reworked, as it interacts poorly with functions on `post-command-hook'
or `after-change-functions'.
* lisp/simple.el: New section added.
* src/cmds.c (remove_excessive_undo_boundaries): Now in lisp.
(self_insert_command): Calls simple.el to amalgamate.
(delete_char): Calls simple.el to amalgamate.
* src/keyboard.c (last_undo_boundary): Removed.
* src/undo.c (run_undoable_change): New function.
* src/data.c (INTBIG_TO_LISP): New macro, with most
of the contents of the old INTEGER_TO_CONS.
(intbig_to_lisp, uintbig_to_lisp): New functions.
* src/lisp.h (INTEGER_TO_CONS):
Simplify by using EXPR_SIGNED and the new functions.
This shrinks code size a bit, and makes it easier to
put a breakpoint on handling of large integers.
* src/lisp.h (LISP_MACRO_DEFUN, LISP_MACRO_DEFUN_VOID): Remove.
All uses rewritten to define the function directly rather than to
use a macro to define the function. This conforms to Standard C,
which does not allow stray semicolons at the top level. I hope it
also avoids the problems with TAGS. Those macros, though clever,
were pretty confusing anyway, and it wasn’t clear they were worth
the aggravation even without the TAGS problem.
* src/lisp.h (LISP_MACRO_DEFUN): Mention in the commentary the
need to end each invocation with a semi-colon.
Add a semi-colon at the end of each invocation of LISP_MACRO_DEFUN
and LISP_MACRO_DEFUN_VOID. This is to avoid missing in TAGS
inline functions defined immediately after each invocation, and
also avoid tagging every invocation of these macros.
* src/alloc.c (PURE_POINTER_P): Remove.
All uses replaced with PURE_P.
(XPNTR_OR_SYMBOL_OFFSET): New function.
(XPNTR): Move here from lisp.h.
Reimplement in terms of XPNTR_OR_SYMBOL_OFFSET.
(mark_maybe_object, valid_lisp_object_p, survives_gc_p):
Remove unnecessary cast.
(purecopy): Use XPNTR_OR_SYMBOL_OFFSET instead of XPNTR,
to avoid an unnecessary runtime test for symbols.
* src/lisp.h (lisp_h_XPNTR, XPNTR): Remove, moving XPNTR to alloc.c.
Only alloc.c needs XPNTR now.
* src/nsterm.m:
* src/lisp.h:
* src/editfns.c:
* src/doprnt.c: Add 'coding' cookies -- these files include
Unicode characters and should be decoded as UTF-8.
Make --enable-gcc-warnings a bit pickier, by also using -Wswitch.
* configure.ac (WERROR_CFLAGS): Don’t add -Wno-switch.
* lib-src/etags.c (main, consider_token, C_entries):
* src/coding.c (encode_invocation_designation):
* src/data.c (Ftype_of):
* src/eval.c (Fdefvaralias, default_toplevel_binding)
(Fbacktrace__locals, mark_specpdl):
* src/lisp.h (record_xmalloc):
* src/syntax.c (scan_lists, scan_sexps_forward):
* src/window.c (window_relative_x_coord):
* src/xdisp.c (push_it, pop_it):
* src/xterm.c (xg_scroll_callback, x_check_fullscreen):
Error out or do nothing (as appropriate) if a switch statement
with an enum value does not cover all of the enum.
* src/dispextern.h (struct iterator_stack_entry.u.comp):
Remove unused member discovered by using -Wswitch.
* src/lisp.h (record_xmalloc): Add a ‘+ 0’ to pacify -Wswitch.
* src/vm-limit.c (check_memory_limits):
Simplify warning-diagnostic computation by using a table.
This removes the need for GCPRO1 etc. Suggested by Stefan Monnier in:
http://lists.gnu.org/archive/html/emacs-devel/2015-08/msg00918.html
* doc/lispref/internals.texi (Writing Emacs Primitives):
* etc/NEWS:
Document the change.
