22 3. A few bugs were flushed and fixed by Douglas Crosher
23 while working the CL-HTTP port.
24
25 4. Ray Toy's enhancements to floating-point type
26 propagation seem real stable and really useful.
27 Add :propagate-float-type to *features* before compiling
28 a world.
29
30 5. A bug in file-author which signalled an error on
31 a pathname argument was fixed. Hemlock's change-log
32 function now works.
33
34 6. A few random bugs in Hemlock were fixed.
35
36 X86 specific changes:
37
38 0. Due to the following, there may be bootstrap problems
39 building this release. Please load p86/boot.lisp
40 prior to running worldcom/worldbuild/comcom. You
41 must build at least 3 generations of compiler before
42 all changes propagate into place. You will need the
43 boot.lisp file only for the first generation build.
44
45 1. Major improvements in debugger support thanks to
46 Douglas Crosher. Previous "unknown code location"
47 notes from debug with confusing debug data are now
48 replaced by useful and accurate information.
49
50 2. More math functions are open coded using Intel's
51 fancy instructions. Sin, Cos and Tan are fast
52 inlined if the argument range is known to be (mod 2^64).
53 Currently arguments outside this range return a zero
54 value. An efficiency note is printed if the compiler
55 can't optimize to fast code.
56
57 3. A problem where PCL CLOS sometimes left dangling
58 pointers in unscanned (by GC) memory has been fixed.
59
60 Bugs:
61
62 Compiler errors are sometimes generated from LABELS
63 forms which use optional arguments. A workaround is
64 to not use optional arguments or include all args at
65 the call site. This seems to be the major outstanding bug.
66
67 There is one known bug in restart-case which does not
68 want to work unless compiled to native code.
69
70 A mis-feature is that compiled code is allocated in static
71 memory. This restriction may go away in a future release.
72 The impact of static allocation of compiled code is that
73 any dead code is not reclaimed and contributes to VM bloat.
74 This is an operational problem only in intensive incremental
75 code development.
76
77 Please send comments and bugs to
78 Paul Werkowski
79 pw@snoopy.mv.com
80
81 or to cmucl-imp@cons.org
82
83 -----------------------------------------------------------------------
84 Informal Release History:
85
86 November 23, 1996. 1.03.6
87
88 Changes since release 1.03.1 include
89
90 * The FreeBSD/X86 source code has been merged with
91 the current CMU tree which had a number of unpublished
92 fixes applied since version 17f.
93
94 * Changes specific to the Linux port have been
95 merged as well. These changes don't effect FreeBSD users.
96
97 * Major improvements to the floating-point code generators
98 thanks to Douglas Crosher.
99
100 * The internal "static-space" has changed along with
101 some internal static variable names. You probably will
102 have to recompile everything. Sorry.
103
104 * The source kit now contains improvements to floating
105 point type propagation, thanks to Ray Toy. These
106 changes are still experimental, and disabled by default
107 in the binary distribution. When enabled, Python is able
108 to optimize the results of intermediate floating point
109 types through most arithmetic operations. This is useful
110 when bounded float types are used,
111 eg (declare (type (double-float 0d0) x y), then
112 (* x y) is known to be non-negative. See file srctran.lisp
113 for hints on how to enable this feature.
114
115 * Many internal changes and improvements too
116 numerous to remember.
117
118
119 Changes since release 1.03 include:
120
121 * A nasty little bug was found in floating-point
122 register loading that caused erroneous data to
123 be created in some conditions. It is now fixed thanks
124 to Douglas Crosher.
125
126 Changes since release 1.02 include:
127
128 * Fixed some obscure bugs in the Object System.
129 Non-standard method-combination now works. These
130 changes are platform-independent.
131
132 * Fixed a problem in the definition of the arithmetic-error
133 condition which caused bad behavior with divide-by-zero
134 and floating-point exceptions.
135
136 * Changes made by Peter Van Eynde for the first Linux release
137 have been merged into the current source tree.
138
139 * Improved version of some irrational functions are included
140 thanks to Raymond Toy.
141
142 * A few random bug fixes here and there.
143
144 Changes since release 1.01 include:
145
146 * Compiled library files for the major subsystems,
147 while not in the basic lisp image, are provided
148 separately on the www.mv.com/users/pw/lisp/dl
149 page. You can download what you want and use
150 bin/config to build a custom system.
151
152 * Fixed problem with handling interrupts occuring
153 during execution of forms enclosed by
154 (without-interrupts ..), for example, interrupting
155 a garbage collection and then continuing caused
156 a bus error or segment violation.
157
158 * Some random optimizations.
159
160 * Reorganized the way foreign symbols are handled.
161 If you get "Unknown foreign symbol" warnings you
162 will need to recompile some code.
163
164 Changes since release 1.0 include:
165
166 * Fix to floating-point code generator that
167 caused incorrect results from FADD or FMUL when
168 both operands were in same register.
169
170 * Support for auto-gc-trigger now notifies
171 lisp to consider whether or not to initiate GC.
172
173 * Fix to save-lisp C support code which lost
174 the restart function address in some cases.
175
176 * Improvements to platform specific parts of
177 debug-internals allow better stack backtracking,
178 with less hackery.
179
180 * Includes all known patches from CMU. The bug
181 which caused the Motif inspector grief is now gone,
182 and the inspector is actually usable and useful.
183
184

  ViewVC Help
Powered by ViewVC 1.1.5