Annotation of OpenXM_contrib/gc/README, Revision 1.1.1.1
1.1 maekawa 1: Copyright 1988, 1989 Hans-J. Boehm, Alan J. Demers
2: Copyright (c) 1991-1996 by Xerox Corporation. All rights reserved.
3: Copyright (c) 1996-1998 by Silicon Graphics. All rights reserved.
4:
5: THIS MATERIAL IS PROVIDED AS IS, WITH ABSOLUTELY NO WARRANTY EXPRESSED
6: OR IMPLIED. ANY USE IS AT YOUR OWN RISK.
7:
8: Permission is hereby granted to use or copy this program
9: for any purpose, provided the above notices are retained on all copies.
10: Permission to modify the code and to distribute modified code is granted,
11: provided the above notices are retained, and a notice that the code was
12: modified is included with the above copyright notice.
13:
14: This is version 4.14 of a conservative garbage collector for C and C++.
15:
16: You might find a more recent version of this at
17:
18: http://reality.sgi.com/boehm/gc.html
19:
20: HISTORY -
21:
22: Early versions of this collector were developed as a part of research
23: projects supported in part by the National Science Foundation
24: and the Defense Advance Research Projects Agency.
25: Much of the code was rewritten by Hans-J. Boehm at Xerox PARC
26: and is now maintained by him at SGI (boehm@sgi.com).
27:
28: Some other contributors:
29:
30: More recent contributors are mentioned in the modification history at the
31: end of this file. My apologies for any omissions.
32:
33: The SPARC specific code was contributed by Mark Weiser
34: (weiser@parc.xerox.com). The Encore Multimax modifications were supplied by
35: Kevin Kenny (kenny@m.cs.uiuc.edu). The adaptation to the RT is largely due
36: to Vernon Lee (scorpion@rice.edu), on machines made available by IBM.
37: Much of the HP specific code and a number of good suggestions for improving the
38: generic code are due to Walter Underwood (wunder@hp-ses.sde.hp.com).
39: Robert Brazile (brazile@diamond.bbn.com) originally supplied the ULTRIX code.
40: Al Dosser (dosser@src.dec.com) and Regis Cridlig (Regis.Cridlig@cl.cam.ac.uk)
41: subsequently provided updates and information on variation between ULTRIX
42: systems. Parag Patel (parag@netcom.com) supplied the A/UX code.
43: Jesper Peterson(jep@mtiame.mtia.oz.au) and
44: Michel Schinz supplied the Amiga port.
45: Thomas Funke (thf@zelator.in-berlin.de(?)) and
46: Brian D.Carlstrom (bdc@clark.lcs.mit.edu) supplied the NeXT ports.
47: Douglas Steel (doug@wg.icl.co.uk) provided ICL DRS6000 code.
48: Bill Janssen (janssen@parc.xerox.com) supplied the SunOS dynamic loader
49: specific code. Manuel Serrano (serrano@cornas.inria.fr) supplied linux and
50: Sony News specific code. Al Dosser provided Alpha/OSF/1 code. He and
51: Dave Detlefs(detlefs@src.dec.com) also provided several generic bug fixes.
52: Alistair G. Crooks(agc@uts.amdahl.com) supplied the NetBSD and 386BSD ports.
53: Jeffrey Hsu (hsu@soda.berkeley.edu) provided the FreeBSD port.
54: Brent Benson (brent@jade.ssd.csd.harris.com) ported the collector to
55: a Motorola 88K processor running CX/UX (Harris NightHawk).
56: Ari Huttunen (Ari.Huttunen@hut.fi) generalized the OS/2 port to
57: nonIBM development environments (a nontrivial task).
58: Patrick Beard (beard@cs.ucdavis.edu) provided the initial MacOS port.
59: David Chase, then at Olivetti Research, suggested several improvements.
60: Scott Schwartz (schwartz@groucho.cse.psu.edu) supplied some of the
61: code to save and print call stacks for leak detection on a SPARC.
62: Jesse Hull and John Ellis supplied the C++ interface code.
63: Zhong Shao performed much of the experimentation that led to the
64: current typed allocation facility. (His dynamic type inference code hasn't
65: made it into the released version of the collector, yet.)
66: (Blame for misinstallation of these modifications goes to the first author,
67: however.)
68:
69: OVERVIEW
70:
71: This is intended to be a general purpose, garbage collecting storage
72: allocator. The algorithms used are described in:
73:
74: Boehm, H., and M. Weiser, "Garbage Collection in an Uncooperative Environment",
75: Software Practice & Experience, September 1988, pp. 807-820.
76:
77: Boehm, H., A. Demers, and S. Shenker, "Mostly Parallel Garbage Collection",
78: Proceedings of the ACM SIGPLAN '91 Conference on Programming Language Design
79: and Implementation, SIGPLAN Notices 26, 6 (June 1991), pp. 157-164.
80:
81: Boehm, H., "Space Efficient Conservative Garbage Collection", Proceedings
82: of the ACM SIGPLAN '91 Conference on Programming Language Design and
83: Implementation, SIGPLAN Notices 28, 6 (June 1993), pp. 197-206.
84:
85: Possible interactions between the collector and optimizing compilers are
86: discussed in
87:
88: Boehm, H., and D. Chase, "A Proposal for GC-safe C Compilation",
89: The Journal of C Language Translation 4, 2 (December 1992).
90:
91: and
92:
93: Boehm H., "Simple GC-safe Compilation", Proceedings
94: of the ACM SIGPLAN '96 Conference on Programming Language Design and
95: Implementation.
96:
97: (Both are also available from
98: http://reality.sgi.com/boehm/papers/, among other places.)
99:
100: Unlike the collector described in the second reference, this collector
101: operates either with the mutator stopped during the entire collection
102: (default) or incrementally during allocations. (The latter is supported
103: on only a few machines.) It does not rely on threads, but is intended
104: to be thread-safe.
105:
106: Some of the ideas underlying the collector have previously been explored
107: by others. (Doug McIlroy wrote a vaguely similar collector that is part of
108: version 8 UNIX (tm).) However none of this work appears to have been widely
109: disseminated.
110:
111: Rudimentary tools for use of the collector as a leak detector are included, as
112: is a fairly sophisticated string package "cord" that makes use of the collector.
113: (See cord/README.)
114:
115:
116: GENERAL DESCRIPTION
117:
118: This is a garbage collecting storage allocator that is intended to be
119: used as a plug-in replacement for C's malloc.
120:
121: Since the collector does not require pointers to be tagged, it does not
122: attempt to ensure that all inaccessible storage is reclaimed. However,
123: in our experience, it is typically more successful at reclaiming unused
124: memory than most C programs using explicit deallocation. Unlike manually
125: introduced leaks, the amount of unreclaimed memory typically stays
126: bounded.
127:
128: In the following, an "object" is defined to be a region of memory allocated
129: by the routines described below.
130:
131: Any objects not intended to be collected must be pointed to either
132: from other such accessible objects, or from the registers,
133: stack, data, or statically allocated bss segments. Pointers from
134: the stack or registers may point to anywhere inside an object.
135: The same is true for heap pointers if the collector is compiled with
136: ALL_INTERIOR_POINTERS defined, as is now the default.
137:
138: Compiling without ALL_INTERIOR_POINTERS may reduce accidental retention
139: of garbage objects, by requiring pointers from the heap to to the beginning
140: of an object. But this no longer appears to be a significant
141: issue for most programs.
142:
143: There are a number of routines which modify the pointer recognition
144: algorithm. GC_register_displacement allows certain interior pointers
145: to be recognized even if ALL_INTERIOR_POINTERS is nor defined.
146: GC_malloc_ignore_off_page allows some pointers into the middle of large objects
147: to be disregarded, greatly reducing the probablility of accidental
148: retention of large objects. For most purposes it seems best to compile
149: with ALL_INTERIOR_POINTERS and to use GC_malloc_ignore_off_page if
150: you get collector warnings from allocations of very large objects.
151: See README.debugging for details.
152:
153: Note that pointers inside memory allocated by the standard "malloc" are not
154: seen by the garbage collector. Thus objects pointed to only from such a
155: region may be prematurely deallocated. It is thus suggested that the
156: standard "malloc" be used only for memory regions, such as I/O buffers, that
157: are guaranteed not to contain pointers to garbage collectable memory.
158: Pointers in C language automatic, static, or register variables,
159: are correctly recognized. (Note that GC_malloc_uncollectable has semantics
160: similar to standard malloc, but allocates objects that are traced by the
161: collector.)
162:
163: The collector does not always know how to find pointers in data
164: areas that are associated with dynamic libraries. This is easy to
165: remedy IF you know how to find those data areas on your operating
166: system (see GC_add_roots). Code for doing this under SunOS, IRIX 5.X and 6.X,
167: HP/UX, Alpha OSF/1, Linux, and win32 is included and used by default. (See
168: README.win32 for win32 details.) On other systems pointers from dynamic
169: library data areas may not be considered by the collector.
170:
171: Note that the garbage collector does not need to be informed of shared
172: read-only data. However if the shared library mechanism can introduce
173: discontiguous data areas that may contain pointers, then the collector does
174: need to be informed.
175:
176: Signal processing for most signals may be deferred during collection,
177: and during uninterruptible parts of the allocation process. Unlike
178: standard ANSI C mallocs, it can be safe to invoke malloc
179: from a signal handler while another malloc is in progress, provided
180: the original malloc is not restarted. (Empirically, many UNIX
181: applications already assume this.) To obtain this level of signal
182: safety, remove the definition of -DNO_SIGNALS in Makefile. This incurs
183: a minor performance penalty, and hence is no longer the default.
184:
185: The allocator/collector can also be configured for thread-safe operation.