* src/alloc.c (gcprolist, dump_zombies, MAX_ZOMBIES, zombies)
(nzombies, ngcs, avg_zombies, max_live, max_zombies, avg_live)
(Fgc_status, check_gcpros, relocatable_string_data_p, gc-precise):
* src/bytecode.c (mark_byte_stack) [BYTE_MARK_STACK]:
* src/eval.c (gcpro_level) [DEBUG_GCPRO]:
* src/lisp.h (struct handler.gcpro, struct gcpro, GC_MARK_STACK)
(GC_USE_GCPROS_AS_BEFORE, GC_MAKE_GCPROS_NOOPS)
(GC_MARK_STACK_CHECK_GCPROS, GC_USE_GCPROS_CHECK_ZOMBIES)
(BYTE_MARK_STACK, GCPRO1, GCPRO2, GCPRO3, GCPRO4, GCPRO5, GCPRO6)
(GCPRO7, UNGCPRO, RETURN_UNGCPRO):
Remove. All uses removed. The code now assumes
GC_MARK_STACK == GC_MAKE_GCPROS_NOOPS.
* src/bytecode.c (relocate_byte_stack):
Rename from unmark_byte_stack, since it now only relocates.
All callers changed.
* src/frame.c (make_frame): Add an IF_LINT to pacify GCC 5.2
with GCPROs removed.
* src/systime.h: Use EMACS_LISP_H as the canary instead of GCPRO1.
* test/automated/finalizer-tests.el (finalizer-basic)
(finalizer-circular-reference, finalizer-cross-reference)
(finalizer-error):
* test/automated/generator-tests.el (cps-test-iter-close-finalizer):
Remove tests, as they depend on gc-precise.
That way, the caller doesn’t have to use curved quotes to
get diagnostics that match the text-quoting-style preferences.
Suggested by Dmitry Gutov in:
http://lists.gnu.org/archive/html/emacs-devel/2015-08/msg00893.html
This means we no longer need %qs, so remove that format.
While we’re at it, fix an unlikely bug and lessen the pressure
on the garbage collector by processing the string once rather
than twice in the usual case.
* doc/lispref/strings.texi (Formatting Strings):
* etc/NEWS: Document this.
* lisp/subr.el (format-message): Remove; now done in C.
* src/callint.c (Fcall_interactively):
* src/editfns.c (Fmessage, Fmessage_box):
Use Fformat_message instead of Finternal__text_restyle
followed by Fformat.
* src/doc.c (LSQM, RSQM): Remove; all uses changed to use
uLSQM and uRSQM.
(Fsubstitute_command_keys): Prefer AUTO_STRING to build_string
when pure ASCII now suffices. Fix unlikely bug when parsing
unibyte string containing non-ASCII bytes. Use inline code
rather than memcpy, as it’s a tiny number of bytes.
(Finternal__text_restyle): Remove; no longer used.
(syms_of_doc): Don’t declare it.
* src/editfns.c (Fformat): Rewrite in terms of new function
‘styled_format’.
(Fformat_message): New function, moved here from subr.el.
(styled_format): New function, with the old guts of Fformat,
except it now optionally transliterates quotes, and it transliterates
traditional grave accent and apostrophe quoting as well.
Remove recently-added q flag; no longer needed or used.
(syms_of_editfns): Define format-message.
* src/lisp.h (uLSQM0, uLSQM1, uLSQM2, uRSQM0, uRSQM1, uRSQM2):
Remove; no longer need to be global symbols.
* src/xdisp.c (vadd_to_log): Use Fformat_message, not Fformat,
so that callers can use `%s'.
* src/image.c (image_size_error, xbm_load_image, xbm_load)
(xpm_load, pbm_load, png_load_body, jpeg_load_body, tiff_load)
(gif_load, imagemagick_load_image, imagemagick_load, svg_load)
(svg_load_image, gs_load, x_kill_gs_process):
* src/lread.c (load_warn_old_style_backquotes):
* src/xfaces.c (load_pixmap):
* src/xselect.c (x_clipboard_manager_error_1):
Use `%s' instead of %qs in formats.
This is a followup to the recent doc string change, and deals with
diagnostics and the like. This patch is more conservative than
the doc string change, in that the behavior of ‘format’ changes
only if its first arg contains curved quotes and the user prefers
straight or grave quotes. (Come to think of it, perhaps we should
be similarly conservative with doc strings too, but that can wait.)
The upside of this conservatism is that existing usage is almost
surely unaffected. The downside is that we'll eventually have to
change Emacs's format strings to use curved quotes in places where
the user might want curved quotes, but that's a simple and
mechanical translation that I'm willing to do later. (Bug#21222)
* doc/lispref/help.texi (Keys in Documentation):
Move description of text-quoting-style from here ...
