Mercurial > hg > xemacs-beta
view src/ntheap.c @ 502:7039e6323819
[xemacs-hg @ 2001-05-04 22:41:46 by ben]
----------------------- byte-comp warning fixes -----------------
New functions for cleanly eliminating byte-compiler warnings.
Their definitions require no changes at all in bytecomp.el,
meaning that any package that wants to use them and be compatible
with older versions of XEmacs need only copy the code and rename
the functions (i.e. prefix them with the package name).
Eliminate byte-compiler warnings using the new functions in
bytecomp-runtime.el.
Move coding-system-put,get,category, since they're not
Mule-specific and are used in prefer-coding-system.
font.el was incredibly ugly. Clean it up. Avoid using defsubst
for any exported functions, to avoid possible compatibility
problems if we later change the internal interface. (It happened
before, with face accessors, between 19.8 and 19.9). Fix tons
of warnings.
Clean up (new function gpm-is-supported-p eliminates duplicate
code in gpm-create/delete-device-hook) and eliminate warnings.
---------- make byte-recompile-directory work in the ---------
core `lisp' dir, even in the absence of
a Mule XEmacs (i.e. make it skip the Mule
files rather than trying to compile them).
now you should be able to do `touch *.el'
in the `lisp' dir, then
M-x byte-recompile-directory, and get no
warnings.
Avoid trying to compile Mule files in byte-recompile-directory
when we're not in a Mule XEmacs, since we're highly likely to get
syntax errors.
Add a coding-system cookie to all Mule files so that
byte-recompile-directory ignores them.
Magic cookie function moved to files.el from code-files.el (for
use by bytecomp even in a non-coding-system XEmacs), and changed
names and semantics for use by bytecomp. NOTE: IMO this is an
internal function that we can change as we like (and there is
absolutely no code anywhere else using the function).
---------------- GUI improvements: menus, help -------------------
Rearrange order of keymap declarations to be alphabetical.
Improve help on help to include all bindings, and group by
category. Add bindings for new Info commands. Remove
warnings. Use command-hyper-apropos in place of command-apropos.
Add a function to do the equivalent of command-apropos.
Evals its help-text argument so you can put expressions there.
Used now by help-for-help.
Add binding to continue text searches. Expand index searches to
work over multiple info documents. Add commands to search
text/index in User and Lispref.
Add new entry, "Uncomment Region" (parallels "Comment Out Region").
Redo Help menu; add bindings for new Info commands to search the
index or text of the User and Lispref manuals. Add command for
mark-paragraph, activate-region. Make Edit->R accelerator be
rectangle, not register (more commonly used), and put rectangle
first. Fix the Edit Init File entry to never load the .elc file.
Simplify the default-popup-menu. Add Cmds->Tabs menu.
Use kp-left not kp_left, etc.
---------------- Miscellaneous bug fixes/cleanup -------------------
byte-compiler-options: Correct doc string.
easy-menu-do-define: fix extra quote.
fill-paragraph-or-region:Rewrite to be more correct -- use
call-interactively so that we always get exactly the same
behavior as if the functions were called directly.
No need to fiddle with zmacs-region-stays, now that bogus
clearing of it (2001-04-28 src/ChangeLog) is removed.
Put dialog titles back in -- this time correctly. Fix various
other problems with leaks and such.
key-sequence-list-description:
Clean up fun to always correctly canonicalize.
Clean up Kinsoku comments, synch comment-region with FSF 20.7.
* simple.el (region-exists-p):
* simple.el (region-active-p):
Add comment about which one is correct to use in menu specs.
* sound.el (load-sound-file):
Minor code clean up.
* startup.el:
* startup.el (command-line-early):
* startup.el (initial-scratch-message):
Comment changes. Add info about sample.init.el to splash screen.
Improve initial-scratch-message and clarify purpose of Scratch
buffer. Fix byte-compile warning.
------------------------ Added features -------------------------
Add new variable to control whether etags checks all parent
directories for tag files. (On by default.)
* hash-table.el: New file, useful utility functions.
