Annotation of OpenXM_contrib2/asir2000/gc/README, Revision 1.2
1.1 noro 1: Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
2: Copyright (c) 1991-1996 by Xerox Corporation. All rights reserved.
1.2 ! noro 3: Copyright (c) 1996-1999 by Silicon Graphics. All rights reserved.
! 4: Copyright (c) 1999 by Hewlett-Packard Company. All rights reserved.
1.1 noro 5:
6: THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
7: OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
8:
9: Permission is hereby granted to use or copy this program
10: for any purpose, provided the above notices are retained on all copies.
11: Permission to modify the code and to distribute modified code is granted,
12: provided the above notices are retained, and a notice that the code was
13: modified is included with the above copyright notice.
14:
1.2 ! noro 15: This is version 5.0alpha4 of a conservative garbage collector for C and C++.
1.1 noro 16:
17: You might find a more recent version of this at
18:
1.2 ! noro 19: http://www.hpl.hp.com/personal/Hans_Boehm/gc
1.1 noro 20:
21: HISTORY -
22:
23: Early versions of this collector were developed as a part of research
24: projects supported in part by the National Science Foundation
25: and the Defense Advance Research Projects Agency.
1.2 ! noro 26: Much of the code was rewritten by Hans-J. Boehm (boehm@acm.org) at Xerox PARC
! 27: and at SGI.
1.1 noro 28:
29: Some other contributors:
30:
31: More recent contributors are mentioned in the modification history at the
32: end of this file. My apologies for any omissions.
33:
34: The SPARC specific code was contributed by Mark Weiser
35: (weiser@parc.xerox.com). The Encore Multimax modifications were supplied by
36: Kevin Kenny (kenny@m.cs.uiuc.edu). The adaptation to the RT is largely due
37: to Vernon Lee (scorpion@rice.edu), on machines made available by IBM.
38: Much of the HP specific code and a number of good suggestions for improving the
39: generic code are due to Walter Underwood (wunder@hp-ses.sde.hp.com).
40: Robert Brazile (brazile@diamond.bbn.com) originally supplied the ULTRIX code.
41: Al Dosser (dosser@src.dec.com) and Regis Cridlig (Regis.Cridlig@cl.cam.ac.uk)
42: subsequently provided updates and information on variation between ULTRIX
43: systems. Parag Patel (parag@netcom.com) supplied the A/UX code.
44: Jesper Peterson(jep@mtiame.mtia.oz.au), Michel Schinz, and
45: Martin Tauchmann (martintauchmann@bigfoot.com) supplied the Amiga port.
46: Thomas Funke (thf@zelator.in-berlin.de(?)) and
47: Brian D.Carlstrom (bdc@clark.lcs.mit.edu) supplied the NeXT ports.
48: Douglas Steel (doug@wg.icl.co.uk) provided ICL DRS6000 code.
49: Bill Janssen (janssen@parc.xerox.com) supplied the SunOS dynamic loader
50: specific code. Manuel Serrano (serrano@cornas.inria.fr) supplied linux and
51: Sony News specific code. Al Dosser provided Alpha/OSF/1 code. He and
52: Dave Detlefs(detlefs@src.dec.com) also provided several generic bug fixes.
53: Alistair G. Crooks(agc@uts.amdahl.com) supplied the NetBSD and 386BSD ports.
54: Jeffrey Hsu (hsu@soda.berkeley.edu) provided the FreeBSD port.
55: Brent Benson (brent@jade.ssd.csd.harris.com) ported the collector to
56: a Motorola 88K processor running CX/UX (Harris NightHawk).
57: Ari Huttunen (Ari.Huttunen@hut.fi) generalized the OS/2 port to
58: nonIBM development environments (a nontrivial task).
59: Patrick Beard (beard@cs.ucdavis.edu) provided the initial MacOS port.
60: David Chase, then at Olivetti Research, suggested several improvements.
61: Scott Schwartz (schwartz@groucho.cse.psu.edu) supplied some of the
62: code to save and print call stacks for leak detection on a SPARC.
63: Jesse Hull and John Ellis supplied the C++ interface code.
64: Zhong Shao performed much of the experimentation that led to the
65: current typed allocation facility. (His dynamic type inference code hasn't
66: made it into the released version of the collector, yet.)
67: (Blame for misinstallation of these modifications goes to the first author,
68: however.)
69:
70: OVERVIEW
71:
72: This is intended to be a general purpose, garbage collecting storage
73: allocator. The algorithms used are described in:
74:
75: Boehm, H., and M. Weiser, "Garbage Collection in an Uncooperative Environment",
76: Software Practice & Experience, September 1988, pp. 807-820.
77:
78: Boehm, H., A. Demers, and S. Shenker, "Mostly Parallel Garbage Collection",
79: Proceedings of the ACM SIGPLAN '91 Conference on Programming Language Design
80: and Implementation, SIGPLAN Notices 26, 6 (June 1991), pp. 157-164.
81:
82: Boehm, H., "Space Efficient Conservative Garbage Collection", Proceedings
83: of the ACM SIGPLAN '91 Conference on Programming Language Design and
84: Implementation, SIGPLAN Notices 28, 6 (June 1993), pp. 197-206.
85:
86: Possible interactions between the collector and optimizing compilers are
87: discussed in
88:
89: Boehm, H., and D. Chase, "A Proposal for GC-safe C Compilation",
90: The Journal of C Language Translation 4, 2 (December 1992).
91:
92: and
93:
94: Boehm H., "Simple GC-safe Compilation", Proceedings
95: of the ACM SIGPLAN '96 Conference on Programming Language Design and
96: Implementation.
97:
98: (Both are also available from
99: http://reality.sgi.com/boehm/papers/, among other places.)
100:
101: Unlike the collector described in the second reference, this collector
102: operates either with the mutator stopped during the entire collection
103: (default) or incrementally during allocations. (The latter is supported
104: on only a few machines.) It does not rely on threads, but is intended
105: to be thread-safe.
106:
107: Some of the ideas underlying the collector have previously been explored
108: by others. (Doug McIlroy wrote a vaguely similar collector that is part of
109: version 8 UNIX (tm).) However none of this work appears to have been widely
110: disseminated.
111:
112: Rudimentary tools for use of the collector as a leak detector are included, as
113: is a fairly sophisticated string package "cord" that makes use of the collector.
114: (See cord/README.)
115:
116:
117: GENERAL DESCRIPTION
118:
119: This is a garbage collecting storage allocator that is intended to be
120: used as a plug-in replacement for C's malloc.
121:
122: Since the collector does not require pointers to be tagged, it does not
123: attempt to ensure that all inaccessible storage is reclaimed. However,
124: in our experience, it is typically more successful at reclaiming unused
125: memory than most C programs using explicit deallocation. Unlike manually
126: introduced leaks, the amount of unreclaimed memory typically stays
127: bounded.
128:
129: In the following, an "object" is defined to be a region of memory allocated
130: by the routines described below.
131:
132: Any objects not intended to be collected must be pointed to either
133: from other such accessible objects, or from the registers,
134: stack, data, or statically allocated bss segments. Pointers from
135: the stack or registers may point to anywhere inside an object.
136: The same is true for heap pointers if the collector is compiled with
137: ALL_INTERIOR_POINTERS defined, as is now the default.
138:
139: Compiling without ALL_INTERIOR_POINTERS may reduce accidental retention
140: of garbage objects, by requiring pointers from the heap to to the beginning
141: of an object. But this no longer appears to be a significant
142: issue for most programs.
143:
144: There are a number of routines which modify the pointer recognition
145: algorithm. GC_register_displacement allows certain interior pointers
146: to be recognized even if ALL_INTERIOR_POINTERS is nor defined.
147: GC_malloc_ignore_off_page allows some pointers into the middle of large objects
148: to be disregarded, greatly reducing the probablility of accidental
149: retention of large objects. For most purposes it seems best to compile
150: with ALL_INTERIOR_POINTERS and to use GC_malloc_ignore_off_page if
151: you get collector warnings from allocations of very large objects.
152: See README.debugging for details.
153:
154: Note that pointers inside memory allocated by the standard "malloc" are not
155: seen by the garbage collector. Thus objects pointed to only from such a
156: region may be prematurely deallocated. It is thus suggested that the
157: standard "malloc" be used only for memory regions, such as I/O buffers, that
158: are guaranteed not to contain pointers to garbage collectable memory.
159: Pointers in C language automatic, static, or register variables,
160: are correctly recognized. (Note that GC_malloc_uncollectable has semantics
161: similar to standard malloc, but allocates objects that are traced by the
162: collector.)
163:
164: The collector does not always know how to find pointers in data
165: areas that are associated with dynamic libraries. This is easy to
166: remedy IF you know how to find those data areas on your operating
167: system (see GC_add_roots). Code for doing this under SunOS, IRIX 5.X and 6.X,
168: HP/UX, Alpha OSF/1, Linux, and win32 is included and used by default. (See
169: README.win32 for win32 details.) On other systems pointers from dynamic
170: library data areas may not be considered by the collector.
171:
172: Note that the garbage collector does not need to be informed of shared
173: read-only data. However if the shared library mechanism can introduce
174: discontiguous data areas that may contain pointers, then the collector does
175: need to be informed.
176:
177: Signal processing for most signals may be deferred during collection,
178: and during uninterruptible parts of the allocation process. Unlike
179: standard ANSI C mallocs, it can be safe to invoke malloc
180: from a signal handler while another malloc is in progress, provided
181: the original malloc is not restarted. (Empirically, many UNIX
182: applications already assume this.) To obtain this level of signal
183: safety, remove the definition of -DNO_SIGNALS in Makefile. This incurs
184: a minor performance penalty, and hence is no longer the default.
185:
186: The allocator/collector can also be configured for thread-safe operation.
187: (Full signal safety can also be achieved, but only at the cost of two system
188: calls per malloc, which is usually unacceptable.)
189:
190: INSTALLATION AND PORTABILITY
191:
192: As distributed, the macro SILENT is defined in Makefile.
193: In the event of problems, this can be removed to obtain a moderate
194: amount of descriptive output for each collection.
195: (The given statistics exhibit a few peculiarities.