* doc/lispref/strings.texi (Formatting Strings):
... to here, and describe new behavior of ‘format’.
* etc/NEWS: Describe new behavior.
* lisp/calc/calc-help.el (calc-describe-thing):
* lisp/emacs-lisp/derived.el (derived-mode-make-docstring):
* lisp/info.el (Info-find-index-name):
Use ‘concat’ rather than ‘format’ to avoid misinterpretation
of recently-added curved quotes.
* src/doc.c (uLSQM0, uLSQM1, uLSQM2, uRSQM0, uRSQM1, uRSQM2):
Move from here ...
* src/lisp.h: ... to here.
* src/doc.c (text_quoting_style): New function.
(Fsubstitute_command_keys): Use it.
* src/editfns.c (Fformat): Implement new behavior.
* src/lisp.h (enum text_quoting_style): New enum.
* src/alloc.c (run_finalizer_handler):
* src/charset.c (load_charset_map_from_vector):
* src/nsimage.m (ns_load_image):
* src/xfaces.c (load_pixmap, load_color2):
Simplify, now that add_to_log has a variable number of args.
* src/image.c (image_error): Take a variable number of args.
Callers simplified.
* src/lisp.h (add_to_log, vadd_to_log): Adjust to new APIs.
* src/xdisp.c (format_nargs, vadd_to_log): New functions.
(add_to_log): Make varargs, and reimplement in terms of vadd_to_log.
* src/xfaces.c (merge_face_ref): Fix typo that omitted color name.
This simplifies time conversions in other time zones.
It also prevents display-time-world tampering with TZ (Bug#21020).
* admin/admin.el (add-release-logs):
Use improved add-log-time-format API.
* admin/merge-gnulib (GNULIB_MODULES): Add time_rz, timegm.
(GNULIB_TOOL_FLAGS): Avoid flexmember, setenv, unsetenv.
* configure.ac (tzalloc): Remove test for this, since
Emacs no longer uses HAVE_TZALLOC directly.
* doc/lispref/os.texi (Time of Day, Time Conversion)
(Time Parsing):
* etc/NEWS: Document the new behavior.
Merge from gnulib, incorporating:
2015-07-25 strftime: fix newly-introduced bug on Solaris
2015-07-23 fprintftime, strftime: use timezone_t args
* lib/gnulib.mk, m4/gnulib-comp.m4: Regenerate.
* lib/strftime.c, lib/strftime.h, lib/time.in.h, m4/sys_time_h.m4:
* m4/time_h.m4:
Update from gnulib.
* lib/time_rz.c, lib/timegm.c, m4/time_rz.m4, m4/timegm.m4:
New files from gnulib.
* lisp/time-stamp.el (time-stamp-string):
* lisp/time.el (display-time-world-list)
(display-time-world-display):
Use new API, with time zone arg.
* lisp/time.el (display-time-world-display):
Fix race when current-time advances while we're running.
* lisp/vc/add-log.el (add-log-iso8601-time-zone)
(add-log-iso8601-time-string): Accept optional time zone arg.
* lisp/vc/add-log.el (add-change-log-entry):
* lisp/vc/log-edit.el (log-edit-changelog-ours-p): Use new arg.
* nt/gnulib.mk: Propagate lib/gnulib.mk changes here.
Add rules for the time module, since they're now needed
for tzalloc etc.
* src/conf_post.h (getenv_TZ, setenv_TZ): New macros.
(emacs_getenv_TZ, emacs_setenv_TZ): New decls.
* src/editfns.c: Include errno.h.
(set_time_zone_rule): Omit unnecessary forward decl.
(initial_tz): Remove, replacing with ...
(local_tz, wall_clock_tz, utc_tz): New static vars and constants.
(tzeqlen): New constant; prefer it to (sizeof "TZ=" - 1).
(emacs_localtime_rz, emacs_mktime_z, xtzalloc, xtzfree)
(tzlookup): New static functions.
(init_editfns): New arg DUMPING. All uses changed.
(init_editfns): Omit most initialization if dumping, not if
!initialized. Initialize wall_clock_tz and local_tz.
(emacs_nmemftime, format_time_string): Time zone argument can now
be any time zone, not just a boolean for UTC or local time. All
callers changed.