186: (Full signal safety can also be achieved, but only at the cost of two system
187: calls per malloc, which is usually unacceptable.)
188:
189: INSTALLATION AND PORTABILITY
190:
191: As distributed, the macro SILENT is defined in Makefile.
192: In the event of problems, this can be removed to obtain a moderate
193: amount of descriptive output for each collection.
194: (The given statistics exhibit a few peculiarities.
195: Things don't appear to add up for a variety of reasons, most notably
196: fragmentation losses. These are probably much more significant for the
197: contrived program "test.c" than for your application.)
198:
199: Note that typing "make test" will automatically build the collector
200: and then run setjmp_test and gctest. Setjmp_test will give you information
201: about configuring the collector, which is useful primarily if you have
202: a machine that's not already supported. Gctest is a somewhat superficial
203: test of collector functionality. Failure is indicated by a core dump or
204: a message to the effect that the collector is broken. Gctest takes about
205: 35 seconds to run on a SPARCstation 2. On a slower machine,
206: expect it to take a while. It may use up to 8 MB of memory. (The
207: multi-threaded version will use more.) "Make test" will also, as
208: its last step, attempt to build and test the "cord" string library.
209: This will fail without an ANSI C compiler.
210:
211: The Makefile will generate a library gc.a which you should link against.
212: Typing "make cords" will add the cord library to gc.a.
213: Note that this requires an ANSI C compiler.
214:
215: It is suggested that if you need to replace a piece of the collector
216: (e.g. GC_mark_rts.c) you simply list your version ahead of gc.a on the
217: work.)
218: ld command line, rather than replacing the one in gc.a. (This will
219: generate numerous warnings under some versions of AIX, but it still
220: works.)
221:
222: All include files that need to be used by clients will be put in the
223: include subdirectory. (Normally this is just gc.h. "Make cords" adds
224: "cord.h" and "ec.h".)
225:
226: The collector currently is designed to run essentially unmodified on
227: machines that use a flat 32-bit or 64-bit address space.
228: That includes the vast majority of Workstations and X86 (X >= 3) PCs.
229: (The list here was deleted because it was getting too long and constantly
230: out of date.)
231: It does NOT run under plain 16-bit DOS or Windows 3.X. There are however
232: various packages (e.g. win32s, djgpp) that allow flat 32-bit address
233: applications to run under those systemsif the have at least an 80386 processor,
234: and several of those are compatible with the collector.
235:
236: In a few cases (Amiga, OS/2, Win32, MacOS) a separate makefile
237: or equivalent is supplied. Many of these have separate README.system
238: files.
239:
240: Dynamic libraries are completely supported only under SunOS
241: (and even that support is not functional on the last Sun 3 release),
242: IRIX 5&6, HP-PA, Win32 (not Win32S) and OSF/1 on DEC AXP machines.
243: On other machines we recommend that you do one of the following:
244:
245: 1) Add dynamic library support (and send us the code).
246: 2) Use static versions of the libraries.
247: 3) Arrange for dynamic libraries to use the standard malloc.
248: This is still dangerous if the library stores a pointer to a
249: garbage collected object. But nearly all standard interfaces
250: prohibit this, because they deal correctly with pointers
251: to stack allocated objects. (Strtok is an exception. Don't
252: use it.)
253:
254: In all cases we assume that pointer alignment is consistent with that
255: enforced by the standard C compilers. If you use a nonstandard compiler
256: you may have to adjust the alignment parameters defined in gc_priv.h.
257:
258: A port to a machine that is not byte addressed, or does not use 32 bit
259: or 64 bit addresses will require a major effort. A port to plain MSDOS
260: or win16 is hard.
261:
262: For machines not already mentioned, or for nonstandard compilers, the
263: following are likely to require change:
264:
265: 1. The parameters in gcconfig.h.
266: The parameters that will usually require adjustment are
267: STACKBOTTOM, ALIGNMENT and DATASTART. Setjmp_test
268: prints its guesses of the first two.
269: DATASTART should be an expression for computing the
270: address of the beginning of the data segment. This can often be
271: &etext. But some memory management units require that there be
272: some unmapped space between the text and the data segment. Thus
273: it may be more complicated. On UNIX systems, this is rarely
274: documented. But the adb "$m" command may be helpful. (Note
275: that DATASTART will usually be a function of &etext. Thus a
276: single experiment is usually insufficient.)
277: STACKBOTTOM is used to initialize GC_stackbottom, which
278: should be a sufficient approximation to the coldest stack address.
279: On some machines, it is difficult to obtain such a value that is
280: valid across a variety of MMUs, OS releases, etc. A number of
281: alternatives exist for using the collector in spite of this. See the
282: discussion in gcconfig.h immediately preceding the various
283: definitions of STACKBOTTOM.
284:
285: 2. mach_dep.c.
286: The most important routine here is one to mark from registers.
287: The distributed file includes a generic hack (based on setjmp) that
288: happens to work on many machines, and may work on yours. Try
289: compiling and running setjmp_t.c to see whether it has a chance of
290: working. (This is not correct C, so don't blame your compiler if it
291: doesn't work. Based on limited experience, register window machines
292: are likely to cause trouble. If your version of setjmp claims that
293: all accessible variables, including registers, have the value they
294: had at the time of the longjmp, it also will not work. Vanilla 4.2 BSD
295: on Vaxen makes such a claim. SunOS does not.)
296: If your compiler does not allow in-line assembly code, or if you prefer
297: not to use such a facility, mach_dep.c may be replaced by a .s file
298: (as we did for the MIPS machine and the PC/RT).
299: At this point enough architectures are supported by mach_dep.c
300: that you will rarely need to do more than adjust for assembler
301: syntax.
302:
303: 3. os_dep.c (and gc_priv.h).
304: Several kinds of operating system dependent routines reside here.
305: Many are optional. Several are invoked only through corresponding
306: macros in gc_priv.h, which may also be redefined as appropriate.
307: The routine GC_register_data_segments is crucial. It registers static
308: data areas that must be traversed by the collector. (User calls to
309: GC_add_roots may sometimes be used for similar effect.)
310: Routines to obtain memory from the OS also reside here.
311: Alternatively this can be done entirely by the macro GET_MEM
312: defined in gc_priv.h. Routines to disable and reenable signals
313: also reside here if they are need by the macros DISABLE_SIGNALS
314: and ENABLE_SIGNALS defined in gc_priv.h.
315: In a multithreaded environment, the macros LOCK and UNLOCK
316: in gc_priv.h will need to be suitably redefined.
317: The incremental collector requires page dirty information, which
318: is acquired through routines defined in os_dep.c. Unless directed
319: otherwise by gcconfig.h, these are implemented as stubs that simply
320: treat all pages as dirty. (This of course makes the incremental
321: collector much less useful.)
322:
323: 4. dyn_load.c
324: This provides a routine that allows the collector to scan data
325: segments associated with dynamic libraries. Often it is not
326: necessary to provide this routine unless user-written dynamic
327: libraries are used.
328:
329: For a different version of UN*X or different machines using the
330: Motorola 68000, Vax, SPARC, 80386, NS 32000, PC/RT, or MIPS architecture,
331: it should frequently suffice to change definitions in gcconfig.h.
332:
333:
334: THE C INTERFACE TO THE ALLOCATOR
335:
336: The following routines are intended to be directly called by the user.
337: Note that usually only GC_malloc is necessary. GC_clear_roots and GC_add_roots
338: calls may be required if the collector has to trace from nonstandard places
339: (e.g. from dynamic library data areas on a machine on which the
340: collector doesn't already understand them.) On some machines, it may
341: be desirable to set GC_stacktop to a good approximation of the stack base.
342: (This enhances code portability on HP PA machines, since there is no
343: good way for the collector to compute this value.) Client code may include
344: "gc.h", which defines all of the following, plus many others.
345:
346: 1) GC_malloc(nbytes)
347: - allocate an object of size nbytes. Unlike malloc, the object is
348: cleared before being returned to the user. Gc_malloc will
349: invoke the garbage collector when it determines this to be appropriate.
350: GC_malloc may return 0 if it is unable to acquire sufficient
351: space from the operating system. This is the most probable
352: consequence of running out of space. Other possible consequences
353: are that a function call will fail due to lack of stack space,
354: or that the collector will fail in other ways because it cannot
355: maintain its internal data structures, or that a crucial system
356: process will fail and take down the machine. Most of these
357: possibilities are independent of the malloc implementation.
358:
359: 2) GC_malloc_atomic(nbytes)
360: - allocate an object of size nbytes that is guaranteed not to contain any
361: pointers. The returned object is not guaranteed to be cleared.
362: (Can always be replaced by GC_malloc, but results in faster collection
363: times. The collector will probably run faster if large character
364: arrays, etc. are allocated with GC_malloc_atomic than if they are
365: statically allocated.)
366:
367: 3) GC_realloc(object, new_size)
368: - change the size of object to be new_size. Returns a pointer to the
369: new object, which may, or may not, be the same as the pointer to
370: the old object. The new object is taken to be atomic iff the old one
371: was. If the new object is composite and larger than the original object,
372: then the newly added bytes are cleared (we hope). This is very likely
373: to allocate a new object, unless MERGE_SIZES is defined in gc_priv.h.
374: Even then, it is likely to recycle the old object only if the object
375: is grown in small additive increments (which, we claim, is generally bad
376: coding practice.)
377:
378: 4) GC_free(object)
379: - explicitly deallocate an object returned by GC_malloc or
380: GC_malloc_atomic. Not necessary, but can be used to minimize
381: collections if performance is critical. Probably a performance
382: loss for very small objects (<= 8 bytes).