196: Things don't appear to add up for a variety of reasons, most notably
197: fragmentation losses. These are probably much more significant for the
198: contrived program "test.c" than for your application.)
199:
200: Note that typing "make test" will automatically build the collector
201: and then run setjmp_test and gctest. Setjmp_test will give you information
202: about configuring the collector, which is useful primarily if you have
203: a machine that's not already supported. Gctest is a somewhat superficial
204: test of collector functionality. Failure is indicated by a core dump or
205: a message to the effect that the collector is broken. Gctest takes about
206: 35 seconds to run on a SPARCstation 2. On a slower machine,
207: expect it to take a while. It may use up to 8 MB of memory. (The
208: multi-threaded version will use more.) "Make test" will also, as
209: its last step, attempt to build and test the "cord" string library.
210: This will fail without an ANSI C compiler.
211:
212: The Makefile will generate a library gc.a which you should link against.
213: Typing "make cords" will add the cord library to gc.a.
214: Note that this requires an ANSI C compiler.
215:
216: It is suggested that if you need to replace a piece of the collector
217: (e.g. GC_mark_rts.c) you simply list your version ahead of gc.a on the
218: work.)
219: ld command line, rather than replacing the one in gc.a. (This will
220: generate numerous warnings under some versions of AIX, but it still
221: works.)
222:
223: All include files that need to be used by clients will be put in the
224: include subdirectory. (Normally this is just gc.h. "Make cords" adds
225: "cord.h" and "ec.h".)
226:
227: The collector currently is designed to run essentially unmodified on
228: machines that use a flat 32-bit or 64-bit address space.
229: That includes the vast majority of Workstations and X86 (X >= 3) PCs.
230: (The list here was deleted because it was getting too long and constantly
231: out of date.)
232: It does NOT run under plain 16-bit DOS or Windows 3.X. There are however
233: various packages (e.g. win32s, djgpp) that allow flat 32-bit address
234: applications to run under those systemsif the have at least an 80386 processor,
235: and several of those are compatible with the collector.
236:
237: In a few cases (Amiga, OS/2, Win32, MacOS) a separate makefile
238: or equivalent is supplied. Many of these have separate README.system
239: files.
240:
241: Dynamic libraries are completely supported only under SunOS
242: (and even that support is not functional on the last Sun 3 release),
243: IRIX 5&6, HP-PA, Win32 (not Win32S) and OSF/1 on DEC AXP machines.
244: On other machines we recommend that you do one of the following:
245:
246: 1) Add dynamic library support (and send us the code).
247: 2) Use static versions of the libraries.
248: 3) Arrange for dynamic libraries to use the standard malloc.
249: This is still dangerous if the library stores a pointer to a
250: garbage collected object. But nearly all standard interfaces
251: prohibit this, because they deal correctly with pointers
252: to stack allocated objects. (Strtok is an exception. Don't
253: use it.)
254:
255: In all cases we assume that pointer alignment is consistent with that
256: enforced by the standard C compilers. If you use a nonstandard compiler
257: you may have to adjust the alignment parameters defined in gc_priv.h.
258:
259: A port to a machine that is not byte addressed, or does not use 32 bit
260: or 64 bit addresses will require a major effort. A port to plain MSDOS
261: or win16 is hard.
262:
263: For machines not already mentioned, or for nonstandard compilers, the
264: following are likely to require change:
265:
266: 1. The parameters in gcconfig.h.
267: The parameters that will usually require adjustment are
268: STACKBOTTOM, ALIGNMENT and DATASTART. Setjmp_test
269: prints its guesses of the first two.
270: DATASTART should be an expression for computing the
271: address of the beginning of the data segment. This can often be
272: &etext. But some memory management units require that there be
273: some unmapped space between the text and the data segment. Thus
274: it may be more complicated. On UNIX systems, this is rarely
275: documented. But the adb "$m" command may be helpful. (Note
276: that DATASTART will usually be a function of &etext. Thus a
277: single experiment is usually insufficient.)
278: STACKBOTTOM is used to initialize GC_stackbottom, which
279: should be a sufficient approximation to the coldest stack address.
280: On some machines, it is difficult to obtain such a value that is
281: valid across a variety of MMUs, OS releases, etc. A number of
282: alternatives exist for using the collector in spite of this. See the
283: discussion in gcconfig.h immediately preceding the various
284: definitions of STACKBOTTOM.
285:
286: 2. mach_dep.c.
287: The most important routine here is one to mark from registers.
288: The distributed file includes a generic hack (based on setjmp) that
289: happens to work on many machines, and may work on yours. Try
290: compiling and running setjmp_t.c to see whether it has a chance of
291: working. (This is not correct C, so don't blame your compiler if it
292: doesn't work. Based on limited experience, register window machines
293: are likely to cause trouble. If your version of setjmp claims that
294: all accessible variables, including registers, have the value they
295: had at the time of the longjmp, it also will not work. Vanilla 4.2 BSD
296: on Vaxen makes such a claim. SunOS does not.)
297: If your compiler does not allow in-line assembly code, or if you prefer
298: not to use such a facility, mach_dep.c may be replaced by a .s file
299: (as we did for the MIPS machine and the PC/RT).
300: At this point enough architectures are supported by mach_dep.c
301: that you will rarely need to do more than adjust for assembler
302: syntax.
303:
304: 3. os_dep.c (and gc_priv.h).
305: Several kinds of operating system dependent routines reside here.
306: Many are optional. Several are invoked only through corresponding
307: macros in gc_priv.h, which may also be redefined as appropriate.
308: The routine GC_register_data_segments is crucial. It registers static
309: data areas that must be traversed by the collector. (User calls to
310: GC_add_roots may sometimes be used for similar effect.)
311: Routines to obtain memory from the OS also reside here.
312: Alternatively this can be done entirely by the macro GET_MEM
313: defined in gc_priv.h. Routines to disable and reenable signals
314: also reside here if they are need by the macros DISABLE_SIGNALS
315: and ENABLE_SIGNALS defined in gc_priv.h.
316: In a multithreaded environment, the macros LOCK and UNLOCK
317: in gc_priv.h will need to be suitably redefined.
318: The incremental collector requires page dirty information, which
319: is acquired through routines defined in os_dep.c. Unless directed
320: otherwise by gcconfig.h, these are implemented as stubs that simply
321: treat all pages as dirty. (This of course makes the incremental
322: collector much less useful.)
323:
324: 4. dyn_load.c
325: This provides a routine that allows the collector to scan data
326: segments associated with dynamic libraries. Often it is not
327: necessary to provide this routine unless user-written dynamic
328: libraries are used.
329:
330: For a different version of UN*X or different machines using the
331: Motorola 68000, Vax, SPARC, 80386, NS 32000, PC/RT, or MIPS architecture,
332: it should frequently suffice to change definitions in gcconfig.h.
333:
334:
335: THE C INTERFACE TO THE ALLOCATOR
336:
337: The following routines are intended to be directly called by the user.
338: Note that usually only GC_malloc is necessary. GC_clear_roots and GC_add_roots
339: calls may be required if the collector has to trace from nonstandard places
340: (e.g. from dynamic library data areas on a machine on which the
341: collector doesn't already understand them.) On some machines, it may
342: be desirable to set GC_stacktop to a good approximation of the stack base.
343: (This enhances code portability on HP PA machines, since there is no
344: good way for the collector to compute this value.) Client code may include
345: "gc.h", which defines all of the following, plus many others.
346:
347: 1) GC_malloc(nbytes)
348: - allocate an object of size nbytes. Unlike malloc, the object is
349: cleared before being returned to the user. Gc_malloc will
350: invoke the garbage collector when it determines this to be appropriate.
351: GC_malloc may return 0 if it is unable to acquire sufficient
352: space from the operating system. This is the most probable
353: consequence of running out of space. Other possible consequences
354: are that a function call will fail due to lack of stack space,
355: or that the collector will fail in other ways because it cannot
356: maintain its internal data structures, or that a crucial system
357: process will fail and take down the machine. Most of these
358: possibilities are independent of the malloc implementation.
359:
360: 2) GC_malloc_atomic(nbytes)
361: - allocate an object of size nbytes that is guaranteed not to contain any
362: pointers. The returned object is not guaranteed to be cleared.
363: (Can always be replaced by GC_malloc, but results in faster collection
364: times. The collector will probably run faster if large character
365: arrays, etc. are allocated with GC_malloc_atomic than if they are
366: statically allocated.)
367:
368: 3) GC_realloc(object, new_size)
369: - change the size of object to be new_size. Returns a pointer to the
370: new object, which may, or may not, be the same as the pointer to
371: the old object. The new object is taken to be atomic iff the old one
372: was. If the new object is composite and larger than the original object,
373: then the newly added bytes are cleared (we hope). This is very likely
374: to allocate a new object, unless MERGE_SIZES is defined in gc_priv.h.
375: Even then, it is likely to recycle the old object only if the object
376: is grown in small additive increments (which, we claim, is generally bad
377: coding practice.)
378:
379: 4) GC_free(object)
380: - explicitly deallocate an object returned by GC_malloc or
381: GC_malloc_atomic. Not necessary, but can be used to minimize
382: collections if performance is critical. Probably a performance
383: loss for very small objects (<= 8 bytes).
384:
385: 5) GC_expand_hp(bytes)
386: - Explicitly increase the heap size. (This is normally done automatically
387: if a garbage collection failed to GC_reclaim enough memory. Explicit
388: calls to GC_expand_hp may prevent unnecessarily frequent collections at
389: program startup.)
390:
391: 6) GC_malloc_ignore_off_page(bytes)
392: - identical to GC_malloc, but the client promises to keep a pointer to
393: the somewhere within the first 256 bytes of the object while it is
394: live. (This pointer should nortmally be declared volatile to prevent
395: interference from compiler optimizations.) This is the recommended
396: way to allocate anything that is likely to be larger than 100Kbytes
397: or so. (GC_malloc may result in failure to reclaim such objects.)
398:
399: 7) GC_set_warn_proc(proc)
400: - Can be used to redirect warnings from the collector. Such warnings
401: should be rare, and should not be ignored during code development.