(Fformat_time_string, Fencode_time, Fcurrent_time_string)
(Fcurrent_time_zone): New optional arg ZONE.
(Fdecode_time, Fset_time_zone_rule): ZONE arg can now also take
the same form as with the other new additions.
(decode_time_zone): Remove; no longer needed.
(tzvalbuf): Now file-scope.
(emacs_getenv_TZ, emacs_setenv_TZ): New functions.
(syms_of_editfns): Define Qwall.
* src/editfns.c (mktime_z) [!HAVE_TZALLOC]:
* src/systime.h (mktime_z, timezone_t, tzalloc, tzfree)
[!HAVE_TZALLOC]:
Remove; now supplied by gnulib.
* src/emacs.c (main):
* src/lisp.h (init_editfns): Adjust to init_editfns API change.
Improve the heuristic for distinguishing stack overflows from
other SIGSEGV causes (Bug#21004). Corinna Vinschen explained that
the getrlimit method wasn't portable to Cygwin; see:
https://www.cygwin.com/ml/cygwin/2015-07/msg00092.html
Corinna suggested pthread_getattr_np but this also has problems.
Instead, replace the low-level system stuff with a simple
heuristic based on known good stack addresses.
* src/eval.c, src/lisp.h (near_C_stack_top): New function.
* src/sysdep.c: Don't include <sys/resource.h>.
(stack_direction): Remove. All uses removed.
(stack_overflow): New function.
(handle_sigsegv): Use it instead of incorrect getrlimit heuristic.
Make SEGV fatal in non-main threads.
Use the same alignment for the !USE_LSB_TAG case as for the
more-typical USE_LSB_TAG case. The attempt to support arbitrary
alignments with !USE_LSB_TAG had subtle bugs in garbage collection
once we changed the representation of symbols so that Qnil == 0.
Problem reported by Eli Zaretskii (Bug#20862).
* src/alloc.c (XMALLOC_HEADER_ALIGNMENT) [XMALLOC_OVERRUN_CHECK]:
* src/alloc.c (vector_alignment, union aligned_Lisp_Symbol)
(union aligned_Lisp_Misc, maybe_lisp_pointer, pure_alloc):
Use same alignment for !USE_LSB_TAG as for USE_LSB_TAG.
* src/alloc.c (POINTERS_MIGHT_HIDE_IN_OBJECTS): Remove.
This optimization in the !USE_LSB_TAG case is no longer valid when
symbols are represented via offsets. Change the only use to
assume that pointers might hide in objects.
* src/lisp.h (alignas) [!USE_LSB_TAG]:
Require support in this case, too.
(TAG_SYMOFFSET, XSYMBOL) [!USE_LSB_TAG]: Do not shift the offset.
This is OK, because the !USE_LSB_TAG case now applies only when
Lisp_Object is wider than void *, so there's no longer any need
to shift the offset. Not shifting the offset means that
symbol representations have the same alignment as pointers,
which the GC assumes.
* src/editfns.c, src/lisp.h (format2): Remove.
It is more trouble than it's worth, now that we have CALLN.
This is just a minor refactoring.
* src/buffer.c (Fkill_buffer):
* src/dbusbind.c (XD_OBJECT_TO_STRING):
* src/fileio.c (barf_or_query_if_file_exists):
Adjust to format2 going away.
* src/print.c (PRINTDECLARE): Remove. Move its contents into
PRINTPREPARE; doable now that we assume C99. All callers changed.
(PRINTCHAR): Remove, as it adds more mystery than clarity.
All callers changed.
(strout): Assume that caller computes length. All callers changed.
(print_c_string): New function.
(write_string, write_string_1): Compute length instead of asking
the caller to compute it. All callers changed.
(write_string): Simplify by using write_string_1.
(write_string_1): Simplify by using print_c_string.
(Fterpri): Compute default val more clearly.
(Fprin1_to_string, print_object):
Assume C99 to avoid unnecessary nesting.
(print_object): Prefer print_c_string to multiple printchar, or
to calling strout with -1 length. Coalesce into sprintf when
this is easy.
src/emacs.c <w32_daemon_event> [WINDOWSNT]: New global var.
(main) [WINDOWSNT]: Initialize it to NULL. Create the event to
signal clients we are ready for connections.
(Fdaemon_initialized): Use DAEMON_RUNNING.