383:
384: 5) GC_expand_hp(bytes)
385: - Explicitly increase the heap size. (This is normally done automatically
386: if a garbage collection failed to GC_reclaim enough memory. Explicit
387: calls to GC_expand_hp may prevent unnecessarily frequent collections at
388: program startup.)
389:
390: 6) GC_malloc_ignore_off_page(bytes)
391: - identical to GC_malloc, but the client promises to keep a pointer to
392: the somewhere within the first 256 bytes of the object while it is
393: live. (This pointer should nortmally be declared volatile to prevent
394: interference from compiler optimizations.) This is the recommended
395: way to allocate anything that is likely to be larger than 100Kbytes
396: or so. (GC_malloc may result in failure to reclaim such objects.)
397:
398: 7) GC_set_warn_proc(proc)
399: - Can be used to redirect warnings from the collector. Such warnings
400: should be rare, and should not be ignored during code development.
401:
402: 8) GC_enable_incremental()
403: - Enables generational and incremental collection. Useful for large
404: heaps on machines that provide access to page dirty information.
405: Some dirty bit implementations may interfere with debugging
406: (by catching address faults) and place restrictions on heap arguments
407: to system calls (since write faults inside a system call may not be
408: handled well).
409:
410: 9) Several routines to allow for registration of finalization code.
411: User supplied finalization code may be invoked when an object becomes
412: unreachable. To call (*f)(obj, x) when obj becomes inaccessible, use
413: GC_register_finalizer(obj, f, x, 0, 0);
414: For more sophisticated uses, and for finalization ordering issues,
415: see gc.h.
416:
417: The global variable GC_free_space_divisor may be adjusted up from its
418: default value of 4 to use less space and more collection time, or down for
419: the opposite effect. Setting it to 1 or 0 will effectively disable collections
420: and cause all allocations to simply grow the heap.
421:
422: The variable GC_non_gc_bytes, which is normally 0, may be changed to reflect
423: the amount of memory allocated by the above routines that should not be
424: considered as a candidate for collection. Careless use may, of course, result
425: in excessive memory consumption.
426:
427: Some additional tuning is possible through the parameters defined
428: near the top of gc_priv.h.
429:
430: If only GC_malloc is intended to be used, it might be appropriate to define:
431:
432: #define malloc(n) GC_malloc(n)
433: #define calloc(m,n) GC_malloc((m)*(n))
434:
435: For small pieces of VERY allocation intensive code, gc_inl.h
436: includes some allocation macros that may be used in place of GC_malloc
437: and friends.
438:
439: All externally visible names in the garbage collector start with "GC_".
440: To avoid name conflicts, client code should avoid this prefix, except when
441: accessing garbage collector routines or variables.
442:
443: There are provisions for allocation with explicit type information.
444: This is rarely necessary. Details can be found in gc_typed.h.
445:
446: THE C++ INTERFACE TO THE ALLOCATOR:
447:
448: The Ellis-Hull C++ interface to the collector is included in
449: the collector distribution. If you intend to use this, type
450: "make c++" after the initial build of the collector is complete.
451: See gc_cpp.h for the definition of the interface. This interface
452: tries to approximate the Ellis-Detlefs C++ garbage collection
453: proposal without compiler changes.
454:
455: Cautions:
456: 1. Arrays allocated without new placement syntax are
457: allocated as uncollectable objects. They are traced by the
458: collector, but will not be reclaimed.
459:
460: 2. Failure to use "make c++" in combination with (1) will
461: result in arrays allocated using the default new operator.
462: This is likely to result in disaster without linker warnings.
463:
464: 3. If your compiler supports an overloaded new[] operator,
465: then gc_cpp.cc and gc_cpp.h should be suitably modified.
466:
467: 4. Many current C++ compilers have deficiencies that
468: break some of the functionality. See the comments in gc_cpp.h
469: for suggested workarounds.
470:
471: USE AS LEAK DETECTOR:
472:
473: The collector may be used to track down leaks in C programs that are
474: intended to run with malloc/free (e.g. code with extreme real-time or
475: portability constraints). To do so define FIND_LEAK in Makefile
476: This will cause the collector to invoke the report_leak
477: routine defined near the top of reclaim.c whenever an inaccessible
478: object is found that has not been explicitly freed. The collector will
479: no longer reclaim inaccessible memory; in this form it is purely a
480: debugging tool.
481: Productive use of this facility normally involves redefining report_leak
482: to do something more intelligent. This typically requires annotating
483: objects with additional information (e.g. creation time stack trace) that
484: identifies their origin. Such code is typically not very portable, and is
485: not included here, except on SPARC machines.
486: If all objects are allocated with GC_DEBUG_MALLOC (see next section),
487: then the default version of report_leak will report the source file
488: and line number at which the leaked object was allocated. This may
489: sometimes be sufficient. (On SPARC/SUNOS4 machines, it will also report
490: a cryptic stack trace. This can often be turned into a sympolic stack
491: trace by invoking program "foo" with "callprocs foo". Callprocs is
492: a short shell script that invokes adb to expand program counter values
493: to symbolic addresses. It was largely supplied by Scott Schwartz.)
494: Note that the debugging facilities described in the next section can
495: sometimes be slightly LESS effective in leak finding mode, since in
496: leak finding mode, GC_debug_free actually results in reuse of the object.
497: (Otherwise the object is simply marked invalid.) Also note that the test
498: program is not designed to run meaningfully in FIND_LEAK mode.
499: Use "make gc.a" to build the collector.
500:
501: DEBUGGING FACILITIES:
502:
503: The routines GC_debug_malloc, GC_debug_malloc_atomic, GC_debug_realloc,
504: and GC_debug_free provide an alternate interface to the collector, which
505: provides some help with memory overwrite errors, and the like.
506: Objects allocated in this way are annotated with additional
507: information. Some of this information is checked during garbage
508: collections, and detected inconsistencies are reported to stderr.
509:
510: Simple cases of writing past the end of an allocated object should
511: be caught if the object is explicitly deallocated, or if the
512: collector is invoked while the object is live. The first deallocation
513: of an object will clear the debugging info associated with an
514: object, so accidentally repeated calls to GC_debug_free will report the
515: deallocation of an object without debugging information. Out of
516: memory errors will be reported to stderr, in addition to returning
517: NIL.
518:
519: GC_debug_malloc checking during garbage collection is enabled
520: with the first call to GC_debug_malloc. This will result in some
521: slowdown during collections. If frequent heap checks are desired,
522: this can be achieved by explicitly invoking GC_gcollect, e.g. from
523: the debugger.
524:
525: GC_debug_malloc allocated objects should not be passed to GC_realloc
526: or GC_free, and conversely. It is however acceptable to allocate only
527: some objects with GC_debug_malloc, and to use GC_malloc for other objects,
528: provided the two pools are kept distinct. In this case, there is a very
529: low probablility that GC_malloc allocated objects may be misidentified as
530: having been overwritten. This should happen with probability at most
531: one in 2**32. This probability is zero if GC_debug_malloc is never called.
532:
533: GC_debug_malloc, GC_malloc_atomic, and GC_debug_realloc take two
534: additional trailing arguments, a string and an integer. These are not
535: interpreted by the allocator. They are stored in the object (the string is
536: not copied). If an error involving the object is detected, they are printed.
537:
538: The macros GC_MALLOC, GC_MALLOC_ATOMIC, GC_REALLOC, GC_FREE, and
539: GC_REGISTER_FINALIZER are also provided. These require the same arguments
540: as the corresponding (nondebugging) routines. If gc.h is included
541: with GC_DEBUG defined, they call the debugging versions of these
542: functions, passing the current file name and line number as the two
543: extra arguments, where appropriate. If gc.h is included without GC_DEBUG
544: defined, then all these macros will instead be defined to their nondebugging
545: equivalents. (GC_REGISTER_FINALIZER is necessary, since pointers to
546: objects with debugging information are really pointers to a displacement
547: of 16 bytes form the object beginning, and some translation is necessary
548: when finalization routines are invoked. For details, about what's stored
549: in the header, see the definition of the type oh in debug_malloc.c)
550:
551: INCREMENTAL/GENERATIONAL COLLECTION:
552:
553: The collector normally interrupts client code for the duration of
554: a garbage collection mark phase. This may be unacceptable if interactive
555: response is needed for programs with large heaps. The collector
556: can also run in a "generational" mode, in which it usually attempts to
557: collect only objects allocated since the last garbage collection.
558: Furthermore, in this mode, garbage collections run mostly incrementally,
559: with a small amount of work performed in response to each of a large number of
560: GC_malloc requests.
561:
562: This mode is enabled by a call to GC_enable_incremental().
563:
564: Incremental and generational collection is effective in reducing
565: pause times only if the collector has some way to tell which objects
566: or pages have been recently modified. The collector uses two sources
567: of information:
568:
569: 1. Information provided by the VM system. This may be provided in
570: one of several forms. Under Solaris 2.X (and potentially under other
571: similar systems) information on dirty pages can be read from the
572: /proc file system. Under other systems (currently SunOS4.X) it is
573: possible to write-protect the heap, and catch the resulting faults.
574: On these systems we require that system calls writing to the heap
575: (other than read) be handled specially by client code.
576: See os_dep.c for details.
577:
578: 2. Information supplied by the programmer. We define "stubborn"
579: objects to be objects that are rarely changed. Such an object
580: can be allocated (and enabled for writing) with GC_malloc_stubborn.
581: Once it has been initialized, the collector should be informed with
582: a call to GC_end_stubborn_change. Subsequent writes that store
583: pointers into the object must be preceded by a call to
584: GC_change_stubborn.
585:
586: This mechanism performs best for objects that are written only for
587: initialization, and such that only one stubborn object is writable
588: at once. It is typically not worth using for short-lived
589: objects. Stubborn objects are treated less efficiently than pointerfree
590: (atomic) objects.