402:
403: 8) GC_enable_incremental()
404: - Enables generational and incremental collection. Useful for large
405: heaps on machines that provide access to page dirty information.
406: Some dirty bit implementations may interfere with debugging
407: (by catching address faults) and place restrictions on heap arguments
408: to system calls (since write faults inside a system call may not be
409: handled well).
410:
411: 9) Several routines to allow for registration of finalization code.
412: User supplied finalization code may be invoked when an object becomes
413: unreachable. To call (*f)(obj, x) when obj becomes inaccessible, use
414: GC_register_finalizer(obj, f, x, 0, 0);
415: For more sophisticated uses, and for finalization ordering issues,
416: see gc.h.
417:
418: The global variable GC_free_space_divisor may be adjusted up from its
419: default value of 4 to use less space and more collection time, or down for
420: the opposite effect. Setting it to 1 or 0 will effectively disable collections
421: and cause all allocations to simply grow the heap.
422:
423: The variable GC_non_gc_bytes, which is normally 0, may be changed to reflect
424: the amount of memory allocated by the above routines that should not be
425: considered as a candidate for collection. Careless use may, of course, result
426: in excessive memory consumption.
427:
428: Some additional tuning is possible through the parameters defined
429: near the top of gc_priv.h.
430:
431: If only GC_malloc is intended to be used, it might be appropriate to define:
432:
433: #define malloc(n) GC_malloc(n)
434: #define calloc(m,n) GC_malloc((m)*(n))
435:
436: For small pieces of VERY allocation intensive code, gc_inl.h
437: includes some allocation macros that may be used in place of GC_malloc
438: and friends.
439:
440: All externally visible names in the garbage collector start with "GC_".
441: To avoid name conflicts, client code should avoid this prefix, except when
442: accessing garbage collector routines or variables.
443:
444: There are provisions for allocation with explicit type information.
445: This is rarely necessary. Details can be found in gc_typed.h.
446:
447: THE C++ INTERFACE TO THE ALLOCATOR:
448:
449: The Ellis-Hull C++ interface to the collector is included in
450: the collector distribution. If you intend to use this, type
451: "make c++" after the initial build of the collector is complete.
452: See gc_cpp.h for the definition of the interface. This interface
453: tries to approximate the Ellis-Detlefs C++ garbage collection
454: proposal without compiler changes.
455:
456: Cautions:
457: 1. Arrays allocated without new placement syntax are
458: allocated as uncollectable objects. They are traced by the
459: collector, but will not be reclaimed.
460:
461: 2. Failure to use "make c++" in combination with (1) will
462: result in arrays allocated using the default new operator.
463: This is likely to result in disaster without linker warnings.
464:
465: 3. If your compiler supports an overloaded new[] operator,
466: then gc_cpp.cc and gc_cpp.h should be suitably modified.
467:
468: 4. Many current C++ compilers have deficiencies that
469: break some of the functionality. See the comments in gc_cpp.h
470: for suggested workarounds.
471:
472: USE AS LEAK DETECTOR:
473:
474: The collector may be used to track down leaks in C programs that are
475: intended to run with malloc/free (e.g. code with extreme real-time or
476: portability constraints). To do so define FIND_LEAK in Makefile
477: This will cause the collector to invoke the report_leak
478: routine defined near the top of reclaim.c whenever an inaccessible
479: object is found that has not been explicitly freed. The collector will
480: no longer reclaim inaccessible memory; in this form it is purely a
481: debugging tool.
482: Productive use of this facility normally involves redefining report_leak
483: to do something more intelligent. This typically requires annotating
484: objects with additional information (e.g. creation time stack trace) that
485: identifies their origin. Such code is typically not very portable, and is
486: not included here, except on SPARC machines.
487: If all objects are allocated with GC_DEBUG_MALLOC (see next section),
488: then the default version of report_leak will report the source file
489: and line number at which the leaked object was allocated. This may
490: sometimes be sufficient. (On SPARC/SUNOS4 machines, it will also report
491: a cryptic stack trace. This can often be turned into a sympolic stack
492: trace by invoking program "foo" with "callprocs foo". Callprocs is
493: a short shell script that invokes adb to expand program counter values
494: to symbolic addresses. It was largely supplied by Scott Schwartz.)
495: Note that the debugging facilities described in the next section can
496: sometimes be slightly LESS effective in leak finding mode, since in
497: leak finding mode, GC_debug_free actually results in reuse of the object.
498: (Otherwise the object is simply marked invalid.) Also note that the test
499: program is not designed to run meaningfully in FIND_LEAK mode.
500: Use "make gc.a" to build the collector.
501:
502: DEBUGGING FACILITIES:
503:
504: The routines GC_debug_malloc, GC_debug_malloc_atomic, GC_debug_realloc,
505: and GC_debug_free provide an alternate interface to the collector, which
506: provides some help with memory overwrite errors, and the like.
507: Objects allocated in this way are annotated with additional
508: information. Some of this information is checked during garbage
509: collections, and detected inconsistencies are reported to stderr.
510:
511: Simple cases of writing past the end of an allocated object should
512: be caught if the object is explicitly deallocated, or if the
513: collector is invoked while the object is live. The first deallocation
514: of an object will clear the debugging info associated with an
515: object, so accidentally repeated calls to GC_debug_free will report the
516: deallocation of an object without debugging information. Out of
517: memory errors will be reported to stderr, in addition to returning
518: NIL.
519:
520: GC_debug_malloc checking during garbage collection is enabled
521: with the first call to GC_debug_malloc. This will result in some
522: slowdown during collections. If frequent heap checks are desired,
523: this can be achieved by explicitly invoking GC_gcollect, e.g. from
524: the debugger.
525:
526: GC_debug_malloc allocated objects should not be passed to GC_realloc
527: or GC_free, and conversely. It is however acceptable to allocate only
528: some objects with GC_debug_malloc, and to use GC_malloc for other objects,
529: provided the two pools are kept distinct. In this case, there is a very
530: low probablility that GC_malloc allocated objects may be misidentified as
531: having been overwritten. This should happen with probability at most
532: one in 2**32. This probability is zero if GC_debug_malloc is never called.
533:
534: GC_debug_malloc, GC_malloc_atomic, and GC_debug_realloc take two
535: additional trailing arguments, a string and an integer. These are not
536: interpreted by the allocator. They are stored in the object (the string is
537: not copied). If an error involving the object is detected, they are printed.
538:
539: The macros GC_MALLOC, GC_MALLOC_ATOMIC, GC_REALLOC, GC_FREE, and
540: GC_REGISTER_FINALIZER are also provided. These require the same arguments
541: as the corresponding (nondebugging) routines. If gc.h is included
542: with GC_DEBUG defined, they call the debugging versions of these
543: functions, passing the current file name and line number as the two
544: extra arguments, where appropriate. If gc.h is included without GC_DEBUG
545: defined, then all these macros will instead be defined to their nondebugging
546: equivalents. (GC_REGISTER_FINALIZER is necessary, since pointers to
547: objects with debugging information are really pointers to a displacement
548: of 16 bytes form the object beginning, and some translation is necessary
549: when finalization routines are invoked. For details, about what's stored
550: in the header, see the definition of the type oh in debug_malloc.c)
551:
552: INCREMENTAL/GENERATIONAL COLLECTION:
553:
554: The collector normally interrupts client code for the duration of
555: a garbage collection mark phase. This may be unacceptable if interactive
556: response is needed for programs with large heaps. The collector
557: can also run in a "generational" mode, in which it usually attempts to
558: collect only objects allocated since the last garbage collection.
559: Furthermore, in this mode, garbage collections run mostly incrementally,
560: with a small amount of work performed in response to each of a large number of
561: GC_malloc requests.
562:
563: This mode is enabled by a call to GC_enable_incremental().
564:
565: Incremental and generational collection is effective in reducing
566: pause times only if the collector has some way to tell which objects
567: or pages have been recently modified. The collector uses two sources
568: of information:
569:
570: 1. Information provided by the VM system. This may be provided in
571: one of several forms. Under Solaris 2.X (and potentially under other
572: similar systems) information on dirty pages can be read from the
573: /proc file system. Under other systems (currently SunOS4.X) it is
574: possible to write-protect the heap, and catch the resulting faults.
575: On these systems we require that system calls writing to the heap
576: (other than read) be handled specially by client code.
577: See os_dep.c for details.
578:
579: 2. Information supplied by the programmer. We define "stubborn"
580: objects to be objects that are rarely changed. Such an object
581: can be allocated (and enabled for writing) with GC_malloc_stubborn.
582: Once it has been initialized, the collector should be informed with
583: a call to GC_end_stubborn_change. Subsequent writes that store
584: pointers into the object must be preceded by a call to
585: GC_change_stubborn.
586:
587: This mechanism performs best for objects that are written only for
588: initialization, and such that only one stubborn object is writable
589: at once. It is typically not worth using for short-lived
590: objects. Stubborn objects are treated less efficiently than pointerfree
591: (atomic) objects.
592:
593: A rough rule of thumb is that, in the absence of VM information, garbage
594: collection pauses are proportional to the amount of pointerful storage
595: plus the amount of modified "stubborn" storage that is reachable during
596: the collection.
597:
598: Initial allocation of stubborn objects takes longer than allocation
599: of other objects, since other data structures need to be maintained.
600:
601: We recommend against random use of stubborn objects in client
602: code, since bugs caused by inappropriate writes to stubborn objects
603: are likely to be very infrequently observed and hard to trace.
604: However, their use may be appropriate in a few carefully written
605: library routines that do not make the objects themselves available
606: for writing by client code.
607:
608:
609: BUGS:
610:
611: Any memory that does not have a recognizable pointer to it will be
612: reclaimed. Exclusive-or'ing forward and backward links in a list
613: doesn't cut it.
614: Some C optimizers may lose the last undisguised pointer to a memory
615: object as a consequence of clever optimizations. This has almost
616: never been observed in practice. Send mail to boehm@acm.org
617: for suggestions on how to fix your compiler.
618: This is not a real-time collector. In the standard configuration,
619: percentage of time required for collection should be constant across
620: heap sizes. But collection pauses will increase for larger heaps.
