| blob | 6f57db117647407369b0059a8a91dc12c4454821 |
1 The collector has at various times been compiled under Windows 95 & later, NT,
2 and XP, with the original Microsoft SDK, with Visual C++ 2.0, 4.0, and 6, with
3 the GNU win32 tools, with Borland 4.5, with Watcom C, and recently
4 with the Digital Mars compiler. It is likely that some of these have been
5 broken in the meantime. Patches are appreciated.
7 For historical reasons,
8 the collector test program "gctest" is linked as a GUI application,
9 but does not open any windows. Its output appears in the file
10 "gc.log". It may be started from the file manager. The hour glass
11 cursor may appear as long as it's running. If it is started from the
12 command line, it will usually run in the background. Wait a few
13 minutes (a few seconds on a modern machine) before you check the output.
14 You should see either a failure indication or a "Collector appears to
15 work" message.
17 The cord test program has not been ported (but should port
18 easily). A toy editor (cord/de.exe) based on cords (heavyweight
19 strings represented as trees) has been ported and is included.
20 It runs fine under either win32 or win32S. It serves as an example
21 of a true Windows application, except that it was written by a
22 nonexpert Windows programmer. (There are some peculiarities
23 in the way files are displayed. The <cr> is displayed explicitly
24 for standard DOS text files. As in the UNIX version, control
25 characters are displayed explicitly, but in this case as red text.
26 This may be suboptimal for some tastes and/or sets of default
27 window colors.)
29 In general -DREDIRECT_MALLOC is unlikely to work unless the
30 application is completely statically linked.
32 The collector normally allocates memory from the OS with VirtualAlloc.
33 This appears to cause problems under Windows NT and Windows 2000 (but
34 not Windows 95/98) if the memory is later passed to CreateDIBitmap.
35 To work around this problem, build the collector with -DUSE_GLOBAL_ALLOC.
36 This is currently incompatible with -DUSE_MUNMAP. (Thanks to Jonathan
37 Clark for tracking this down. There's some chance this may be fixed
38 in 6.1alpha4, since we now separate heap sections with an unused page.)
40 Microsoft Tools
41 ---------------
42 For Microsoft development tools, rename NT_MAKEFILE as
43 MAKEFILE. (Make sure that the CPU environment variable is defined
44 to be i386.) In order to use the gc_cpp.h C++ interface, all
45 client code should include gc_cpp.h.
47 For historical reasons,
48 the collector test program "gctest" is linked as a GUI application,
49 but does not open any windows. Its output appears in the file
50 "gc.log". It may be started from the file manager. The hour glass
51 cursor may appear as long as it's running. If it is started from the
52 command line, it will usually run in the background. Wait a few
53 minutes (a few seconds on a modern machine) before you check the output.
54 You should see either a failure indication or a "Collector appears to
55 work" message.
57 If you would prefer a VC++.NET project file, ask boehm@acm.org. One has
58 been contributed, but it seems to contain some absolute paths etc., so
59 it can presumably only be a starting point, and is not in the standard
60 distribution. It is unclear (to me, Hans Boehm) whether it is feasible to
61 change that.
63 Clients may need to define GC_NOT_DLL before including gc.h, if the
64 collector was built as a static library (as it normally is in the
65 absence of thread support).
67 GNU Tools
68 ---------
69 For GNU-win32, use the regular makefile, possibly after uncommenting
70 the line "include Makefile.DLLs". The latter should be necessary only
71 if you want to package the collector as a DLL.
72 [Is the following sentence obsolete? -HB] The GNU-win32 port is
73 believed to work only for b18, not b19, probably due to linker changes
74 in b19. This is probably fixable with a different definition of
75 DATASTART and DATAEND in gcconfig.h.
77 The collector should also be buildable under Cygwin with either the
78 old standard Makefile, or with the "configure;make" machinery.
80 Borland Tools
81 -------------
82 [Rarely tested.]
83 For Borland tools, use BCC_MAKEFILE. Note that
84 Borland's compiler defaults to 1 byte alignment in structures (-a1),
85 whereas Visual C++ appears to default to 8 byte alignment (/Zp8).
86 The garbage collector in its default configuration EXPECTS AT
87 LEAST 4 BYTE ALIGNMENT. Thus the BORLAND DEFAULT MUST
88 BE OVERRIDDEN. (In my opinion, it should usually be anyway.
89 I expect that -a1 introduces major performance penalties on a
90 486 or Pentium.) Note that this changes structure layouts. (As a last
91 resort, gcconfig.h can be changed to allow 1 byte alignment. But
92 this has significant negative performance implications.)
93 The Makefile is set up to assume Borland 4.5. If you have another
94 version, change the line near the top. By default, it does not
95 require the assembler. If you do have the assembler, I recommend
96 removing the -DUSE_GENERIC.