[WINDOWSNT]: MS-Windows specific code to signal clients we are
ready for connections.
src/lisp.h (DAEMON_RUNNING): New macro, encapsulates Posix and
MS-Windows conditions for running in daemon mode.
src/minibuf.c (read_minibuf): Use DAEMON_RUNNING.
src/keyboard.c (kbd_buffer_get_event): Use DAEMON_RUNNING.
src/dispnew.c (init_display) [WINDOWSNT]: Initialize frames/terminal
even in daemon mode.
nt/inc/ms-w32.h (W32_DAEMON_EVENT): New macro.
lib-src/emacsclient.c (decode_options) [WINDOWSNT]: Don't reject empty
arguments for --alternate-editor.
(print_help_and_exit) [WINDOWSNT]: Don't refrain from advertising
empty arguments for --alternate-editor.
(start_daemon_and_retry_set_socket) [WINDOWSNT]: MS-Windows
specific code to start Emacs in daemon mode and wait for it to be
ready for client connections.
lisp/server.el (server-process-filter): Force GUI frames on
MS-Windows in daemon mode, even if a TTY frame was requested.
lisp/frameset.el (frameset-keep-original-display-p): Don't assume
windows-nt cannot be in daemon mode.
lisp/frame.el (window-system-for-display): Don't assume windows-nt
cannot be in daemon mode.
* lisp/cedet/semantic/doc.el (semantic-documentation-comment-preceding-tag):
Rename from semantic-documentation-comment-preceeding-tag. All
uses changed. Leave an obsolete alias behind.
* src/lisp.h (DEFINE_NON_NIL_Q_SYMBOL_MACROS):
Rename from DEFINE_NONNIL_Q_SYMBOL_MACROS. All uses changed.
See the thread containing:
http://lists.gnu.org/archive/html/emacs-devel/2015-02/msg00720.html
* lib-src/make-docfile.c (write_globals): Generate code that #defines
Qxxx macros other than Qnil only if DEFINE_NONNIL_Q_SYMBOL_MACROS.
Qnil is safe to define even in plugins, since it must be zero for
other reasons.
* src/lisp.h (DEFINE_LISP_SYMBOL): New macro, replacing and simplifying
DEFINE_LISP_SYMBOL_BEGIN / DEFINE_LISP_SYMBOL_END. All uses changed.
(DEFINE_NONNIL_Q_SYMBOL_MACROS): New macro, defaulting to true.
Suggested by Stefan Monnier in:
http://lists.gnu.org/archive/html/emacs-devel/2015-01/msg00588.html
* alloc.c (allocate_pseudovector):
Use memclear, not memsetnil, to remove a 'verify'.
* callint.c (Fcall_interactively):
* dispnew.c (realloc_glyph_pool):
* xdisp.c (init_iterator):
Use memclear, not memset, to remove a 'verify'.
* lisp.h (memclear): Rename from memsetnil, and take a byte
count rather than a word count. All callers changed.
This mostly undoes the previous change in this area. See:
http://lists.gnu.org/archive/html/emacs-devel/2015-01/msg00570.html
* alloc.c (allocate_pseudovector):
* callint.c (Fcall_interactively):
* dispnew.c (realloc_glyph_pool):
* fringe.c (init_fringe):
* lisp.h (memsetnil):
* xdisp.c (init_iterator):
Simplify by assuming that Qnil is zero, but verify the assumption.
* lisp.h (NIL_IS_ZERO): Revert back to this symbol, removing
NIL_IS_NONZERO. All uses changed.
* alloc.c (allocate_pseudovector):
* callint.c (Fcall_interactively):
* coding.c (syms_of_coding):
* dispnew.c (realloc_glyph_pool):
* fringe.c (init_fringe):
* lisp.h (memsetnil):
* xdisp.c (init_iterator):
Port to the currently-hypothetical case where Qnil is nonzero.
* dispnew.c (adjust_glyph_matrix): Remove unnecessary verification,
as there are no Lisp_Object values in the data here.
* lisp.h (NIL_IS_NONZERO): New symbol, replacing NIL_IS_ZERO.
All uses changed. Define only if not already defined, so that one
can debug with -DNIL_IS_NONZERO.
* xdisp.c (init_iterator): Remove unnecessary initializations to 0.