591:
592: A rough rule of thumb is that, in the absence of VM information, garbage
593: collection pauses are proportional to the amount of pointerful storage
594: plus the amount of modified "stubborn" storage that is reachable during
595: the collection.
596:
597: Initial allocation of stubborn objects takes longer than allocation
598: of other objects, since other data structures need to be maintained.
599:
600: We recommend against random use of stubborn objects in client
601: code, since bugs caused by inappropriate writes to stubborn objects
602: are likely to be very infrequently observed and hard to trace.
603: However, their use may be appropriate in a few carefully written
604: library routines that do not make the objects themselves available
605: for writing by client code.
606:
607:
608: BUGS:
609:
610: Any memory that does not have a recognizable pointer to it will be
611: reclaimed. Exclusive-or'ing forward and backward links in a list
612: doesn't cut it.
613: Some C optimizers may lose the last undisguised pointer to a memory
614: object as a consequence of clever optimizations. This has almost
615: never been observed in practice. Send mail to boehm@sgi.com
616: for suggestions on how to fix your compiler.
617: This is not a real-time collector. In the standard configuration,
618: percentage of time required for collection should be constant across
619: heap sizes. But collection pauses will increase for larger heaps.
620: (On SPARCstation 2s collection times will be on the order of 300 msecs
621: per MB of accessible memory that needs to be scanned. Your mileage
622: may vary.) The incremental/generational collection facility helps,
623: but is portable only if "stubborn" allocation is used.
624: Please address bug reports to boehm@sgi.com. If you are
625: contemplating a major addition, you might also send mail to ask whether
626: it's already been done (or whether we tried and discarded it).
627:
628: RECENT VERSIONS:
629:
630: Version 1.3 and immediately preceding versions contained spurious
631: assembly language assignments to TMP_SP. Only the assignment in the PC/RT
632: code is necessary. On other machines, with certain compiler options,
633: the assignments can lead to an unsaved register being overwritten.
634: Known to cause problems under SunOS 3.5 WITHOUT the -O option. (With
635: -O the compiler recognizes it as dead code. It probably shouldn't,
636: but that's another story.)
637:
638: Version 1.4 and earlier versions used compile time determined values
639: for the stack base. This no longer works on Sun 3s, since Sun 3/80s use
640: a different stack base. We now use a straightforward heuristic on all
641: machines on which it is known to work (incl. Sun 3s) and compile-time
642: determined values for the rest. There should really be library calls
643: to determine such values.
644:
645: Version 1.5 and earlier did not ensure 8 byte alignment for objects
646: allocated on a sparc based machine.
647:
648: Version 1.8 added ULTRIX support in gc_private.h.
649:
650: Version 1.9 fixed a major bug in gc_realloc.
651:
652: Version 2.0 introduced a consistent naming convention for collector
653: routines and added support for registering dynamic library data segments
654: in the standard mark_roots.c. Most of the data structures were revamped.
655: The treatment of interior pointers was completely changed. Finalization
656: was added. Support for locking was added. Object kinds were added.
657: We added a black listing facility to avoid allocating at addresses known
658: to occur as integers somewhere in the address space. Much of this
659: was accomplished by adapting ideas and code from the PCR collector.
660: The test program was changed and expanded.
661:
662: Version 2.1 was the first stable version since 1.9, and added support
663: for PPCR.
664:
665: Version 2.2 added debugging allocation, and fixed various bugs. Among them:
666: - GC_realloc could fail to extend the size of the object for certain large object sizes.
667: - A blatant subscript range error in GC_printf, which unfortunately
668: wasn't exercised on machines with sufficient stack alignment constraints.
669: - GC_register_displacement did the wrong thing if it was called after
670: any allocation had taken place.
671: - The leak finding code would eventually break after 2048 byte
672: byte objects leaked.
673: - interface.c didn't compile.
674: - The heap size remained much too small for large stacks.
675: - The stack clearing code behaved badly for large stacks, and perhaps
676: on HP/PA machines.
677:
678: Version 2.3 added ALL_INTERIOR_POINTERS and fixed the following bugs:
679: - Missing declaration of etext in the A/UX version.
680: - Some PCR root-finding problems.
681: - Blacklisting was not 100% effective, because the plausible future
682: heap bounds were being miscalculated.
683: - GC_realloc didn't handle out-of-memory correctly.
684: - GC_base could return a nonzero value for addresses inside free blocks.
685: - test.c wasn't really thread safe, and could erroneously report failure
686: in a multithreaded environment. (The locking primitives need to be
687: replaced for other threads packages.)
688: - GC_CONS was thoroughly broken.
689: - On a SPARC with dynamic linking, signals stayed diabled while the
690: client code was running.
691: (Thanks to Manuel Serrano at INRIA for reporting the last two.)
692:
693: Version 2.4 added GC_free_space_divisor as a tuning knob, added
694: support for OS/2 and linux, and fixed the following bugs:
695: - On machines with unaligned pointers (e.g. Sun 3), every 128th word could
696: fail to be considered for marking.
697: - Dynamic_load.c erroneously added 4 bytes to the length of the data and
698: bss sections of the dynamic library. This could result in a bad memory
699: reference if the actual length was a multiple of a page. (Observed on
700: Sun 3. Can probably also happen on a Sun 4.)
701: (Thanks to Robert Brazile for pointing out that the Sun 3 version
702: was broken. Dynamic library handling is still broken on Sun 3s
703: under 4.1.1U1, but apparently not 4.1.1. If you have such a machine,
704: use -Bstatic.)
705:
706: Version 2.5 fixed the following bugs:
707: - Removed an explicit call to exit(1)
708: - Fixed calls to GC_printf and GC_err_printf, so the correct number of
709: arguments are always supplied. The OS/2 C compiler gets confused if
710: the number of actuals and the number of formals differ. (ANSI C
711: doesn't require this to work. The ANSI sanctioned way of doing things
712: causes too many compatibility problems.)
713:
714: Version 3.0 added generational/incremental collection and stubborn
715: objects.
716:
717: Version 3.1 added the following features:
718: - A workaround for a SunOS 4.X SPARC C compiler
719: misfeature that caused problems when the collector was turned into
720: a dynamic library.
721: - A fix for a bug in GC_base that could result in a memory fault.
722: - A fix for a performance bug (and several other misfeatures) pointed
723: out by Dave Detlefs and Al Dosser.
724: - Use of dirty bit information for static data under Solaris 2.X.
725: - DEC Alpha/OSF1 support (thanks to Al Dosser).
726: - Incremental collection on more platforms.
727: - A more refined heap expansion policy. Less space usage by default.
728: - Various minor enhancements to reduce space usage, and to reduce
729: the amount of memory scanned by the collector.
730: - Uncollectable allocation without per object overhead.
731: - More conscientious handling of out-of-memory conditions.
732: - Fixed a bug in debugging stubborn allocation.
733: - Fixed a bug that resulted in occasional erroneous reporting of smashed
734: objects with debugging allocation.
735: - Fixed bogus leak reports of size 4096 blocks with FIND_LEAK.
736:
737: Version 3.2 fixed a serious and not entirely repeatable bug in
738: the incremental collector. It appeared only when dirty bit info
739: on the roots was available, which is normally only under Solaris.
740: It also added GC_general_register_disappearing_link, and some
741: testing code. Interface.c disappeared.
742:
743: Version 3.3 fixes several bugs and adds new ports:
744: - PCR-specific bugs.
745: - Missing locking in GC_free, redundant FASTUNLOCK
746: in GC_malloc_stubborn, and 2 bugs in
747: GC_unregister_disappearing_link.
748: All of the above were pointed out by Neil Sharman
749: (neil@cs.mu.oz.au).
750: - Common symbols allocated by the SunOS4.X dynamic loader
751: were not included in the root set.
752: - Bug in GC_finalize (reported by Brian Beuning and Al Dosser)
753: - Merged Amiga port from Jesper Peterson (untested)
754: - Merged NeXT port from Thomas Funke (significantly
755: modified and untested)
756:
757: Version 3.4:
758: - Fixed a performance bug in GC_realloc.
759: - Updated the amiga port.
760: - Added NetBSD and 386BSD ports.
761: - Added cord library.
762: - Added trivial performance enhancement for
763: ALL_INTERIOR_POINTERS. (Don't scan last word.)
764:
765: Version 3.5
766: - Minor collections now mark from roots only once, if that
767: doesn't cause an excessive pause.
768: - The stack clearing heuristic was refined to prevent anomalies
769: with very heavily recursive programs and sparse stacks.
770: - Fixed a bug that prevented mark stack growth in some cases.
771: GC_objects_are_marked should be set to TRUE after a call
772: to GC_push_roots and as part of GC_push_marked, since
773: both can now set mark bits. I think this is only a performance
774: bug, but I wouldn't bet on it. It's certainly very hard to argue
775: that the old version was correct.
776: - Fixed an incremental collection bug that prevented it from
777: working at all when HBLKSIZE != getpagesize()
778: - Changed dynamic_loading.c to include gc_priv.h before testing
779: DYNAMIC_LOADING. SunOS dynamic library scanning
780: must have been broken in 3.4.
781: - Object size rounding now adapts to program behavior.
782: - Added a workaround (provided by Manuel Serrano and
783: colleagues) to a long-standing SunOS 4.X (and 3.X?) ld bug
784: that I had incorrectly assumed to have been squished.
785: The collector was broken if the text segment size was within
786: 32 bytes of a multiple of 8K bytes, and if the beginning of
787: the data segment contained interesting roots. The workaround
788: assumes a demand-loadable executable. The original may have
789: have "worked" in some other cases.