621: (On SPARCstation 2s collection times will be on the order of 300 msecs
622: per MB of accessible memory that needs to be scanned. Your mileage
623: may vary.) The incremental/generational collection facility helps,
624: but is portable only if "stubborn" allocation is used.
625: Please address bug reports to boehm@acm.org. If you are
626: contemplating a major addition, you might also send mail to ask whether
627: it's already been done (or whether we tried and discarded it).
628:
629: RECENT VERSIONS:
630:
631: Version 1.3 and immediately preceding versions contained spurious
632: assembly language assignments to TMP_SP. Only the assignment in the PC/RT
633: code is necessary. On other machines, with certain compiler options,
634: the assignments can lead to an unsaved register being overwritten.
635: Known to cause problems under SunOS 3.5 WITHOUT the -O option. (With
636: -O the compiler recognizes it as dead code. It probably shouldn't,
637: but that's another story.)
638:
639: Version 1.4 and earlier versions used compile time determined values
640: for the stack base. This no longer works on Sun 3s, since Sun 3/80s use
641: a different stack base. We now use a straightforward heuristic on all
642: machines on which it is known to work (incl. Sun 3s) and compile-time
643: determined values for the rest. There should really be library calls
644: to determine such values.
645:
646: Version 1.5 and earlier did not ensure 8 byte alignment for objects
647: allocated on a sparc based machine.
648:
649: Version 1.8 added ULTRIX support in gc_private.h.
650:
651: Version 1.9 fixed a major bug in gc_realloc.
652:
653: Version 2.0 introduced a consistent naming convention for collector
654: routines and added support for registering dynamic library data segments
655: in the standard mark_roots.c. Most of the data structures were revamped.
656: The treatment of interior pointers was completely changed. Finalization
657: was added. Support for locking was added. Object kinds were added.
658: We added a black listing facility to avoid allocating at addresses known
659: to occur as integers somewhere in the address space. Much of this
660: was accomplished by adapting ideas and code from the PCR collector.
661: The test program was changed and expanded.
662:
663: Version 2.1 was the first stable version since 1.9, and added support
664: for PPCR.
665:
666: Version 2.2 added debugging allocation, and fixed various bugs. Among them:
667: - GC_realloc could fail to extend the size of the object for certain large object sizes.
668: - A blatant subscript range error in GC_printf, which unfortunately
669: wasn't exercised on machines with sufficient stack alignment constraints.
670: - GC_register_displacement did the wrong thing if it was called after
671: any allocation had taken place.
672: - The leak finding code would eventually break after 2048 byte
673: byte objects leaked.
674: - interface.c didn't compile.
675: - The heap size remained much too small for large stacks.
676: - The stack clearing code behaved badly for large stacks, and perhaps
677: on HP/PA machines.
678:
679: Version 2.3 added ALL_INTERIOR_POINTERS and fixed the following bugs:
680: - Missing declaration of etext in the A/UX version.
681: - Some PCR root-finding problems.
682: - Blacklisting was not 100% effective, because the plausible future
683: heap bounds were being miscalculated.
684: - GC_realloc didn't handle out-of-memory correctly.
685: - GC_base could return a nonzero value for addresses inside free blocks.
686: - test.c wasn't really thread safe, and could erroneously report failure
687: in a multithreaded environment. (The locking primitives need to be
688: replaced for other threads packages.)
689: - GC_CONS was thoroughly broken.
690: - On a SPARC with dynamic linking, signals stayed diabled while the
691: client code was running.
692: (Thanks to Manuel Serrano at INRIA for reporting the last two.)
693:
694: Version 2.4 added GC_free_space_divisor as a tuning knob, added
695: support for OS/2 and linux, and fixed the following bugs:
696: - On machines with unaligned pointers (e.g. Sun 3), every 128th word could
697: fail to be considered for marking.
698: - Dynamic_load.c erroneously added 4 bytes to the length of the data and
699: bss sections of the dynamic library. This could result in a bad memory
700: reference if the actual length was a multiple of a page. (Observed on
701: Sun 3. Can probably also happen on a Sun 4.)
702: (Thanks to Robert Brazile for pointing out that the Sun 3 version
703: was broken. Dynamic library handling is still broken on Sun 3s
704: under 4.1.1U1, but apparently not 4.1.1. If you have such a machine,
705: use -Bstatic.)
706:
707: Version 2.5 fixed the following bugs:
708: - Removed an explicit call to exit(1)
709: - Fixed calls to GC_printf and GC_err_printf, so the correct number of
710: arguments are always supplied. The OS/2 C compiler gets confused if
711: the number of actuals and the number of formals differ. (ANSI C
712: doesn't require this to work. The ANSI sanctioned way of doing things
713: causes too many compatibility problems.)
714:
715: Version 3.0 added generational/incremental collection and stubborn
716: objects.
717:
718: Version 3.1 added the following features:
719: - A workaround for a SunOS 4.X SPARC C compiler
720: misfeature that caused problems when the collector was turned into
721: a dynamic library.
722: - A fix for a bug in GC_base that could result in a memory fault.
723: - A fix for a performance bug (and several other misfeatures) pointed
724: out by Dave Detlefs and Al Dosser.
725: - Use of dirty bit information for static data under Solaris 2.X.
726: - DEC Alpha/OSF1 support (thanks to Al Dosser).
727: - Incremental collection on more platforms.
728: - A more refined heap expansion policy. Less space usage by default.
729: - Various minor enhancements to reduce space usage, and to reduce
730: the amount of memory scanned by the collector.
731: - Uncollectable allocation without per object overhead.
732: - More conscientious handling of out-of-memory conditions.
733: - Fixed a bug in debugging stubborn allocation.
734: - Fixed a bug that resulted in occasional erroneous reporting of smashed
735: objects with debugging allocation.
736: - Fixed bogus leak reports of size 4096 blocks with FIND_LEAK.
737:
738: Version 3.2 fixed a serious and not entirely repeatable bug in
739: the incremental collector. It appeared only when dirty bit info
740: on the roots was available, which is normally only under Solaris.
741: It also added GC_general_register_disappearing_link, and some
742: testing code. Interface.c disappeared.
743:
744: Version 3.3 fixes several bugs and adds new ports:
745: - PCR-specific bugs.
746: - Missing locking in GC_free, redundant FASTUNLOCK
747: in GC_malloc_stubborn, and 2 bugs in
748: GC_unregister_disappearing_link.
749: All of the above were pointed out by Neil Sharman
750: (neil@cs.mu.oz.au).
751: - Common symbols allocated by the SunOS4.X dynamic loader
752: were not included in the root set.
753: - Bug in GC_finalize (reported by Brian Beuning and Al Dosser)
754: - Merged Amiga port from Jesper Peterson (untested)
755: - Merged NeXT port from Thomas Funke (significantly
756: modified and untested)
757:
758: Version 3.4:
759: - Fixed a performance bug in GC_realloc.
760: - Updated the amiga port.
761: - Added NetBSD and 386BSD ports.
762: - Added cord library.
763: - Added trivial performance enhancement for
764: ALL_INTERIOR_POINTERS. (Don't scan last word.)
765:
766: Version 3.5
767: - Minor collections now mark from roots only once, if that
768: doesn't cause an excessive pause.
769: - The stack clearing heuristic was refined to prevent anomalies
770: with very heavily recursive programs and sparse stacks.
771: - Fixed a bug that prevented mark stack growth in some cases.
772: GC_objects_are_marked should be set to TRUE after a call
773: to GC_push_roots and as part of GC_push_marked, since
774: both can now set mark bits. I think this is only a performance
775: bug, but I wouldn't bet on it. It's certainly very hard to argue
776: that the old version was correct.
777: - Fixed an incremental collection bug that prevented it from
778: working at all when HBLKSIZE != getpagesize()
779: - Changed dynamic_loading.c to include gc_priv.h before testing
780: DYNAMIC_LOADING. SunOS dynamic library scanning
781: must have been broken in 3.4.
782: - Object size rounding now adapts to program behavior.
783: - Added a workaround (provided by Manuel Serrano and
784: colleagues) to a long-standing SunOS 4.X (and 3.X?) ld bug
785: that I had incorrectly assumed to have been squished.
786: The collector was broken if the text segment size was within
787: 32 bytes of a multiple of 8K bytes, and if the beginning of
788: the data segment contained interesting roots. The workaround
789: assumes a demand-loadable executable. The original may have
790: have "worked" in some other cases.
791: - Added dynamic library support under IRIX5.
792: - Added support for EMX under OS/2 (thanks to Ari Huttunen).
793:
794: Version 3.6:
795: - fixed a bug in the mark stack growth code that was introduced
796: in 3.4.
797: - fixed Makefile to work around DEC AXP compiler tail recursion
798: bug.
799:
800: Version 3.7:
801: - Added a workaround for an HP/UX compiler bug.
802: - Fixed another stack clearing performance bug. Reworked
803: that code once more.
804:
805: Version 4.0:
806: - Added support for Solaris threads (which was possible
807: only by reimplementing some fraction of Solaris threads,
808: since Sun doesn't currently make the thread debugging
809: interface available).
810: - Added non-threads win32 and win32S support.
811: - (Grudgingly, with suitable muttering of obscenities) renamed
812: files so that the collector distribution could live on a FAT
813: file system. Files that are guaranteed to be useless on
814: a PC still have long names. Gc_inline.h and gc_private.h
815: still exist, but now just include gc_inl.h and gc_priv.h.
816: - Fixed a really obscure bug in finalization that could cause
817: undetected mark stack overflows. (I would be surprised if
818: any real code ever tickled this one.)
819: - Changed finalization code to dynamically resize the hash
820: tables it maintains. (This probably does not matter for well-
821: -written code. It no doubt does for C++ code that overuses
822: destructors.)
823: - Added typed allocation primitives. Rewrote the marker to
824: accommodate them with more reasonable efficiency. This
825: change should also speed up marking for GC_malloc allocated
826: objects a little. See gc_typed.h for new primitives.
827: - Improved debugging facilities slightly. Allocation time
828: stack traces are now kept by default on SPARC/SUNOS4.
