1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
|
------------------------------------------------------------------------------
NEWS for version 0.4.0
------------------------------------------------------------------------------
o work around for gcc 2.95.4 and 3.0.4 for non conforming
__builtin_return_address() (Edward Welbourne).
o new '--no-wrapper' option to ccmalloc script allows to disable C++ support
by not linking against the C++ wrapper.
o malloc in wrapper.c only fills the new allocated block with the magic
char 0x42 if the allocation of the block was successful. This allows to
stress test the memory system even with ccmalloc compiled in (like
calling 'malloc((1 << 31))').
------------------------------------------------------------------------------
NEWS for version 0.3.9
------------------------------------------------------------------------------
o fixed configure to work on FreeBSD
o patched configure to correctly set 'HAVE_ATEXIT' and also detect the path
of the default dynamically linked libc. This patch fixes some problems
with Redhat 7.1 which uses libc6.
o '--version' and '--help' for 'ccmalloc' script
o finally allowed 'free(0)'
------------------------------------------------------------------------------
NEWS for version 0.3.8
------------------------------------------------------------------------------
o checking statically allocated dynamic data for C++ seems to work again
o better support for C++: increased buffer size, better lexer for parsing
demangling info of C++ template symbols from gdb (Henning Moll).
------------------------------------------------------------------------------
NEWS for version 0.3.7
------------------------------------------------------------------------------
o fixed 'ccmalloc' script to skip files compiled with '-S' and '-c'
o fixed 'ccmalloc' script to filter out '-o <object-file>' as well
o do not log chain that contain a reference to 'ccmalloc_init'
------------------------------------------------------------------------------
NEWS for version 0.3.4
------------------------------------------------------------------------------
o Started to work on compatibility with Sun's workshop compiler CC.
The idea is to compile 'callchain.c' with 'gcc' and the 'C++' wrapper
'ccmalloc.cc' with 'CC'. Then it should be possible to link and compile
your own project with 'CC'. Just link as follows
CC -o binary your-obj1.o your-obj2.o ccmalloc-CC.o -lccmalloc -ldl
I also added a script 'ccmalloc' that hides this additional arguments to
the compiler and makes the usage of ccmalloc similar to the usage of
'purify'. Now the standard way of using ccmalloc with C++ becomes
ccmalloc CC -o binary your-obj1.o your-obj2.o
or with 'gcc'
ccmalloc gcc -o binary your-obj1.o your-obj2.o
However, for some unknown reason 'CC' still produces link errors and
this feature does not work yet.
o Added patches by Fritz Elfert for better command line argument detection
under Linux (and actually reenabled the whole detection code).
------------------------------------------------------------------------------
NEWS for version 0.3.3 (patches from Henning Moll)
------------------------------------------------------------------------------
o Fixed off by one bug in reporting line number
o Used 'set width 0' command for GDB to produce easier to parse one line
output. This removes some 'could not read info from gdb' messages, i.e.
with long symbol names as in STL.
o Skipped 'Current language: ...' message from GDB for better support of
multiple language debugging.
o Better heuristics for skipping library callchains.
------------------------------------------------------------------------------
NEWS for version 0.3.2
------------------------------------------------------------------------------
o Initialized 'logfile' properly
o Added '-x' for 'nm' on Solaris
o Forced 'gdb' not to read initialization file '.gdbinit'
------------------------------------------------------------------------------
NEWS for version 0.3.0
------------------------------------------------------------------------------
o Removed support for SunOS (only Solaris and Linux left)
o Added the ability for ccmalloc to report only the call chains for wasted
allocations, i.e. allocations which were not completely freed properly
(Ian Ishmael).
------------------------------------------------------------------------------
NEWS for version 0.2.4
------------------------------------------------------------------------------
o Moved `banner()' out of the initialization code. This fixed the `ctime'
bug: Because `banner()' calls `ctime' itself, and `ctime' might call
malloc, the memory potentially allocated by ctime must be public memory,
since it could be deallocated from a following call to `ctime'
originating outside from `ccmalloc'.
------------------------------------------------------------------------------
NEWS for version 0.2.3
------------------------------------------------------------------------------
new features:
=============
o Change the cmp_CallChains_by_*_for_Qsort functions to not return 0 so
often by comparing them on address data, too. (This means we have to
read in the data from gdb before sorting the callchains.) This allows
one to compare resulting log files from different runs of the same
program with, e.g. diff or emacs ediff, without getting too many
irrelevant differences (Johannes).
o Last, but not least, read address data for dynamic libraries from gdb
as well. Only works for those dynamic libraries that the program is
linked with, not the ones that are dlopen'ed explicitly. Controlled
by a flag called `load-dynlibs' (Johannes).
o new flag `align-8-byte' to align memory to 8 byte on Linux.
bugfixes:
=========
o The configure script now tries to find the standard C library (libc.so)
and checks if it can be opened with dlopen. This should make ccmalloc
run on Redhat 5.0 Linux out of the box.
o GPL ;-)
o Fix the src/Makefile so that it also installs the config file and
so that install depends on ccmalloc.o (Johannes).
------------------------------------------------------------------------------
NEWS for version 0.2.2
------------------------------------------------------------------------------
bug fixes:
==========
o `wrapper.c' did not handle realloc correctly while in state INITIALIZING.
This lead to `*** realloc(0x8065f3d) called with non valid argument'
before the program was even started. This bug occured when using KDE
together with Qt. Now even the static allocator in wrapper.c saves the
size of the allocated data.
------------------------------------------------------------------------------
NEWS for version 0.2.1
------------------------------------------------------------------------------
bug fixes:
==========
o returned data is now 8 byte aligned on SunOS and Solaris. This avoids
a Bus Error when allocating heap memory for doubles.
(a lot of people reported that bug)
o shrinking allocated data with realloc works now
(a lot of people reported that bug)
o fixed dereferencing bug in cmpAddr (Johannes Keukelaar)
o calls fflush before killing itself
new features:
=============
o `only-log-chain' and `dont-log-chain' (Johannes Keukelaar)
o `logpid' (Didier Remy)
------------------------------------------------------------------------------
(C) 1997-2003 Armin Biere
$Id: NEWS,v 1.19 2003/02/03 08:03:54 biere Exp $
------------------------------------------------------------------------------
|