790: - Added dynamic library support under IRIX5.
791: - Added support for EMX under OS/2 (thanks to Ari Huttunen).
792:
793: Version 3.6:
794: - fixed a bug in the mark stack growth code that was introduced
795: in 3.4.
796: - fixed Makefile to work around DEC AXP compiler tail recursion
797: bug.
798:
799: Version 3.7:
800: - Added a workaround for an HP/UX compiler bug.
801: - Fixed another stack clearing performance bug. Reworked
802: that code once more.
803:
804: Version 4.0:
805: - Added support for Solaris threads (which was possible
806: only by reimplementing some fraction of Solaris threads,
807: since Sun doesn't currently make the thread debugging
808: interface available).
809: - Added non-threads win32 and win32S support.
810: - (Grudgingly, with suitable muttering of obscenities) renamed
811: files so that the collector distribution could live on a FAT
812: file system. Files that are guaranteed to be useless on
813: a PC still have long names. Gc_inline.h and gc_private.h
814: still exist, but now just include gc_inl.h and gc_priv.h.
815: - Fixed a really obscure bug in finalization that could cause
816: undetected mark stack overflows. (I would be surprised if
817: any real code ever tickled this one.)
818: - Changed finalization code to dynamically resize the hash
819: tables it maintains. (This probably does not matter for well-
820: -written code. It no doubt does for C++ code that overuses
821: destructors.)
822: - Added typed allocation primitives. Rewrote the marker to
823: accommodate them with more reasonable efficiency. This
824: change should also speed up marking for GC_malloc allocated
825: objects a little. See gc_typed.h for new primitives.
826: - Improved debugging facilities slightly. Allocation time
827: stack traces are now kept by default on SPARC/SUNOS4.
828: (Thanks to Scott Schwartz.)
829: - Added better support for small heap applications.
830: - Significantly extended cord package. Fixed a bug in the
831: implementation of lazily read files. Printf and friends now
832: have cord variants. Cord traversals are a bit faster.
833: - Made ALL_INTERIOR_POINTERS recognition the default.
834: - Fixed de so that it can run in constant space, independent
835: of file size. Added simple string searching to cords and de.
836: - Added the Hull-Ellis C++ interface.
837: - Added dynamic library support for OSF/1.
838: (Thanks to Al Dosser and Tim Bingham at DEC.)
839: - Changed argument to GC_expand_hp to be expressed
840: in units of bytes instead of heap blocks. (Necessary
841: since the heap block size now varies depending on
842: configuration. The old version was never very clean.)
843: - Added GC_get_heap_size(). The previous "equivalent"
844: was broken.
845: - Restructured the Makefile a bit.
846:
847: Since version 4.0:
848: - Changed finalization implementation to guarantee that
849: finalization procedures are called outside of the allocation
850: lock, making direct use of the interface a little less dangerous.
851: MAY BREAK EXISTING CLIENTS that assume finalizers
852: are protected by a lock. Since there seem to be few multithreaded
853: clients that use finalization, this is hopefully not much of
854: a problem.
855: - Fixed a gross bug in CORD_prev.
856: - Fixed a bug in blacklst.c that could result in unbounded
857: heap growth during startup on machines that do not clear
858: memory obtained from the OS (e.g. win32S).
859: - Ported de editor to win32/win32S. (This is now the only
860: version with a mouse-sensitive UI.)
861: - Added GC_malloc_ignore_off_page to allocate large arrays
862: in the presence of ALL_INTERIOR_POINTERS.
863: - Changed GC_call_with_alloc_lock to not disable signals in
864: the single-threaded case.
865: - Reduced retry count in GC_collect_or_expand for garbage
866: collecting when out of memory.
867: - Made uncollectable allocations bypass black-listing, as they
868: should.
869: - Fixed a bug in typed_test in test.c that could cause (legitimate)
870: GC crashes.
871: - Fixed some potential synchronization problems in finalize.c
872: - Fixed a real locking problem in typd_mlc.c.
873: - Worked around an AIX 3.2 compiler feature that results in
874: out of bounds memory references.
875: - Partially worked around an IRIX5.2 beta problem (which may
876: or may not persist to the final release).
877: - Fixed a bug in the heap integrity checking code that could
878: result in explicitly deallocated objects being identified as
879: smashed. Fixed a bug in the dbg_mlc stack saving code
880: that caused old argument pointers to be considered live.
881: - Fixed a bug in CORD_ncmp (and hence CORD_str).
882: - Repaired the OS2 port, which had suffered from bit rot
883: in 4.0. Worked around what appears to be CSet/2 V1.0
884: optimizer bug.
885: - Fixed a Makefile bug for target "c++".
886:
887: Since version 4.1:
888: - Multiple bug fixes/workarounds in the Solaris threads version.
889: (It occasionally failed to locate some register contents for
890: marking. It also turns out that thr_suspend and friends are
891: unreliable in Solaris 2.3. Dirty bit reads appear
892: to be unreliable under some weird
893: circumstances. My stack marking code
894: contained a serious performance bug. The new code is
895: extremely defensive, and has not failed in several cpu
896: hours of testing. But no guarantees ...)
897: - Added MacOS support (thanks to Patrick Beard.)
898: - Fixed several syntactic bugs in gc_c++.h and friends. (These
899: didn't bother g++, but did bother most other compilers.)
900: Fixed gc_c++.h finalization interface. (It didn't.)
901: - 64 bit alignment for allocated objects was not guaranteed in a
902: few cases in which it should have been.
903: - Added GC_malloc_atomic_ignore_off_page.
904: - Added GC_collect_a_little.
905: - Added some prototypes to gc.h.
906: - Some other minor bug fixes (notably in Makefile).
907: - Fixed OS/2 / EMX port (thanks to Ari Huttunen).
908: - Fixed AmigaDOS port. (thanks to Michel Schinz).
909: - Fixed the DATASTART definition under Solaris. There
910: was a 1 in 16K chance of the collector missing the first
911: 64K of static data (and thus crashing).
912: - Fixed some blatant anachronisms in the README file.
913: - Fixed PCR-Makefile for upcoming PPCR release.
914:
915: Since version 4.2:
916: - Fixed SPARC alignment problem with GC_DEBUG.
917: - Fixed Solaris threads /proc workaround. The real
918: problem was an interaction with mprotect.
919: - Incorporated fix from Patrick Beard for gc_c++.h (now gc_cpp.h).
920: - Slightly improved allocator space utilization by
921: fixing the GC_size_map mechanism.
922: - Integrated some Sony News and MIPS RISCos 4.51
923: patches. (Thanks to Nobuyuki Hikichi of
924: Software Research Associates, Inc. Japan)
925: - Fixed HP_PA alignment problem. (Thanks to
926: xjam@cork.cs.berkeley.edu.)
927: - Added GC_same_obj and friends. Changed GC_base
928: to return 0 for pointers past the end of large objects.
929: Improved GC_base performance with ALL_INTERIOR_POINTERS
930: on machines with a slow integer mod operation.
931: Added GC_PTR_ADD, GC_PTR_STORE, etc. to prepare
932: for preprocessor.
933: - changed the default on most UNIX machines to be that
934: signals are not disabled during critical GC operations.
935: This is still ANSI-conforming, though somewhat dangerous
936: in the presence of signal handlers. But the performance
937: cost of the alternative is sometimes problematic.
938: Can be changed back with a minor Makefile edit.
939: - renamed IS_STRING in gc.h, to CORD_IS_STRING, thus
940: following my own naming convention. Added the function
941: CORD_to_const_char_star.
942: - Fixed a gross bug in GC_finalize. Symptom: occasional
943: address faults in that function. (Thanks to Anselm
944: Baird-Smith (Anselm.BairdSmith@inria.fr)
945: - Added port to ICL DRS6000 running DRS/NX. Restructured
946: things a bit to factor out common code, and remove obsolete
947: code. Collector should now run under SUNOS5 with either
948: mprotect or /proc dirty bits. (Thanks to Douglas Steel
949: (doug@wg.icl.co.uk)).
950: - More bug fixes and workarounds for Solaris 2.X. (These were
951: mostly related to putting the collector in a dynamic library,
952: which didn't really work before. Also SOLARIS_THREADS
953: didn't interact well with dl_open.) Thanks to btlewis@eng.sun.com.
954: - Fixed a serious performance bug on the DEC Alpha. The text
955: segment was getting registered as part of the root set.
956: (Amazingly, the result was still fast enough that the bug
957: was not conspicuous.) The fix works on OSF/1, version 1.3.
958: Hopefully it also works on other versions of OSF/1 ...
959: - Fixed a bug in GC_clear_roots.
960: - Fixed a bug in GC_generic_malloc_words_small that broke
961: gc_inl.h. (Reported by Antoine de Maricourt. I broke it
962: in trying to tweak the Mac port.)
963: - Fixed some problems with cord/de under Linux.
964: - Fixed some cord problems, notably with CORD_riter4.
965: - Added DG/UX port.
966: Thanks to Ben A. Mesander (ben@piglet.cr.usgs.gov)
967: - Added finalization registration routines with weaker ordering
968: constraints. (This is necessary for C++ finalization with
969: multiple inheritance, since the compiler often adds self-cycles.)
970: - Filled the holes in the SCO port. (Thanks to Michael Arnoldus
971: <chime@proinf.dk>.)
972: - John Ellis' additions to the C++ support: From John:
973:
974: * I completely rewrote the documentation in the interface gc_c++.h
975: (later renamed gc_cpp.h). I've tried to make it both clearer and more
976: precise.
977:
978: * The definition of accessibility now ignores pointers from an
979: finalizable object (an object with a clean-up function) to itself.