829: (Thanks to Scott Schwartz.)
830: - Added better support for small heap applications.
831: - Significantly extended cord package. Fixed a bug in the
832: implementation of lazily read files. Printf and friends now
833: have cord variants. Cord traversals are a bit faster.
834: - Made ALL_INTERIOR_POINTERS recognition the default.
835: - Fixed de so that it can run in constant space, independent
836: of file size. Added simple string searching to cords and de.
837: - Added the Hull-Ellis C++ interface.
838: - Added dynamic library support for OSF/1.
839: (Thanks to Al Dosser and Tim Bingham at DEC.)
840: - Changed argument to GC_expand_hp to be expressed
841: in units of bytes instead of heap blocks. (Necessary
842: since the heap block size now varies depending on
843: configuration. The old version was never very clean.)
844: - Added GC_get_heap_size(). The previous "equivalent"
845: was broken.
846: - Restructured the Makefile a bit.
847:
848: Since version 4.0:
849: - Changed finalization implementation to guarantee that
850: finalization procedures are called outside of the allocation
851: lock, making direct use of the interface a little less dangerous.
852: MAY BREAK EXISTING CLIENTS that assume finalizers
853: are protected by a lock. Since there seem to be few multithreaded
854: clients that use finalization, this is hopefully not much of
855: a problem.
856: - Fixed a gross bug in CORD_prev.
857: - Fixed a bug in blacklst.c that could result in unbounded
858: heap growth during startup on machines that do not clear
859: memory obtained from the OS (e.g. win32S).
860: - Ported de editor to win32/win32S. (This is now the only
861: version with a mouse-sensitive UI.)
862: - Added GC_malloc_ignore_off_page to allocate large arrays
863: in the presence of ALL_INTERIOR_POINTERS.
864: - Changed GC_call_with_alloc_lock to not disable signals in
865: the single-threaded case.
866: - Reduced retry count in GC_collect_or_expand for garbage
867: collecting when out of memory.
868: - Made uncollectable allocations bypass black-listing, as they
869: should.
870: - Fixed a bug in typed_test in test.c that could cause (legitimate)
871: GC crashes.
872: - Fixed some potential synchronization problems in finalize.c
873: - Fixed a real locking problem in typd_mlc.c.
874: - Worked around an AIX 3.2 compiler feature that results in
875: out of bounds memory references.
876: - Partially worked around an IRIX5.2 beta problem (which may
877: or may not persist to the final release).
878: - Fixed a bug in the heap integrity checking code that could
879: result in explicitly deallocated objects being identified as
880: smashed. Fixed a bug in the dbg_mlc stack saving code
881: that caused old argument pointers to be considered live.
882: - Fixed a bug in CORD_ncmp (and hence CORD_str).
883: - Repaired the OS2 port, which had suffered from bit rot
884: in 4.0. Worked around what appears to be CSet/2 V1.0
885: optimizer bug.
886: - Fixed a Makefile bug for target "c++".
887:
888: Since version 4.1:
889: - Multiple bug fixes/workarounds in the Solaris threads version.
890: (It occasionally failed to locate some register contents for
891: marking. It also turns out that thr_suspend and friends are
892: unreliable in Solaris 2.3. Dirty bit reads appear
893: to be unreliable under some weird
894: circumstances. My stack marking code
895: contained a serious performance bug. The new code is
896: extremely defensive, and has not failed in several cpu
897: hours of testing. But no guarantees ...)
898: - Added MacOS support (thanks to Patrick Beard.)
899: - Fixed several syntactic bugs in gc_c++.h and friends. (These
900: didn't bother g++, but did bother most other compilers.)
901: Fixed gc_c++.h finalization interface. (It didn't.)
902: - 64 bit alignment for allocated objects was not guaranteed in a
903: few cases in which it should have been.
904: - Added GC_malloc_atomic_ignore_off_page.
905: - Added GC_collect_a_little.
906: - Added some prototypes to gc.h.
907: - Some other minor bug fixes (notably in Makefile).
908: - Fixed OS/2 / EMX port (thanks to Ari Huttunen).
909: - Fixed AmigaDOS port. (thanks to Michel Schinz).
910: - Fixed the DATASTART definition under Solaris. There
911: was a 1 in 16K chance of the collector missing the first
912: 64K of static data (and thus crashing).
913: - Fixed some blatant anachronisms in the README file.
914: - Fixed PCR-Makefile for upcoming PPCR release.
915:
916: Since version 4.2:
917: - Fixed SPARC alignment problem with GC_DEBUG.
918: - Fixed Solaris threads /proc workaround. The real
919: problem was an interaction with mprotect.
920: - Incorporated fix from Patrick Beard for gc_c++.h (now gc_cpp.h).
921: - Slightly improved allocator space utilization by
922: fixing the GC_size_map mechanism.
923: - Integrated some Sony News and MIPS RISCos 4.51
924: patches. (Thanks to Nobuyuki Hikichi of
925: Software Research Associates, Inc. Japan)
926: - Fixed HP_PA alignment problem. (Thanks to
927: xjam@cork.cs.berkeley.edu.)
928: - Added GC_same_obj and friends. Changed GC_base
929: to return 0 for pointers past the end of large objects.
930: Improved GC_base performance with ALL_INTERIOR_POINTERS
931: on machines with a slow integer mod operation.
932: Added GC_PTR_ADD, GC_PTR_STORE, etc. to prepare
933: for preprocessor.
934: - changed the default on most UNIX machines to be that
935: signals are not disabled during critical GC operations.
936: This is still ANSI-conforming, though somewhat dangerous
937: in the presence of signal handlers. But the performance
938: cost of the alternative is sometimes problematic.
939: Can be changed back with a minor Makefile edit.
940: - renamed IS_STRING in gc.h, to CORD_IS_STRING, thus
941: following my own naming convention. Added the function
942: CORD_to_const_char_star.
943: - Fixed a gross bug in GC_finalize. Symptom: occasional
944: address faults in that function. (Thanks to Anselm
945: Baird-Smith (Anselm.BairdSmith@inria.fr)
946: - Added port to ICL DRS6000 running DRS/NX. Restructured
947: things a bit to factor out common code, and remove obsolete
948: code. Collector should now run under SUNOS5 with either
949: mprotect or /proc dirty bits. (Thanks to Douglas Steel
950: (doug@wg.icl.co.uk)).
951: - More bug fixes and workarounds for Solaris 2.X. (These were
952: mostly related to putting the collector in a dynamic library,
953: which didn't really work before. Also SOLARIS_THREADS
954: didn't interact well with dl_open.) Thanks to btlewis@eng.sun.com.
955: - Fixed a serious performance bug on the DEC Alpha. The text
956: segment was getting registered as part of the root set.
957: (Amazingly, the result was still fast enough that the bug
958: was not conspicuous.) The fix works on OSF/1, version 1.3.
959: Hopefully it also works on other versions of OSF/1 ...
960: - Fixed a bug in GC_clear_roots.
961: - Fixed a bug in GC_generic_malloc_words_small that broke
962: gc_inl.h. (Reported by Antoine de Maricourt. I broke it
963: in trying to tweak the Mac port.)
964: - Fixed some problems with cord/de under Linux.
965: - Fixed some cord problems, notably with CORD_riter4.
966: - Added DG/UX port.
967: Thanks to Ben A. Mesander (ben@piglet.cr.usgs.gov)
968: - Added finalization registration routines with weaker ordering
969: constraints. (This is necessary for C++ finalization with
970: multiple inheritance, since the compiler often adds self-cycles.)
971: - Filled the holes in the SCO port. (Thanks to Michael Arnoldus
972: <chime@proinf.dk>.)
973: - John Ellis' additions to the C++ support: From John:
974:
975: * I completely rewrote the documentation in the interface gc_c++.h
976: (later renamed gc_cpp.h). I've tried to make it both clearer and more
977: precise.
978:
979: * The definition of accessibility now ignores pointers from an
980: finalizable object (an object with a clean-up function) to itself.
981: This allows objects with virtual base classes to be finalizable by the
982: collector. Compilers typically implement virtual base classes using
983: pointers from an object to itself, which under the old definition of
984: accessibility prevented objects with virtual base classes from ever
985: being collected or finalized.
986:
987: * gc_cleanup now includes gc as a virtual base. This was enabled by
988: the change in the definition of accessibility.
989:
990: * I added support for operator new[]. Since most (all?) compilers
991: don't yet support operator new[], it is conditionalized on
992: -DOPERATOR_NEW_ARRAY. The code is untested, but its trivial and looks
993: correct.
994:
995: * The test program test_gc_c++ (later renamed test_cpp.cc)
996: tries to test for the C++-specific functionality not tested by the
997: other programs.
998: - Added <unistd.h> include to misc.c. (Needed for ppcr.)
999: - Added PowerMac port. (Thanks to Patrick Beard again.)
1000: - Fixed "srcdir"-related Makefile problems. Changed things so
1001: that all externally visible include files always appear in the
1002: include subdirectory of the source. Made gc.h directly
1003: includable from C++ code. (These were at Per
1004: Bothner's suggestion.)
1005: - Changed Intel code to also mark from ebp (Kevin Warne's
1006: suggestion).
1007: - Renamed C++ related files so they could live in a FAT
1008: file system. (Charles Fiterman's suggestion.)
1009: - Changed Windows NT Makefile to include C++ support in
1010: gc.lib. Added C++ test as Makefile target.
1011:
1012: Since version 4.3:
1013: - ASM_CLEAR_CODE was erroneously defined for HP
1014: PA machines, resulting in a compile error.
1015: - Fixed OS/2 Makefile to create a library. (Thanks to
1016: Mark Boulter (mboulter@vnet.ibm.com)).
1017: - Gc_cleanup objects didn't work if they were created on
1018: the stack. Fixed.
1019: - One copy of Gc_cpp.h in the distribution was out of
1020: synch, and failed to document some known compiler
1021: problems with explicit destructor invocation. Partially
1022: fixed. There are probably other compilers on which
1023: gc_cleanup is miscompiled.
1024: - Fixed Makefile to pass C compiler flags to C++ compiler.
1025: - Added Mac fixes.