* dumped-lisp.el (preloaded-file-list): Dump hash-table.el.
------------ notable bug fix: Windows event code --------------
Get critical quit working.
------------ notable bug fix and new feature: regex code --------------
Shy groups were implemented in a horrible, half-assed way that
would cause them to screw up regex searching in most cases.
Fixed to work correctly.
Also extended back-reference syntax past 9. Only is recognized
as such if there are at least that many non-shy groups; and
optionally will warn about such uses, to catch old code that
might be using them differently. (Added variable to control
this in search.c -- `warn-about-possibly-incompatible-back-
references', on by default for the moment. Declared in lisp.h.
---------------- process/SIGIO improvements -------------------
define USE_GETADDRINFO to replace more complex conditional,
and use it. the code conditionalized on this in
unix_open_network_stream had *serious* problems handling errors.
it's now fixed, and major amounts of duplicate code between
the two versions were combined.
don't disable SIGIO and other interrupts unless
CONNECT_NEEDS_SLOWED_INTERRUPTS is defined -- don't penalize OS's
without bugs. similarly for a freebsd bug that was affecting all
OS's.
* s\ultrix.h:
define CONNECT_NEEDS_SLOWED_INTERRUPTS, since that's the OS
mentioned as having a kernel bug.
* sysdep.c (request_sigio_on_device):
* sysdep.c (unrequest_sigio_on_device):
fix SIGIO problems on Linux. add check for O_ASYNC in case it's
defined and FASYNC isn't. add comment about other ways to do
SIGIO on Linux.
* callproc.c (Fold_call_process_internal):
* process.c (Fstart_process_internal):
Deal with the possibility that `default-directory' doesn't
have terminating slash. Correct comments about vfork.
---------------- Miscellaneous bug fixes/cleanup -------------------
* callint.c (Finteractive):
Add lots of documentation -- exactly what the Lisp equivalents of
all the interactive specs are.
* console.h (struct console): change type of quit_char to Emchar.
* event-msw.c (lstream_type_create_mswindows_selectable): spacing
change.
Eliminate events-mod.h and combine into events.h.
* emacs.c:
* emacs.c (make_arg_list_1):
* emacs.c (main_1):
A couple of char->Extbyte changes, add a comment.
* glyphs-msw.c:
Correct indentation of function defns to not exceed 80 cols.
Try (sort of) to fix some code that sets the colors of the
progress gauge. (Commented out)
* keymap.c (syms_of_keymap):
use DEFSYMBOL.
* process.c (read_process_output):
No need to fiddle with zmacs_region_stays, now that bogus
clearing of it (see below) is removed.
* search.c (Freplace_match): warning fix.
author | ben |
---|---|
date | Fri, 04 May 2001 22:42:35 +0000 |
parents | abe6d1db359e |
children | b39c14581166 |
line wrap: on
line source
/* Heap management routines for XEmacs on Windows NT. Copyright (C) 1994 Free Software Foundation, Inc. This file is part of XEmacs. XEmacs is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2, or (at your option) any later version. XEmacs is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License along with XEmacs; see the file COPYING. If not, write to the Free Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. Geoff Voelker (voelker@cs.washington.edu) 7-29-94 */ /* Adapted for XEmacs by David Hobley <david@spook-le0.cia.com.au> */ /* Synced with FSF Emacs 19.34.6 by Marc Paquette <marcpa@cam.org> */ #include <config.h> #include "lisp.h" /* for VALMASK */ #include <stdlib.h> #include "ntheap.h" /* This gives us the page size and the size of the allocation unit on NT. */ SYSTEM_INFO sysinfo_cache; unsigned long syspage_mask = 0; /* These are defined to get Emacs to compile, but are not used. */ int edata; int etext; /* Cache information describing the NT system for later use. */ void cache_system_info (void) { /* Cache page size, allocation unit, processor type, etc. */ GetSystemInfo (&sysinfo_cache); syspage_mask = sysinfo_cache.dwPageSize - 1; } /* Round ADDRESS up to be aligned