980: This allows objects with virtual base classes to be finalizable by the
981: collector. Compilers typically implement virtual base classes using
982: pointers from an object to itself, which under the old definition of
983: accessibility prevented objects with virtual base classes from ever
984: being collected or finalized.
985:
986: * gc_cleanup now includes gc as a virtual base. This was enabled by
987: the change in the definition of accessibility.
988:
989: * I added support for operator new[]. Since most (all?) compilers
990: don't yet support operator new[], it is conditionalized on
991: -DOPERATOR_NEW_ARRAY. The code is untested, but its trivial and looks
992: correct.
993:
994: * The test program test_gc_c++ (later renamed test_cpp.cc)
995: tries to test for the C++-specific functionality not tested by the
996: other programs.
997: - Added <unistd.h> include to misc.c. (Needed for ppcr.)
998: - Added PowerMac port. (Thanks to Patrick Beard again.)
999: - Fixed "srcdir"-related Makefile problems. Changed things so
1000: that all externally visible include files always appear in the
1001: include subdirectory of the source. Made gc.h directly
1002: includable from C++ code. (These were at Per
1003: Bothner's suggestion.)
1004: - Changed Intel code to also mark from ebp (Kevin Warne's
1005: suggestion).
1006: - Renamed C++ related files so they could live in a FAT
1007: file system. (Charles Fiterman's suggestion.)
1008: - Changed Windows NT Makefile to include C++ support in
1009: gc.lib. Added C++ test as Makefile target.
1010:
1011: Since version 4.3:
1012: - ASM_CLEAR_CODE was erroneously defined for HP
1013: PA machines, resulting in a compile error.
1014: - Fixed OS/2 Makefile to create a library. (Thanks to
1015: Mark Boulter (mboulter@vnet.ibm.com)).
1016: - Gc_cleanup objects didn't work if they were created on
1017: the stack. Fixed.
1018: - One copy of Gc_cpp.h in the distribution was out of
1019: synch, and failed to document some known compiler
1020: problems with explicit destructor invocation. Partially
1021: fixed. There are probably other compilers on which
1022: gc_cleanup is miscompiled.
1023: - Fixed Makefile to pass C compiler flags to C++ compiler.
1024: - Added Mac fixes.
1025: - Fixed os_dep.c to work around what appears to be
1026: a new and different VirtualQuery bug under newer
1027: versions of win32S.
1028: - GC_non_gc_bytes was not correctly maintained by
1029: GC_free. Fixed. Thanks to James Clark (jjc@jclark.com).
1030: - Added GC_set_max_heap_size.
1031: - Changed allocation code to ignore blacklisting if it is preventing
1032: use of a very large block of memory. This has the advantage
1033: that naive code allocating very large objects is much more
1034: likely to work. The downside is you might no
1035: longer find out that such code should really use
1036: GC_malloc_ignore_off_page.
1037: - Changed GC_printf under win32 to close and reopen the file
1038: between calls. FAT file systems otherwise make the log file
1039: useless for debugging.
1040: - Added GC_try_to_collect and GC_get_bytes_since_gc. These
1041: allow starting an abortable collection during idle times.
1042: This facility does not require special OS support. (Thanks to
1043: Michael Spertus of Geodesic Systems for suggesting this. It was
1044: actually an easy addition. Kumar Srikantan previously added a similar
1045: facility to a now ancient version of the collector. At the time
1046: this was much harder, and the result was less convincing.)
1047: - Added some support for the Borland development environment. (Thanks
1048: to John Ellis and Michael Spertus.)
1049: - Removed a misfeature from checksums.c that caused unexpected
1050: heap growth. (Thanks to Scott Schwartz.)
1051: - Changed finalize.c to call WARN if it encounters a finalization cycle.
1052: WARN is defined in gc_priv.h to write a message, usually to stdout.
1053: In many environments, this may be inappropriate.
1054: - Renamed NO_PARAMS in gc.h to GC_NO_PARAMS, thus adhering to my own
1055: naming convention.
1056: - Added GC_set_warn_proc to intercept warnings.
1057: - Fixed Amiga port. (Thanks to Michel Schinz (schinz@alphanet.ch).)
1058: - Fixed a bug in mark.c that could result in an access to unmapped
1059: memory from GC_mark_from_mark_stack on machines with unaligned
1060: pointers.
1061: - Fixed a win32 specific performance bug that could result in scanning of
1062: objects allocated with the system malloc.
1063: - Added REDIRECT_MALLOC.
1064:
1065: Since version 4.4:
1066: - Fixed many minor and one major README bugs. (Thanks to Franklin Chen
1067: (chen@adi.com) for pointing out many of them.)
1068: - Fixed ALPHA/OSF/1 dynamic library support. (Thanks to Jonathan Bachrach
1069: (jonathan@harlequin.com)).
1070: - Added incremental GC support (MPROTECT_VDB) for Linux (with some
1071: help from Bruno Haible).
1072: - Altered SPARC recognition tests in gc.h and config.h (mostly as
1073: suggested by Fergus Henderson).
1074: - Added basic incremental GC support for win32, as implemented by
1075: Windows NT and Windows 95. GC_enable_incremental is a noop
1076: under win32s, which doesn't implement enough of the VM interface.
1077: - Added -DLARGE_CONFIG.
1078: - Fixed GC_..._ignore_off_page to also function without
1079: -DALL_INTERIOR_POINTERS.
1080: - (Hopefully) fixed RS/6000 port. (Only the test was broken.)
1081: - Fixed a performance bug in the nonincremental collector running
1082: on machines supporting incremental collection with MPROTECT_VDB
1083: (e.g. SunOS 4, DEC AXP). This turned into a correctness bug under
1084: win32s with win32 incremental collection. (Not all memory protection
1085: was disabled.)
1086: - Fixed some ppcr related bit rot.
1087: - Caused dynamic libraries to be unregistered before reregistering.
1088: The old way turned out to be a performance bug on some machines.
1089: - GC_root_size was not properly maintained under MSWIN32.
1090: - Added -DNO_DEBUGGING and GC_dump.
1091: - Fixed a couple of bugs arising with SOLARIS_THREADS +
1092: REDIRECT_MALLOC.
1093: - Added NetBSD/M68K port. (Thanks to Peter Seebach
1094: <seebs@taniemarie.solon.com>.)
1095: - Fixed a serious realloc bug. For certain object sizes, the collector
1096: wouldn't scan the expanded part of the object. (Thanks to Clay Spence
1097: (cds@peanut.sarnoff.com) for noticing the problem, and helping me to
1098: track it down.)
1099:
1100: Since version 4.5:
1101: - Added Linux ELF support. (Thanks to Arrigo Triulzi <arrigo@ic.ac.uk>.)
1102: - GC_base crashed if it was called before any other GC_ routines.
1103: This could happen if a gc_cleanup object was allocated outside the heap
1104: before any heap allocation.
1105: - The heap expansion heuristic was not stable if all objects had finalization
1106: enabled. Fixed finalize.c to count memory in finalization queue and
1107: avoid explicit deallocation. Changed alloc.c to also consider this count.
1108: (This is still not recommended. It's expensive if nothing else.) Thanks
1109: to John Ellis for pointing this out.
1110: - GC_malloc_uncollectable(0) was broken. Thanks to Phong Vo for pointing
1111: this out.
1112: - The collector didn't compile under Linux 1.3.X. (Thanks to Fred Gilham for
1113: pointing this out.) The current workaround is ugly, but expected to be
1114: temporary.
1115: - Fixed a formatting problem for SPARC stack traces.
1116: - Fixed some '=='s in os_dep.c that should have been assignments.
1117: Fortunately these were in code that should never be executed anyway.
1118: (Thanks to Fergus Henderson.)
1119: - Fixed the heap block allocator to only drop blacklisted blocks in small
1120: chunks. Made BL_LIMIT self adjusting. (Both of these were in response
1121: to heap growth observed by Paul Graham.)
1122: - Fixed the Metrowerks/68K Mac code to also mark from a6. (Thanks
1123: to Patrick Beard.)
1124: - Significantly updated README.debugging.
1125: - Fixed some problems with longjmps out of signal handlers, especially under
1126: Solaris. Added a workaround for the fact that siglongjmp doesn't appear to
1127: do the right thing with -lthread under Solaris.
1128: - Added MSDOS/djgpp port. (Thanks to Mitch Harris (maharri@uiuc.edu).)
1129: - Added "make reserved_namespace" and "make user_namespace". The
1130: first renames ALL "GC_xxx" identifiers as "_GC_xxx". The second is the
1131: inverse transformation. Note that doing this is guaranteed to break all
1132: clients written for the other names.
1133: - descriptor field for kind NORMAL in GC_obj_kinds with ADD_BYTE_AT_END
1134: defined should be -ALIGNMENT not WORDS_TO_BYTES(-1). This is
1135: a serious bug on machines with pointer alignment of less than a word.
1136: - GC_ignore_self_finalize_mark_proc didn't handle pointers to very near the
1137: end of the object correctly. Caused failures of the C++ test on a DEC Alpha
1138: with g++.
1139: - gc_inl.h still had problems. Partially fixed. Added warnings at the
1140: beginning to hopefully specify the remaining dangers.
1141: - Added DATAEND definition to config.h.
1142: - Fixed some of the .h file organization. Fixed "make floppy".
1143:
1144: Since version 4.6:
1145: - Fixed some compilation problems with -DCHECKSUMS (thanks to Ian Searle)
1146: - Updated some Mac specific files to synchronize with Patrick Beard.
1147: - Fixed a serious bug for machines with non-word-aligned pointers.
1148: (Thanks to Patrick Beard for pointing out the problem. The collector
1149: should fail almost any conceivable test immediately on such machines.)