1026: - Fixed os_dep.c to work around what appears to be
1027: a new and different VirtualQuery bug under newer
1028: versions of win32S.
1029: - GC_non_gc_bytes was not correctly maintained by
1030: GC_free. Fixed. Thanks to James Clark (jjc@jclark.com).
1031: - Added GC_set_max_heap_size.
1032: - Changed allocation code to ignore blacklisting if it is preventing
1033: use of a very large block of memory. This has the advantage
1034: that naive code allocating very large objects is much more
1035: likely to work. The downside is you might no
1036: longer find out that such code should really use
1037: GC_malloc_ignore_off_page.
1038: - Changed GC_printf under win32 to close and reopen the file
1039: between calls. FAT file systems otherwise make the log file
1040: useless for debugging.
1041: - Added GC_try_to_collect and GC_get_bytes_since_gc. These
1042: allow starting an abortable collection during idle times.
1043: This facility does not require special OS support. (Thanks to
1044: Michael Spertus of Geodesic Systems for suggesting this. It was
1045: actually an easy addition. Kumar Srikantan previously added a similar
1046: facility to a now ancient version of the collector. At the time
1047: this was much harder, and the result was less convincing.)
1048: - Added some support for the Borland development environment. (Thanks
1049: to John Ellis and Michael Spertus.)
1050: - Removed a misfeature from checksums.c that caused unexpected
1051: heap growth. (Thanks to Scott Schwartz.)
1052: - Changed finalize.c to call WARN if it encounters a finalization cycle.
1053: WARN is defined in gc_priv.h to write a message, usually to stdout.
1054: In many environments, this may be inappropriate.
1055: - Renamed NO_PARAMS in gc.h to GC_NO_PARAMS, thus adhering to my own
1056: naming convention.
1057: - Added GC_set_warn_proc to intercept warnings.
1058: - Fixed Amiga port. (Thanks to Michel Schinz (schinz@alphanet.ch).)
1059: - Fixed a bug in mark.c that could result in an access to unmapped
1060: memory from GC_mark_from_mark_stack on machines with unaligned
1061: pointers.
1062: - Fixed a win32 specific performance bug that could result in scanning of
1063: objects allocated with the system malloc.
1064: - Added REDIRECT_MALLOC.
1065:
1066: Since version 4.4:
1067: - Fixed many minor and one major README bugs. (Thanks to Franklin Chen
1068: (chen@adi.com) for pointing out many of them.)
1069: - Fixed ALPHA/OSF/1 dynamic library support. (Thanks to Jonathan Bachrach
1070: (jonathan@harlequin.com)).
1071: - Added incremental GC support (MPROTECT_VDB) for Linux (with some
1072: help from Bruno Haible).
1073: - Altered SPARC recognition tests in gc.h and config.h (mostly as
1074: suggested by Fergus Henderson).
1075: - Added basic incremental GC support for win32, as implemented by
1076: Windows NT and Windows 95. GC_enable_incremental is a noop
1077: under win32s, which doesn't implement enough of the VM interface.
1078: - Added -DLARGE_CONFIG.
1079: - Fixed GC_..._ignore_off_page to also function without
1080: -DALL_INTERIOR_POINTERS.
1081: - (Hopefully) fixed RS/6000 port. (Only the test was broken.)
1082: - Fixed a performance bug in the nonincremental collector running
1083: on machines supporting incremental collection with MPROTECT_VDB
1084: (e.g. SunOS 4, DEC AXP). This turned into a correctness bug under
1085: win32s with win32 incremental collection. (Not all memory protection
1086: was disabled.)
1087: - Fixed some ppcr related bit rot.
1088: - Caused dynamic libraries to be unregistered before reregistering.
1089: The old way turned out to be a performance bug on some machines.
1090: - GC_root_size was not properly maintained under MSWIN32.
1091: - Added -DNO_DEBUGGING and GC_dump.
1092: - Fixed a couple of bugs arising with SOLARIS_THREADS +
1093: REDIRECT_MALLOC.
1094: - Added NetBSD/M68K port. (Thanks to Peter Seebach
1095: <seebs@taniemarie.solon.com>.)
1096: - Fixed a serious realloc bug. For certain object sizes, the collector
1097: wouldn't scan the expanded part of the object. (Thanks to Clay Spence
1098: (cds@peanut.sarnoff.com) for noticing the problem, and helping me to
1099: track it down.)
1100:
1101: Since version 4.5:
1102: - Added Linux ELF support. (Thanks to Arrigo Triulzi <arrigo@ic.ac.uk>.)
1103: - GC_base crashed if it was called before any other GC_ routines.
1104: This could happen if a gc_cleanup object was allocated outside the heap
1105: before any heap allocation.
1106: - The heap expansion heuristic was not stable if all objects had finalization
1107: enabled. Fixed finalize.c to count memory in finalization queue and
1108: avoid explicit deallocation. Changed alloc.c to also consider this count.
1109: (This is still not recommended. It's expensive if nothing else.) Thanks
1110: to John Ellis for pointing this out.
1111: - GC_malloc_uncollectable(0) was broken. Thanks to Phong Vo for pointing
1112: this out.
1113: - The collector didn't compile under Linux 1.3.X. (Thanks to Fred Gilham for
1114: pointing this out.) The current workaround is ugly, but expected to be
1115: temporary.
1116: - Fixed a formatting problem for SPARC stack traces.
1117: - Fixed some '=='s in os_dep.c that should have been assignments.
1118: Fortunately these were in code that should never be executed anyway.
1119: (Thanks to Fergus Henderson.)
1120: - Fixed the heap block allocator to only drop blacklisted blocks in small
1121: chunks. Made BL_LIMIT self adjusting. (Both of these were in response
1122: to heap growth observed by Paul Graham.)
1123: - Fixed the Metrowerks/68K Mac code to also mark from a6. (Thanks
1124: to Patrick Beard.)
1125: - Significantly updated README.debugging.
1126: - Fixed some problems with longjmps out of signal handlers, especially under
1127: Solaris. Added a workaround for the fact that siglongjmp doesn't appear to
1128: do the right thing with -lthread under Solaris.
1129: - Added MSDOS/djgpp port. (Thanks to Mitch Harris (maharri@uiuc.edu).)
1130: - Added "make reserved_namespace" and "make user_namespace". The
1131: first renames ALL "GC_xxx" identifiers as "_GC_xxx". The second is the
1132: inverse transformation. Note that doing this is guaranteed to break all
1133: clients written for the other names.
1134: - descriptor field for kind NORMAL in GC_obj_kinds with ADD_BYTE_AT_END
1135: defined should be -ALIGNMENT not WORDS_TO_BYTES(-1). This is
1136: a serious bug on machines with pointer alignment of less than a word.
1137: - GC_ignore_self_finalize_mark_proc didn't handle pointers to very near the
1138: end of the object correctly. Caused failures of the C++ test on a DEC Alpha
1139: with g++.
1140: - gc_inl.h still had problems. Partially fixed. Added warnings at the
1141: beginning to hopefully specify the remaining dangers.
1142: - Added DATAEND definition to config.h.
1143: - Fixed some of the .h file organization. Fixed "make floppy".
1144:
1145: Since version 4.6:
1146: - Fixed some compilation problems with -DCHECKSUMS (thanks to Ian Searle)
1147: - Updated some Mac specific files to synchronize with Patrick Beard.
1148: - Fixed a serious bug for machines with non-word-aligned pointers.
1149: (Thanks to Patrick Beard for pointing out the problem. The collector
1150: should fail almost any conceivable test immediately on such machines.)
1151:
1152: Since version 4.7:
1153: - Changed a "comment" in a MacOS specific part of mach-dep.c that caused
1154: gcc to fail on other platforms.
1155:
1156: Since version 4.8
1157: - More README.debugging fixes.
1158: - Objects ready for finalization, but not finalized in the same GC
1159: cycle, could be prematurely collected. This occasionally happened
1160: in test_cpp.
1161: - Too little memory was obtained from the system for very large
1162: objects. That could cause a heap explosion if these objects were
1163: not contiguous (e.g. under PCR), and too much of them was blacklisted.
1164: - Due to an improper initialization, the collector was too hesitant to
1165: allocate blacklisted objects immediately after system startup.
1166: - Moved GC_arrays from the data into the bss segment by not explicitly
1167: initializing it to zero. This significantly
1168: reduces the size of executables, and probably avoids some disk accesses
1169: on program startup. It's conceivable that it might break a port that I
1170: didn't test.
1171: - Fixed EMX_MAKEFILE to reflect the gc_c++.h to gc_cpp.h renaming which
1172: occurred a while ago.
1173:
1174: Since 4.9:
1175: - Fixed a typo around a call to GC_collect_or_expand in alloc.c. It broke
1176: handling of out of memory. (Thanks to Patrick Beard for noticing.)
1177:
1178: Since 4.10:
1179: - Rationalized (hopefully) GC_try_to_collect in an incremental collection
1180: environment. It appeared to not handle a call while a collection was in
1181: progress, and was otherwise too conservative.
1182: - Merged GC_reclaim_or_delete_all into GC_reclaim_all to get rid of some
1183: code.
1184: - Added Patrick Beard's Mac fixes, with substantial completely untested
1185: modifications.
1186: - Fixed the MPROTECT_VDB code to deal with large pages and imprecise
1187: fault addresses (as on an UltraSPARC running Solaris 2.5). Note that this
1188: was not a problem in the default configuration, which uses PROC_VDB.
1189: - The DEC Alpha assembly code needed to restore $gp between calls.
1190: Thanks to Fergus Henderson for tracking this down and supplying a
1191: patch.
1192: - The write command for "de" was completely broken for large files.
1193: I used the easiest portable fix, which involved changing the semantics
1194: so that f.new is written instead of overwriting f. That's safer anyway.
1195: - Added README.solaris2 with a discussion of the possible problems of
1196: mixing the collector's sbrk allocation with malloc/realloc.
1197: - Changed the data segment starting address for SGI machines. The
1198: old code failed under IRIX6.
1199: - Required double word alignment for MIPS.
1200: - Various minor fixes to remove warnings.