98 Incremental Collection
99 ----------------------
100 There is some support for incremental collection. This is
101 currently pretty simple-minded. Pages are protected. Protection
102 faults are caught by a handler installed at the bottom of the handler
103 stack. This is both slow and interacts poorly with a debugger.
104 Whenever possible, I recommend adding a call to
105 GC_enable_incremental at the last possible moment, after most
106 debugging is complete. Unlike the UNIX versions, no system
107 calls are wrapped by the collector itself. It may be necessary
108 to wrap ReadFile calls that use a buffer in the heap, so that the
109 call does not encounter a protection fault while it's running.
110 (As usual, none of this is an issue unless GC_enable_incremental
111 is called.)
113 Note that incremental collection is disabled with -DSMALL_CONFIG.
115 Threads
116 -------
118 James Clark has contributed the necessary code to support win32 threads
119 with the collector in a DLL.
120 Use NT_THREADS_MAKEFILE (a.k.a gc.mak) instead of NT_MAKEFILE
121 to build this version. Note that this requires some files whose names
122 are more than 8 + 3 characters long. Thus you should unpack the tar file
123 so that long file names are preserved. To build the garbage collector
124 test with VC++ from the command line, use
126 nmake /F ".\gc.mak" CFG="gctest - Win32 Release"
128 This requires that the subdirectory gctest\Release exist.
129 The test program and DLL will reside in the Release directory.
131 This version relies on the collector residing in a dll.
133 This version currently supports incremental collection only if it is
134 enabled before any additional threads are created.
136 Since 6.3alpha2, threads are also better supported in static library builds
137 with Microsoft tools (use NT_STATIC_THREADS_MAKEFILE) and with the GNU
138 tools. In all cases,the collector must be built with GC_WIN32_THREADS
139 defined, even if the Cygwin pthreads interface is used.
140 (NT_STATIC_THREADS_MAKEFILE does this implicitly. Under Cygwin,
141 ./configure --enable-threads=posix defines GC_WIN32_THREADS.) Threads must be
142 created with GC_CreateThread. This can be accomplished by
143 including gc.h and then calling CreateThread, which is redefined
144 by gc.h.
146 For the statically linked versions, it is required that GC_init()
147 be called before other GC calls, since there seems to be no implicit way
148 to initialize the allocation lock. The easiest way to ensure this in
149 portable code is to call GC_INIT() from the main executable (not
150 a dynamic library) before calling any other GC_ routines.
152 We strongly advise against using the TerminateThread() win32 API call,
153 especially with the garbage collector. Any use is likely to provoke a
154 crash in the GC, since it makes it impossible for the collector to
155 correctly track threads.
158 Watcom compiler
159 ---------------
161 Ivan V. Demakov's README for the Watcom port:
163 The collector has been compiled with Watcom C 10.6 and 11.0.
164 It runs under win32, win32s, and even under msdos with dos4gw
165 dos-extender. It should also run under OS/2, though this isn't
166 tested. Under win32 the collector can be built either as dll
167 or as static library.
169 Note that all compilations were done under Windows 95 or NT.
170 For unknown reason compiling under Windows 3.11 for NT (one
171 attempt has been made) leads to broken executables.
173 Incremental collection is not supported.
175 cord is not ported.
177 Before compiling you may need to edit WCC_MAKEFILE to set target
178 platform, library type (dynamic or static), calling conventions, and
179 optimization options.
181 To compile the collector and testing programs use the command:
182 wmake -f WCC_MAKEFILE
184 All programs using gc should be compiled with 4-byte alignment.
185 For further explanations on this see comments about Borland.
187 If the gc is compiled as dll, the macro ``GC_DLL'' should be defined before
188 including "gc.h" (for example, with -DGC_DLL compiler option). It's
189 important, otherwise resulting programs will not run.
191 Ivan Demakov (email: ivan@tgrad.nsk.su)
193 Win32S
194 ------
196 [The following is probably obsolete. The win32s support is still in the
197 collector, but I doubt anyone cares, or has tested it recently.]
199 The collector runs under both win32s and win32, but with different semantics.
200 Under win32, all writable pages outside of the heaps and stack are
201 scanned for roots. Thus the collector sees pointers in DLL data
202 segments. Under win32s, only the main data segment is scanned.
203 (The main data segment should always be scanned. Under some
204 versions of win32s, other regions may also be scanned.)
205 Thus all accessible objects should be accessible from local variables
206 or variables in the main data segment. Alternatively, other data
207 segments (e.g. in DLLs) may be registered with the collector by
208 calling GC_init() and then GC_register_root_section(a), where
209 a is the address of some variable inside the data segment. (Duplicate
210 registrations are ignored, but not terribly quickly.)
212 (There are two reasons for this. We didn't want to see many 16:16
213 pointers. And the VirtualQuery call has different semantics under
214 the two systems, and under different versions of win32s.)