1150:
1151: Since version 4.7:
1152: - Changed a "comment" in a MacOS specific part of mach-dep.c that caused
1153: gcc to fail on other platforms.
1154:
1155: Since version 4.8
1156: - More README.debugging fixes.
1157: - Objects ready for finalization, but not finalized in the same GC
1158: cycle, could be prematurely collected. This occasionally happened
1159: in test_cpp.
1160: - Too little memory was obtained from the system for very large
1161: objects. That could cause a heap explosion if these objects were
1162: not contiguous (e.g. under PCR), and too much of them was blacklisted.
1163: - Due to an improper initialization, the collector was too hesitant to
1164: allocate blacklisted objects immediately after system startup.
1165: - Moved GC_arrays from the data into the bss segment by not explicitly
1166: initializing it to zero. This significantly
1167: reduces the size of executables, and probably avoids some disk accesses
1168: on program startup. It's conceivable that it might break a port that I
1169: didn't test.
1170: - Fixed EMX_MAKEFILE to reflect the gc_c++.h to gc_cpp.h renaming which
1171: occurred a while ago.
1172:
1173: Since 4.9:
1174: - Fixed a typo around a call to GC_collect_or_expand in alloc.c. It broke
1175: handling of out of memory. (Thanks to Patrick Beard for noticing.)
1176:
1177: Since 4.10:
1178: - Rationalized (hopefully) GC_try_to_collect in an incremental collection
1179: environment. It appeared to not handle a call while a collection was in
1180: progress, and was otherwise too conservative.
1181: - Merged GC_reclaim_or_delete_all into GC_reclaim_all to get rid of some
1182: code.
1183: - Added Patrick Beard's Mac fixes, with substantial completely untested
1184: modifications.
1185: - Fixed the MPROTECT_VDB code to deal with large pages and imprecise
1186: fault addresses (as on an UltraSPARC running Solaris 2.5). Note that this
1187: was not a problem in the default configuration, which uses PROC_VDB.
1188: - The DEC Alpha assembly code needed to restore $gp between calls.
1189: Thanks to Fergus Henderson for tracking this down and supplying a
1190: patch.
1191: - The write command for "de" was completely broken for large files.
1192: I used the easiest portable fix, which involved changing the semantics
1193: so that f.new is written instead of overwriting f. That's safer anyway.
1194: - Added README.solaris2 with a discussion of the possible problems of
1195: mixing the collector's sbrk allocation with malloc/realloc.
1196: - Changed the data segment starting address for SGI machines. The
1197: old code failed under IRIX6.
1198: - Required double word alignment for MIPS.
1199: - Various minor fixes to remove warnings.
1200: - Attempted to fix some Solaris threads problems reported by Zhiying Chen.
1201: In particular, the collector could try to fork a thread with the
1202: world stopped as part of GC_thr_init. It also failed to deal with
1203: the case in which the original thread terminated before the whole
1204: process did.
1205: - Added -DNO_EXECUTE_PERMISSION. This has a major performance impact
1206: on the incremental collector under Irix, and perhaps under other
1207: operating systems.
1208: - Added some code to support allocating the heap with mmap. This may
1209: be preferable under some circumstances.
1210: - Integrated dynamic library support for HP.
1211: (Thanks to Knut Tvedten <knuttv@ifi.uio.no>.)
1212: - Integrated James Clark's win32 threads support, and made a number
1213: of changes to it, many of which were suggested by Pontus Rydin.
1214: This is still not 100% solid.
1215: - Integrated Alistair Crooks' support for UTS4 running on an Amdahl
1216: 370-class machine.
1217: - Fixed a serious bug in explicitly typed allocation. Objects requiring
1218: large descriptors where handled in a way that usually resulted in
1219: a segmentation fault in the marker. (Thanks to Jeremy Fitzhardinge
1220: for helping to track this down.)
1221: - Added partial support for GNU win32 development. (Thanks to Fergus
1222: Henderson.)
1223: - Added optional support for Java-style finalization semantics. (Thanks
1224: to Patrick Bridges.) This is recommended only for Java implementations.
1225: - GC_malloc_uncollectable faulted instead of returning 0 when out of
1226: memory. (Thanks to dan@math.uiuc.edu for noticing.)
1227: - Calls to GC_base before the collector was initialized failed on a
1228: DEC Alpha. (Thanks to Matthew Flatt.)
1229: - Added base pointer checking to GC_REGISTER_FINALIZER in debugging
1230: mode, at the suggestion of Jeremy Fitzhardinge.
1231: - GC_debug_realloc failed for uncollectable objects. (Thanks to
1232: Jeremy Fitzhardinge.)
1233: - Explicitly typed allocation could crash if it ran out of memory.
1234: (Thanks to Jeremy Fitzhardinge.)
1235: - Added minimal support for a DEC Alpha running Linux.
1236: - Fixed a problem with allocation of objects whose size overflowed
1237: ptrdiff_t. (This now fails unconditionally, as it should.)
1238: - Added the beginning of Irix pthread support.
1239: - Integrated Xiaokun Zhu's fixes for djgpp 2.01.
1240: - Added SGI-style STL allocator support (gc_alloc.h).
1241: - Fixed a serious bug in README.solaris2. Multithreaded programs must include
1242: gc.h with SOLARIS_THREADS defined.
1243: - Changed GC_free so it actually deallocates uncollectable objects.
1244: (Thanks to Peter Chubb for pointing out the problem.)
1245: - Added Linux ELF support for dynamic libararies. (Thanks again to
1246: Patrick Bridges.)
1247: - Changed the Borland cc configuration so that the assembler is not
1248: required.
1249: - Fixed a bug in the C++ test that caused it to fail in 64-bit
1250: environments.
1251:
1252: Since 4.11:
1253: - Fixed ElfW definition in dyn_load.c. (Thanks to Fergus Henderson.)
1254: This prevented the dynamic library support from compiling on some
1255: older ELF Linux systems.
1256: - Fixed UTS4 port (which I apparently mangled during the integration)
1257: (Thanks to again to Alistair Crooks.)
1258: - "Make C++" failed on Suns with SC4.0, due to a problem with "bool".
1259: Fixed in gc_priv.h.
1260: - Added more pieces for GNU win32. (Thanks to Timothy N. Newsham.)
1261: The current state of things should suffice for at least some
1262: applications.
1263: - Changed the out of memory retry count handling as suggested by
1264: Kenjiro Taura. (This matters only if GC_max_retries > 0, which
1265: is no longer the default.)
1266: - If a /proc read failed repeatedly, GC_written_pages was not updated
1267: correctly. (Thanks to Peter Chubb for diagnosing this.)
1268: - Under unlikely circumstances, the allocator could infinite loop in
1269: an out of memory situation. (Thanks again to Kenjiro Taura for
1270: identifying the problem and supplying a fix.)
1271: - Fixed a syntactic error in the DJGPP code. (Thanks to Fergus
1272: Henderson for finding this by inspection.) Also fixed a test program
1273: problem with DJGPP (Thanks to Peter Monks.)
1274: - Atomic uncollectable objects were not treated correctly by the
1275: incremental collector. This resulted in weird log statistics and
1276: occasional performance problems. (Thanks to Peter Chubb for pointing
1277: this out.)
1278: - Fixed some problems resulting from compilers that dont define
1279: __STDC__. In this case void * and char * were used inconsistently
1280: in some cases. (Void * should not have been used at all. If
1281: you have an ANSI superset compiler that does not define __STDC__,
1282: please compile with -D__STDC__=0. Thanks to Manuel Serrano and others
1283: for pointing out the problem.)
1284: - Fixed a compilation problem on Irix with -n32 and -DIRIX_THREADS.
1285: Also fixed some other IRIX_THREADS problems which may or may not have
1286: had observable symptoms.
1287: - Fixed an HP PA compilation problem in dyn_load.c. (Thanks to
1288: Philippe Queinnec.)
1289: - SEGV fault handlers sometimes did not get reset correctly. (Thanks
1290: to David Pickens.)
1291: - Added a fix for SOLARIS_THREADS on Intel. (Thanks again to David
1292: Pickens.) This probably needs more work to become functional.
1293: - Fixed struct sigcontext_struct in os_dep.c for compilation under
1294: Linux 2.1.X. (Thanks to Fergus Henderson.)
1295: - Changed the DJGPP STACKBOTTOM and DATASTART values to those suggested
1296: by Kristian Kristensen. These may still not be right, but it is
1297: it is likely to work more often than what was there before. They may
1298: even be exactly right.
1299: - Added a #include <string.h> to test_cpp.cc. This appears to help
1300: with HP/UX and gcc. (Thanks to assar@sics.se.)
1301: - Version 4.11 failed to run in incremental mode on recent 64-bit Irix
1302: kernels. This was a problem related to page unaligned heap segments.
1303: Changed the code to page align heap sections on all platforms.
1304: (I had mistakenly identified this as a kernel problem earlier.
1305: It was not.)
1306: - Version 4.11 did not make allocated storage executable, except on
1307: one or two platforms, due to a bug in a #if test. (Thanks to Dave
1308: Grove for pointing this out.)
1309: - Added sparc_sunos4_mach_dep.s to support Sun's compilers under SunOS4.
1310: - Added GC_exclude_static_roots.
1311: - Fixed the object size mapping algorithm. This shouldn't matter,
1312: but the old code was ugly.
1313: - Heap checking code could die if one of the allocated objects was
1314: larger than its base address. (Unsigned underflow problem. Thanks
1315: to Clay Spence for isolating the problem.)
1316: - Added RS6000 (AIX) dynamic library support and fixed STACK_BOTTOM.
1317: (Thanks to Fred Stearns.)
1318: - Added Fergus Henderson's patches for improved robustness with large
1319: heaps and lots of blacklisting.