1201: - Attempted to fix some Solaris threads problems reported by Zhiying Chen.
1202: In particular, the collector could try to fork a thread with the
1203: world stopped as part of GC_thr_init. It also failed to deal with
1204: the case in which the original thread terminated before the whole
1205: process did.
1206: - Added -DNO_EXECUTE_PERMISSION. This has a major performance impact
1207: on the incremental collector under Irix, and perhaps under other
1208: operating systems.
1209: - Added some code to support allocating the heap with mmap. This may
1210: be preferable under some circumstances.
1211: - Integrated dynamic library support for HP.
1212: (Thanks to Knut Tvedten <knuttv@ifi.uio.no>.)
1213: - Integrated James Clark's win32 threads support, and made a number
1214: of changes to it, many of which were suggested by Pontus Rydin.
1215: This is still not 100% solid.
1216: - Integrated Alistair Crooks' support for UTS4 running on an Amdahl
1217: 370-class machine.
1218: - Fixed a serious bug in explicitly typed allocation. Objects requiring
1219: large descriptors where handled in a way that usually resulted in
1220: a segmentation fault in the marker. (Thanks to Jeremy Fitzhardinge
1221: for helping to track this down.)
1222: - Added partial support for GNU win32 development. (Thanks to Fergus
1223: Henderson.)
1224: - Added optional support for Java-style finalization semantics. (Thanks
1225: to Patrick Bridges.) This is recommended only for Java implementations.
1226: - GC_malloc_uncollectable faulted instead of returning 0 when out of
1227: memory. (Thanks to dan@math.uiuc.edu for noticing.)
1228: - Calls to GC_base before the collector was initialized failed on a
1229: DEC Alpha. (Thanks to Matthew Flatt.)
1230: - Added base pointer checking to GC_REGISTER_FINALIZER in debugging
1231: mode, at the suggestion of Jeremy Fitzhardinge.
1232: - GC_debug_realloc failed for uncollectable objects. (Thanks to
1233: Jeremy Fitzhardinge.)
1234: - Explicitly typed allocation could crash if it ran out of memory.
1235: (Thanks to Jeremy Fitzhardinge.)
1236: - Added minimal support for a DEC Alpha running Linux.
1237: - Fixed a problem with allocation of objects whose size overflowed
1238: ptrdiff_t. (This now fails unconditionally, as it should.)
1239: - Added the beginning of Irix pthread support.
1240: - Integrated Xiaokun Zhu's fixes for djgpp 2.01.
1241: - Added SGI-style STL allocator support (gc_alloc.h).
1242: - Fixed a serious bug in README.solaris2. Multithreaded programs must include
1243: gc.h with SOLARIS_THREADS defined.
1244: - Changed GC_free so it actually deallocates uncollectable objects.
1245: (Thanks to Peter Chubb for pointing out the problem.)
1246: - Added Linux ELF support for dynamic libararies. (Thanks again to
1247: Patrick Bridges.)
1248: - Changed the Borland cc configuration so that the assembler is not
1249: required.
1250: - Fixed a bug in the C++ test that caused it to fail in 64-bit
1251: environments.
1252:
1253: Since 4.11:
1254: - Fixed ElfW definition in dyn_load.c. (Thanks to Fergus Henderson.)
1255: This prevented the dynamic library support from compiling on some
1256: older ELF Linux systems.
1257: - Fixed UTS4 port (which I apparently mangled during the integration)
1258: (Thanks to again to Alistair Crooks.)
1259: - "Make C++" failed on Suns with SC4.0, due to a problem with "bool".
1260: Fixed in gc_priv.h.
1261: - Added more pieces for GNU win32. (Thanks to Timothy N. Newsham.)
1262: The current state of things should suffice for at least some
1263: applications.
1264: - Changed the out of memory retry count handling as suggested by
1265: Kenjiro Taura. (This matters only if GC_max_retries > 0, which
1266: is no longer the default.)
1267: - If a /proc read failed repeatedly, GC_written_pages was not updated
1268: correctly. (Thanks to Peter Chubb for diagnosing this.)
1269: - Under unlikely circumstances, the allocator could infinite loop in
1270: an out of memory situation. (Thanks again to Kenjiro Taura for
1271: identifying the problem and supplying a fix.)
1272: - Fixed a syntactic error in the DJGPP code. (Thanks to Fergus
1273: Henderson for finding this by inspection.) Also fixed a test program
1274: problem with DJGPP (Thanks to Peter Monks.)
1275: - Atomic uncollectable objects were not treated correctly by the
1276: incremental collector. This resulted in weird log statistics and
1277: occasional performance problems. (Thanks to Peter Chubb for pointing
1278: this out.)
1279: - Fixed some problems resulting from compilers that dont define
1280: __STDC__. In this case void * and char * were used inconsistently
1281: in some cases. (Void * should not have been used at all. If
1282: you have an ANSI superset compiler that does not define __STDC__,
1283: please compile with -D__STDC__=0. Thanks to Manuel Serrano and others
1284: for pointing out the problem.)
1285: - Fixed a compilation problem on Irix with -n32 and -DIRIX_THREADS.
1286: Also fixed some other IRIX_THREADS problems which may or may not have
1287: had observable symptoms.
1288: - Fixed an HP PA compilation problem in dyn_load.c. (Thanks to
1289: Philippe Queinnec.)
1290: - SEGV fault handlers sometimes did not get reset correctly. (Thanks
1291: to David Pickens.)
1292: - Added a fix for SOLARIS_THREADS on Intel. (Thanks again to David
1293: Pickens.) This probably needs more work to become functional.
1294: - Fixed struct sigcontext_struct in os_dep.c for compilation under
1295: Linux 2.1.X. (Thanks to Fergus Henderson.)
1296: - Changed the DJGPP STACKBOTTOM and DATASTART values to those suggested
1297: by Kristian Kristensen. These may still not be right, but it is
1298: it is likely to work more often than what was there before. They may
1299: even be exactly right.
1300: - Added a #include <string.h> to test_cpp.cc. This appears to help
1301: with HP/UX and gcc. (Thanks to assar@sics.se.)
1302: - Version 4.11 failed to run in incremental mode on recent 64-bit Irix
1303: kernels. This was a problem related to page unaligned heap segments.
1304: Changed the code to page align heap sections on all platforms.
1305: (I had mistakenly identified this as a kernel problem earlier.
1306: It was not.)
1307: - Version 4.11 did not make allocated storage executable, except on
1308: one or two platforms, due to a bug in a #if test. (Thanks to Dave
1309: Grove for pointing this out.)
1310: - Added sparc_sunos4_mach_dep.s to support Sun's compilers under SunOS4.
1311: - Added GC_exclude_static_roots.
1312: - Fixed the object size mapping algorithm. This shouldn't matter,
1313: but the old code was ugly.
1314: - Heap checking code could die if one of the allocated objects was
1315: larger than its base address. (Unsigned underflow problem. Thanks
1316: to Clay Spence for isolating the problem.)
1317: - Added RS6000 (AIX) dynamic library support and fixed STACK_BOTTOM.
1318: (Thanks to Fred Stearns.)
1319: - Added Fergus Henderson's patches for improved robustness with large
1320: heaps and lots of blacklisting.
1321: - Added Peter Chubb's changes to support Solaris Pthreads, to support
1322: MMAP allocation in Solaris, to allow Solaris to find dynamic libraries
1323: through /proc, to add malloc_typed_ignore_off_page, and a few other
1324: minor features and bug fixes.
1325: - The Solaris 2 port should not use sbrk. I received confirmation from
1326: Sun that the use of sbrk and malloc in the same program is not
1327: supported. The collector now defines USE_MMAP by default on Solaris.
1328: - Replaced the djgpp makefile with Gary Leavens' version.
1329: - Fixed MSWIN32 detection test.
1330: - Added Fergus Henderson's patches to allow putting the collector into
1331: a DLL under GNU win32.
1332: - Added Ivan V. Demakov's port to Watcom C on X86.
1333: - Added Ian Piumarta's Linux/PowerPC port.
1334: - On Brian Burton's suggestion added PointerFreeGC to the placement
1335: options in gc_cpp.h. This is of course unsafe, and may be controversial.
1336: On the other hand, it seems to be needed often enough that it's worth
1337: adding as a standard facility.
1338:
1339: Since 4.12:
1340: - Fixed a crucial bug in the Watcom port. There was a redundant decl
1341: of GC_push_one in gc_priv.h.
1342: - Added FINALIZE_ON_DEMAND.
1343: - Fixed some pre-ANSI cc problems in test.c.
1344: - Removed getpagesize() use for Solaris. It seems to be missing in one
1345: or two versions.
1346: - Fixed bool handling for SPARCCompiler version 4.2.
1347: - Fixed some files in include that had gotten unlinked from the main
1348: copy.
1349: - Some RS/6000 fixes (missing casts). Thanks to Toralf Foerster.
1350: - Fixed several problems in GC_debug_realloc, affecting mostly the
1351: FIND_LEAK case.
1352: - GC_exclude_static_roots contained a buggy unsigned comparison to
1353: terminate a loop. (Thanks to Wilson Ho.)
1354: - CORD_str failed if the substring occurred at the last possible position.
1355: (Only affects cord users.)
1356: - Fixed Linux code to deal with RedHat 5.0 and integrated Peter Bigot's
1357: os_dep.c code for dealing with various Linux versions.
1358: - Added workaround for Irix pthreads sigaction bug and possible signal
1359: misdirection problems.
1360: Since alpha1:
1361: - Changed RS6000 STACKBOTTOM.
1362: - Integrated Patrick Beard's Mac changes.
1363: - Alpha1 didn't compile on Irix m.n, m < 6.
1364: - Replaced Makefile.dj with a new one from Gary Leavens.
1365: - Added Andrew Stitcher's changes to support SCO OpenServer.
1366: - Added PRINT_BLACK_LIST, to allow debugging of high densities of false
1367: pointers.
1368: - Added code to debug allocator to keep track of return address
1369: in GC_malloc caller, thus giving a bit more context.
1370: - Changed default behavior of large block allocator to more
1371: aggressively avoid fragmentation. This is likely to slow down the
1372: collector when it succeeds at reducing space cost.