with ALIGN. */ unsigned char * round_to_next (unsigned char *address, unsigned long align) { unsigned long tmp; tmp = (unsigned long) address; tmp = (tmp + align - 1) / align; return (unsigned char *) (tmp * align); } /* Info for keeping track of our heap. */ unsigned char *data_region_base = UNINIT_PTR; unsigned char *data_region_end = UNINIT_PTR; unsigned char *real_data_region_end = UNINIT_PTR; unsigned long data_region_size = UNINIT_LONG; unsigned long reserved_heap_size = UNINIT_LONG; /* The start of the data segment. */ unsigned char * get_data_start (void) { return data_region_base; } /* The end of the data segment. */ unsigned char * get_data_end (void) { return data_region_end; } static unsigned char * allocate_heap (void) { /* The base address for our GNU malloc heap is chosen in conjunction with the link settings for temacs.exe which control the stack size, the initial default process heap size and the executable image base address. The link settings and the malloc heap base below must all correspond; the relationship between these values depends on how NT and Win95 arrange the virtual address space for a process (and on the size of the code and data segments in temacs.exe). The most important thing is to make base address for the executable image high enough to leave enough room between it and the 4MB floor of the process address space on Win95 for the primary thread stack, the process default heap, and other assorted odds and ends (eg. environment strings, private system dll memory etc) that are allocated before temacs has a chance to grab its malloc arena. The malloc heap base can then be set several MB higher than the executable image base, leaving enough room for the code and data segments. Because some parts of Emacs can use rather a lot of stack space (for instance, the regular expression routines can potentially allocate several MB of stack space) we allow 8MB for the stack. Allowing 1MB for the default process heap, and 1MB for odds and ends, we can base the executable at 16MB and still have a generous safety margin. At the moment, the executable has about 810KB of code (for x86) and about 550KB of data - on RISC platforms the code size could be roughly double, so if we allow 4MB for the executable we will have plenty of room for expansion. Thus we would like to set the malloc heap base to 20MB. However, Win95 refuses to allocate the heap starting at this address, so we set the base to 27MB to make it happy. Since Emacs now leaves 28 bits available for pointers, this lets us use the remainder of the region below the 256MB line for our malloc arena - 229MB is still a pretty decent arena to play in! */ unsigned long base = 0x01B00000; /* 27MB */ /* Temporary hack for the non-starting problem - use 28 (256Mb) rather than VALBITS (1Gb) */ unsigned long end = 1 << 28; /* 256MB */ void *ptr = NULL; #define NTHEAP_PROBE_BASE 1 #if NTHEAP_PROBE_BASE /* This is never normally defined */ /* Try various addresses looking for one the kernel will let us have. */ while (!ptr && (base < end)) { reserved_heap_size = end - base; ptr = VirtualAlloc ((void *) base, get_reserved_heap_size (), MEM_RESERVE, PAGE_NOACCESS); base += 0x00100000; /* 1MB increment */ } #else reserved_heap_size = end - base; ptr = VirtualAlloc ((void *) base, get_reserved_heap_size (), MEM_RESERVE, PAGE_NOACCESS); #endif return (unsigned char*) ptr; } /* Emulate Unix sbrk. */ void * sbrk (unsigned long increment) { void *result; long size = (long) increment; /* Allocate our heap if we haven't done so already. */ if (data_region_base == UNINIT_PTR) { data_region_base = allocate_heap (); if (!data_region_base) return NULL; data_region_end = data_region_base; real_data_region_end = data_region_end; data_region_size = get_reserved_heap_size (); } result = data_region_end; /* If size is negative, shrink the heap by decommitting pages. */ if (size < 0) { int new_size; unsigned char *new_data_region_end; size = -size; /* Sanity checks. */ if ((data_region_end - size) < data_region_base) return NULL; /* We can only decommit full pages, so allow for partial deallocation [cga]. */ new_data_region_end = (data_region_end - size); new_data_region_end = (unsigned char *) ((long) (new_data_region_end + syspage_mask) & ~syspage_mask); new_size = real_data_region_end - new_data_region_end; real_data_region_end = new_data_region_end; if (new_size > 0) { /* Decommit size bytes from the end of the heap. */ if (!VirtualFree (real_data_region_end, new_size, MEM_DECOMMIT)) return NULL; } data_region_end -= size; } /* If size is positive, grow the heap by committing reserved pages. */ else if (size > 0) { /* Sanity checks. */ if ((data_region_end + size) > (data_region_base + get_reserved_heap_size ())) return NULL; /* Commit more of our heap. */ if (VirtualAlloc (data_region_end, size, MEM_COMMIT, PAGE_READWRITE) == NULL) return NULL; data_region_end += size; /* We really only commit full pages, so record where the real end of committed memory is [cga]. */ real_data_region_end = (unsigned char *) ((long) (data_region_end + syspage_mask) & ~syspage_mask); } return result; } #if !defined (CANNOT_DUMP) && !defined(HEAP_IN_DATA) && !defined(PDUMP) /* Recreate the heap from the data that was dumped to the executable. EXECUTABLE_PATH tells us where to find the executable. */ void recreate_heap (char *executable_path) { /* First reserve the upper part of our heap. (We reserve first because there have been problems in the past where doing the mapping first has loaded DLLs into the VA space of our heap.) */ /* Query the region at the end of the committed heap */ void *tmp; MEMORY_BASIC_INFORMATION info; DWORD size; unsigned char* base = get_heap_end (); unsigned char* end = base + get_reserved_heap_size () - get_committed_heap_size (); VirtualQuery (base, &info, sizeof info); if (info.State != MEM_FREE) { /* Oops, something has already reserved or commited it, nothing we can do but exit */ char buf[256]; wsprintf(buf, "XEmacs cannot start because the memory region required by the heap is not available.\n" "(BaseAddress = 0x%lx, AllocationBase = 0x%lx, Size = 0x%lx, State = %s, Type = %s)", info.BaseAddress, info.AllocationBase, info.RegionSize, info.State == MEM_COMMIT ? "COMMITED" : "RESERVED", info.Type == MEM_IMAGE ? "IMAGE" : info.Type == MEM_MAPPED ? "MAPPED" : "PRIVATE"); MessageBox(NULL, buf, "XEmacs", MB_OK | MB_ICONSTOP); exit(1); } /* Now try and reserve as much as possible */ size = min (info.RegionSize, end - base); tmp = VirtualAlloc (base, size, MEM_RESERVE, PAGE_NOACCESS); if (!tmp) { /* Can't reserve it, nothing we can do but exit */ char buf[256]; wsprintf(buf, "XEmacs cannot start because it couldn't reserve space required for the heap.\n" "(VirtualAlloc at 0x%lx of 0x%lx failed (%d))", base, size, GetLastError()); MessageBox(NULL, buf, "XEmacs", MB_OK | MB_ICONSTOP); exit (1); } /* We read in the data for the .bss section from the executable first and map in the heap from the executable second to prevent any funny interactions between file I/O and file mapping. */ read_in_bss (executable_path); map_in_heap (executable_path); /* Update system version information to match current system. */ cache_system_info (); } #endif /* CANNOT_DUMP */ /* Round the heap up to the given alignment. */ void round_heap (unsigned long align) { unsigned long needs_to_be; unsigned long need_to_alloc; needs_to_be = (unsigned long) round_to_next (get_heap_end (), align); need_to_alloc = needs_to_be - (unsigned long) get_heap_end (); if (need_to_alloc) sbrk (need_to_alloc); } #if (_MSC_VER >= 1000) /* MSVC 4.2 invokes these functions from mainCRTStartup to initialize a heap via HeapCreate. They are normally defined by the runtime, but we override them here so that the unnecessary HeapCreate call is not performed. */ int __cdecl _heap_init (void) { /* Stepping through the assembly indicates that mainCRTStartup is expecting a nonzero success return value. */ return 1; } void __cdecl _heap_term (void) { return; } #endif