1320: - Added Peter Chubb's changes to support Solaris Pthreads, to support
1321: MMAP allocation in Solaris, to allow Solaris to find dynamic libraries
1322: through /proc, to add malloc_typed_ignore_off_page, and a few other
1323: minor features and bug fixes.
1324: - The Solaris 2 port should not use sbrk. I received confirmation from
1325: Sun that the use of sbrk and malloc in the same program is not
1326: supported. The collector now defines USE_MMAP by default on Solaris.
1327: - Replaced the djgpp makefile with Gary Leavens' version.
1328: - Fixed MSWIN32 detection test.
1329: - Added Fergus Henderson's patches to allow putting the collector into
1330: a DLL under GNU win32.
1331: - Added Ivan V. Demakov's port to Watcom C on X86.
1332: - Added Ian Piumarta's Linux/PowerPC port.
1333: - On Brian Burton's suggestion added PointerFreeGC to the placement
1334: options in gc_cpp.h. This is of course unsafe, and may be controversial.
1335: On the other hand, it seems to be needed often enough that it's worth
1336: adding as a standard facility.
1337:
1338: Since 4.12:
1339: - Fixed a crucial bug in the Watcom port. There was a redundant decl
1340: of GC_push_one in gc_priv.h.
1341: - Added FINALIZE_ON_DEMAND.
1342: - Fixed some pre-ANSI cc problems in test.c.
1343: - Removed getpagesize() use for Solaris. It seems to be missing in one
1344: or two versions.
1345: - Fixed bool handling for SPARCCompiler version 4.2.
1346: - Fixed some files in include that had gotten unlinked from the main
1347: copy.
1348: - Some RS/6000 fixes (missing casts). Thanks to Toralf Foerster.
1349: - Fixed several problems in GC_debug_realloc, affecting mostly the
1350: FIND_LEAK case.
1351: - GC_exclude_static_roots contained a buggy unsigned comparison to
1352: terminate a loop. (Thanks to Wilson Ho.)
1353: - CORD_str failed if the substring occurred at the last possible position.
1354: (Only affects cord users.)
1355: - Fixed Linux code to deal with RedHat 5.0 and integrated Peter Bigot's
1356: os_dep.c code for dealing with various Linux versions.
1357: - Added workaround for Irix pthreads sigaction bug and possible signal
1358: misdirection problems.
1359: Since alpha1:
1360: - Changed RS6000 STACKBOTTOM.
1361: - Integrated Patrick Beard's Mac changes.
1362: - Alpha1 didn't compile on Irix m.n, m < 6.
1363: - Replaced Makefile.dj with a new one from Gary Leavens.
1364: - Added Andrew Stitcher's changes to support SCO OpenServer.
1365: - Added PRINT_BLACK_LIST, to allow debugging of high densities of false
1366: pointers.
1367: - Added code to debug allocator to keep track of return address
1368: in GC_malloc caller, thus giving a bit more context.
1369: - Changed default behavior of large block allocator to more
1370: aggressively avoid fragmentation. This is likely to slow down the
1371: collector when it succeeds at reducing space cost.
1372: - Integrated Fergus Henderson's CYGWIN32 changes. They are untested,
1373: but needed for newer versions.
1374: - USE_MMAP had some serious bugs. This caused the collector to fail
1375: consistently on Solaris with -DSMALL_CONFIG.
1376: - Added Linux threads support, thanks largely to Fergus Henderson.
1377: Since alpha2:
1378: - Fixed more Linux threads problems.
1379: - Changed default GC_free_space_divisor to 3 with new large block allocation.
1380: (Thanks to Matthew Flatt for some measurements that suggest the old
1381: value sometimes favors space too much over time.)
1382: - More CYGWIN32 fixes.
1383: - Integrated Tyson-Dowd's Linux-M68K port.
1384: - Minor HP PA and DEC UNIX fixes from Fergus Henderson.
1385: - Integrated Christoffe Raffali's Linux-SPARC changes.
1386: - Allowed for one more GC fixup iteration after a full GC in incremental
1387: mode. Some quick measurements suggested that this significantly
1388: reduces pause times even with smaller GC_RATE values.
1389: - Moved some more GC data structures into GC_arrays. This decreases
1390: pause times and GC overhead, but makes debugging slightly less convenient.
1391: - Fixed namespace pollution problem ("excl_table").
1392: - Made GC_incremental a constant for -DSMALL_CONFIG, hopefully shrinking
1393: that slightly.
1394: - Added some win32 threads fixes.
1395: - Integrated Ivan Demakov and David Stes' Watcom fixes.
1396: - Various other minor fixes contributed by many people.
1397: - Renamed config.h to gcconfig.h, since config.h tends to be used for
1398: many other things.
1399: - Integrated Matthew Flatt's support for 68K MacOS "far globals".
1400: - Fixed up some of the dynamic library Makefile targets for consistency
1401: across platforms.
1402: - Fixed a USE_MMAP typo that caused out-of-memory handling to fail
1403: on Solaris.
1404: - Added code to test.c to test thread creation a bit more.
1405: - Integrated GC_win32_free_heap, as suggested by Ivan Demakov.
1406: - Fixed Solaris 2.7 stack base finding problem. (This may actually
1407: have been done in an earlier alpha release.)
1408: Since alpha3:
1409: - Fixed MSWIN32 recognition test, which interfered with cygwin.
1410: - Removed unnecessary gc_watcom.asm from distribution. Removed
1411: some obsolete README.win32 text.
1412: - Added Alpha Linux incremental GC support. (Thanks to Philipp Tomsich
1413: for code for retrieving the fault address in a signal handler.)
1414: Changed Linux signal handler context argument to be a pointer.
1415: - Took care of some new warnings generated by the 7.3 SGI compiler.
1416: - Integrated Phillip Musumeci's FreeBSD/ELF fixes.
1417: - -DIRIX_THREADS was broken with the -o32 ABI (typo in gc_priv.h>
1418:
1419: Since 4.13:
1420: - Fixed GC_print_source_ptr to not use a prototype.
1421: - generalized CYGWIN test.
1422: - gc::new did the wrong thing with PointerFreeGC placement.
1423: (Thanks to Rauli Ruohonen.)
1424: - In the ALL_INTERIOR_POINTERS (default) case, some callee-save register
1425: values could fail to be scanned if the register was saved and
1426: reused in a GC frame. This showed up in verbose mode with gctest
1427: compiled with an unreleased SGI compiler. I vaguely recall an old
1428: bug report that may have been related. The bug was probably quite old.
1429: (The problem was that the stack scanning could be deferred until
1430: after the relevant frame was overwritten, and the new save location
1431: might be outside the scanned area. Fixed by more eager stack scanning.)
1432: - PRINT_BLACK_LIST had some problems. A few source addresses were garbage.
1433: - Replaced Makefile.dj and added -I flags to cord make targets.
1434: (Thanks to Gary Leavens.)
1435: - GC_try_to_collect was broken with the nonincremental collector.
1436: - gc_cleanup destructors could pass the wrong address to
1437: GC_register_finalizer_ignore_self in the presence of multiple
1438: inheritance. (Thanks to Darrell Schiebel.)
1439: - Changed PowerPC Linux stack finding code.
1440:
1441: Since 4.14alpha1
1442: - -DSMALL_CONFIG did not work reliably with large (> 4K) pages.
1443: Recycling the mark stack during expansion could result in a size
1444: zero heap segment, which confused things. (This was probably also an
1445: issue with the normal config and huge pages.)
1446: - Did more work to make sure that callee-save registers were scanned
1447: completely, even with the setjmp-based code. Added USE_GENERIC_PUSH_REGS
1448: macro to facilitate testing on machines I have access to.
1449: - Added code to explicitly push register contents for win32 threads.
1450: This seems to be necessary. (Thanks to Pierre de Rop.)
1451:
1452: Since 4.14alpha2
1453: - changed STACKBOTTOM for DJGPP (Thanks to Salvador Eduardo Tropea).
1454:
1455: To do:
1456: - Very large root set sizes (> 16 MB or so) could cause the collector
1457: to abort with an unexpected mark stack overflow. (Thanks again to
1458: Peter Chubb.) NOT YET FIXED. Workaround is to increase the initial
1459: size.
1460: - The SGI version of the collector marks from mmapped pages, even
1461: if they are not part of dynamic library static data areas. This
1462: causes performance problems with some SGI libraries that use mmap
1463: as a bitmap allocator. NOT YET FIXED. It may be possible to turn
1464: off DYNAMIC_LOADING in the collector as a workaround. It may also
1465: be possible to conditionally intercept mmap and use GC_exclude_static_roots.
1466: The real fix is to walk rld data structures, which looks possible.
1467: - Integrate MIT and DEC pthreads ports.
1468: - Deal with very uneven black-listing distributions. If all the black listed
1469: blocks reside in the newly allocated heap section, the heuristic for
1470: temporarily ignoring black-listing fails, and the heap grows too much.
1471: (This was observed in only one case, and could be worked around, but ...)
1472: - I've started work on rewriting the large block allocator to use approximate
1473: best fit. There are rare cases in which the current allocator results in
1474: excessive large block fragmentation, even with the 4.13 fixes. This should
1475: also reduce large block allocation time, whcih has become occasionally
1476: noticable in 4.13.
1477:
FreeBSD-CVSweb <freebsd-cvsweb@FreeBSD.org>
![[BACK]](/icons/cvsweb/back.gif){kind=icon}
Return to README CVS log ![[TXT]](/icons/cvsweb/text.gif){kind=icon}
![[DIR]](/icons/cvsweb/dir.gif){kind=icon}
Up to [local] / OpenXM_contrib / gc