1373: - Integrated Fergus Henderson's CYGWIN32 changes. They are untested,
1374: but needed for newer versions.
1375: - USE_MMAP had some serious bugs. This caused the collector to fail
1376: consistently on Solaris with -DSMALL_CONFIG.
1377: - Added Linux threads support, thanks largely to Fergus Henderson.
1378: Since alpha2:
1379: - Fixed more Linux threads problems.
1380: - Changed default GC_free_space_divisor to 3 with new large block allocation.
1381: (Thanks to Matthew Flatt for some measurements that suggest the old
1382: value sometimes favors space too much over time.)
1383: - More CYGWIN32 fixes.
1384: - Integrated Tyson-Dowd's Linux-M68K port.
1385: - Minor HP PA and DEC UNIX fixes from Fergus Henderson.
1386: - Integrated Christoffe Raffali's Linux-SPARC changes.
1387: - Allowed for one more GC fixup iteration after a full GC in incremental
1388: mode. Some quick measurements suggested that this significantly
1389: reduces pause times even with smaller GC_RATE values.
1390: - Moved some more GC data structures into GC_arrays. This decreases
1391: pause times and GC overhead, but makes debugging slightly less convenient.
1392: - Fixed namespace pollution problem ("excl_table").
1393: - Made GC_incremental a constant for -DSMALL_CONFIG, hopefully shrinking
1394: that slightly.
1395: - Added some win32 threads fixes.
1396: - Integrated Ivan Demakov and David Stes' Watcom fixes.
1397: - Various other minor fixes contributed by many people.
1398: - Renamed config.h to gcconfig.h, since config.h tends to be used for
1399: many other things.
1400: - Integrated Matthew Flatt's support for 68K MacOS "far globals".
1401: - Fixed up some of the dynamic library Makefile targets for consistency
1402: across platforms.
1403: - Fixed a USE_MMAP typo that caused out-of-memory handling to fail
1404: on Solaris.
1405: - Added code to test.c to test thread creation a bit more.
1406: - Integrated GC_win32_free_heap, as suggested by Ivan Demakov.
1407: - Fixed Solaris 2.7 stack base finding problem. (This may actually
1408: have been done in an earlier alpha release.)
1409: Since alpha3:
1410: - Fixed MSWIN32 recognition test, which interfered with cygwin.
1411: - Removed unnecessary gc_watcom.asm from distribution. Removed
1412: some obsolete README.win32 text.
1413: - Added Alpha Linux incremental GC support. (Thanks to Philipp Tomsich
1414: for code for retrieving the fault address in a signal handler.)
1415: Changed Linux signal handler context argument to be a pointer.
1416: - Took care of some new warnings generated by the 7.3 SGI compiler.
1417: - Integrated Phillip Musumeci's FreeBSD/ELF fixes.
1418: - -DIRIX_THREADS was broken with the -o32 ABI (typo in gc_priv.h>
1419:
1420: Since 4.13:
1421: - Fixed GC_print_source_ptr to not use a prototype.
1422: - generalized CYGWIN test.
1423: - gc::new did the wrong thing with PointerFreeGC placement.
1424: (Thanks to Rauli Ruohonen.)
1425: - In the ALL_INTERIOR_POINTERS (default) case, some callee-save register
1426: values could fail to be scanned if the register was saved and
1427: reused in a GC frame. This showed up in verbose mode with gctest
1428: compiled with an unreleased SGI compiler. I vaguely recall an old
1429: bug report that may have been related. The bug was probably quite old.
1430: (The problem was that the stack scanning could be deferred until
1431: after the relevant frame was overwritten, and the new save location
1432: might be outside the scanned area. Fixed by more eager stack scanning.)
1433: - PRINT_BLACK_LIST had some problems. A few source addresses were garbage.
1434: - Replaced Makefile.dj and added -I flags to cord make targets.
1435: (Thanks to Gary Leavens.)
1436: - GC_try_to_collect was broken with the nonincremental collector.
1437: - gc_cleanup destructors could pass the wrong address to
1438: GC_register_finalizer_ignore_self in the presence of multiple
1439: inheritance. (Thanks to Darrell Schiebel.)
1440: - Changed PowerPC Linux stack finding code.
1441:
1442: Since 4.14alpha1
1443: - -DSMALL_CONFIG did not work reliably with large (> 4K) pages.
1444: Recycling the mark stack during expansion could result in a size
1445: zero heap segment, which confused things. (This was probably also an
1446: issue with the normal config and huge pages.)
1447: - Did more work to make sure that callee-save registers were scanned
1448: completely, even with the setjmp-based code. Added USE_GENERIC_PUSH_REGS
1449: macro to facilitate testing on machines I have access to.
1450: - Added code to explicitly push register contents for win32 threads.
1451: This seems to be necessary. (Thanks to Pierre de Rop.)
1452:
1453: Since 4.14alpha2
1454: - changed STACKBOTTOM for DJGPP (Thanks to Salvador Eduardo Tropea).
1455:
1456: Since 4.14
1457: - Reworked large block allocator. Now uses multiple doubly linked free
1458: lists to approximate best fit.
1459: - Changed heap expansion heuristic. Entirely free blocks are no longer
1460: counted towards the heap size. This seems to have a major impact on
1461: heap size stability; the old version could expand the heap way too
1462: much in the presence of large block fragmentation.
1463: - added -DGC_ASSERTIONS and some simple assertions inside the collector.
1464: This is mainlyt for collector debugging.
1465: - added -DUSE_MUNMAP to allow the heap to shrink. Suupported on only
1466: a few UNIX-like platforms for now.
1467: - added GC_dump_regions() for debugging of fragmentation issues.
1468: - Changed PowerPC pointer alignment under Linux to 4. (This needs
1469: checking by someone who has one. The suggestions came to me via a
1470: rather circuitous path.)
1471: - Changed the Linux/Alpha port to walk the data segment backwards until
1472: it encounters a SIGSEGV. The old way to find the start of the data
1473: segment broke with a recent release.
1474: - cordxtra.c needed to call GC_REGISTER_FINALIZER instead of
1475: GC_register_finalizer, so that it would continue to work with GC_DEBUG.
1476: - allochblk sometimes cleared the wrong block for debugging purposes
1477: when it dropped blacklisted blocks. This could result in spurious
1478: error reports with GC_DEBUG.
1479: - added MACOS X Server support. (Thanks to Andrew Stone.)
1480: - Changed the Solaris threads code to ignore stack limits > 8 MB with
1481: a warning. Empirically, it is not safe to access arbitrary pages
1482: in such large stacks. And the dirty bit implementation does not
1483: guarantee that none of them will be accessed.
1484: - Integrated Martin Tauchmann's Amiga changes.
1485: - Integrated James Dominy's OpenBSD/SPARC port.
1486:
1487: Since 5.0alpha1
1488: - Fixed bugs introduced in alpha1 (OpenBSD & large block initialization).
1489: - Added -DKEEP_BACK_PTRS and backptr.h interface. (The implementation
1490: idea came from Al Demers.)
1491:
1.2 ! noro 1492: Since 5.0alpha2
! 1493: - Added some highly incomplete code to support a copied young generation.
! 1494: Comments on nursery.h are appreciated.
! 1495: - Changed -DFIND_LEAK, -DJAVA_FINALIZATION, and -DFINALIZE_ON_DEMAND,
! 1496: so the same effect could be obtained with a runtime switch. This is
! 1497: a step towards standardizing on a single dynamic GC library.
! 1498: - Significantly changed the way leak detection is handled, as a consequence
! 1499: of the above.
! 1500:
! 1501: Since 5.0 alpha3
! 1502: - Added protection fault handling patch for Linux/M68K from Fergus
! 1503: Henderson and Roman Hodek.
! 1504: - Removed the tests for SGI_SOURCE in new_gc_alloc.h. This was causing that
! 1505: interface to fail on nonSGI platforms.
! 1506: - Changed the Linux stack finding code to use /proc, after chnging it
! 1507: to use HEURISTIC1. (Thanks to David Mossberger for pointing out the
! 1508: /proc hook.)
! 1509: - Added HP/UX incremental GC support and HP/UX 11 thread support.
! 1510: - Added basic Linux/IA64 support.
! 1511: - Integrated Anthony Green's PicoJava support.
! 1512: - Integrated Scott Ananian's StrongARM/NetBSD support.
! 1513: - Fixed some fairly serious performance bugs in the incremental
! 1514: collector. These have probably been there essentially forever.
! 1515: (Mark bits were sometimes set before scanning dirty pages.
! 1516: The reclaim phase unnecessarily dirtied full small object pages.)
! 1517: - Changed the reclaim phase to ignore nearly full pages to avoid
! 1518: touching them.
! 1519: - Limited GC_black_list_spacing to roughly the heap growth increment.
! 1520: - Changed full collection triggering heuristic to decrease full GC
! 1521: frequency by default, but to explicitly trigger full GCs during
! 1522: heap growth. This doesn't always improve things, but on average it's
! 1523: probably a win.
! 1524: - GC_debug_free(0, ...) failed. Thanks to Fergus Henderson for the
! 1525: bug report and fix.
! 1526:
1.1 noro 1527: To do:
1528: - Very large root set sizes (> 16 MB or so) could cause the collector
1529: to abort with an unexpected mark stack overflow. (Thanks again to
1530: Peter Chubb.) NOT YET FIXED. Workaround is to increase the initial
1531: size.
1532: - The SGI version of the collector marks from mmapped pages, even
1533: if they are not part of dynamic library static data areas. This
1534: causes performance problems with some SGI libraries that use mmap
1535: as a bitmap allocator. NOT YET FIXED. It may be possible to turn
1536: off DYNAMIC_LOADING in the collector as a workaround. It may also
1537: be possible to conditionally intercept mmap and use GC_exclude_static_roots.
1538: The real fix is to walk rld data structures, which looks possible.
1539: - Integrate MIT and DEC pthreads ports.
1.2 ! noro 1540: - Incremental collector should handle large objects better. Currently,
! 1541: it looks like the whole object is treated as dirty if any part of it
! 1542: is.
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>