* lisp.h (XTYPE): Remove forward declaration. The recent merge
from emacs-24 fixed the problem in a better way, by moving XPNTR's
definition to after XTYPE's.
lisp.h (XPNTR): Move definition to after XTYPE, to avoid
compilation error in an unoptimized build when !USE_LSB_TAG.
src/w32heap.c (DUMPED_HEAP_SIZE): For 32-bit wide-int build, use the
same larger value as for the 64-bit build.
src/w32term.h (SCROLL_BAR_PACK): Cast the result to UINT_PTR to
avoid compiler warnings.
src/w32proc.c (Fw32_get_codepage_charset, Fw32_set_keyboard_layout):
Avoid compiler warnings about cast from integer to pointer of
different size.
src/w32menu.c (menubar_selection_callback, w32_menu_show): Cast to
UINT_PTR instead of EMACS_INT, to avoid compiler warnings about
casting from integer to pointer of different size.
(add_menu_item): Pass the help-echo string as a pointer to
Lisp_String, not as a Lisp_Object.
(w32_menu_display_help): Use make_lisp_ptr to reconstruct a Lisp
string object from its C pointer.
src/w32fns.c (w32_msg_pump) <WM_EMACS_UNREGISTER_HOT_KEY>: Use
make_lisp_ptr instead of XIL, to reconstruct a Lisp_Cons from its
C pointer.
<WM_EMACS_TOGGLE_LOCK_KEY>: msg.lparam is now a C integer.
(Fx_create_frame): Type-cast the result of XFASTINT to avoild
compiler warnings about size differences.
(Fw32_unregister_hot_key): Pass the tail of w32_grabbed_keys as a
pointer to a Lisp_Cons struct, not as a disguised EMACS_INT.
(Fw32_toggle_lock_key): Pass the new state of the key as a C
integer; use -1 for nil. Doc fix.
src/.gdbinit (xgetsym): New subroutine.
(xsymname, xsymbol): Use it.
(xprintsym): No need to call xgetptr.
Problem reported by Eli Zaretskii in:
http://lists.gnu.org/archive/html/emacs-devel/2015-01/msg00265.html
* lisp.h (struct Lisp_Sub_Char_Table): Check that offset matches
what we think it is, rather than checking only its alignment (and
doing so incorrectly on MinGW).
E.g., when configuring --with-wide-int CFLAGS='-O0' on x86,
the inline function XTYPE needs to be declared before being used.
* lisp.h (XTYPE): New forward declaration.
Prefer symbol indexes to struct Lisp_Symbol * casted and then
widened, as the latter had trouble with GCC on Fedora 21 when
configured --with-wide-int and when used in static initializers.
* lib-src/make-docfile.c (write_globals): Define and use symbols like
iQnil (a small integer, like 0) rather than aQnil (an address
constant).
* src/alloc.c (garbage_collect_1, which_symbols):
* src/lread.c (init_obarray):
Prefer builtin_lisp_symbol when it can be used.
* src/dispextern.h (struct image_type.type):
* src/font.c (font_property_table.key):
* src/frame.c (struct frame_parm_table.sym):
* src/keyboard.c (scroll_bar_parts, struct event_head):
* src/xdisp.c (struct props.name):
Use the index of a builtin symbol rather than its address.
All uses changed.
* src/lisp.h (TAG_SYMPTR, XSYMBOL_INIT): Remove, replacing with ...
(TAG_SYMOFFSET, SYMBOL_INDEX): ... new macros that deal with
symbol indexes rather than pointers, and which work better on MSB
hosts because they shift right before tagging. All uses changed.
(DEFINE_LISP_SYMBOL_BEGIN, DEFINE_LISP_SYMBOL_END):
No longer noops on wide-int hosts, since they work now.
(builtin_lisp_symbol): New function.
* make-docfile.c (close_emacs_globals): Align lispsym to GCALIGNMENT.
The alignment is required on all platforms; it just happens to have
been properly aligned on the previous platforms we tested.
* lisp.h (lisp_h_XSYMBOL, XSYMBOL): Assume USE_LSB_TAG in the
macro-implemented version. For the non-USE_LSB_TAG case, supply
a new inline function that is the inverse of the new TAG_SYMPTR.
(lisp_h_XUNTAGBASE, XUNTAGBASE): Remove. All uses removed.
(TAG_SYMPTR) [!USE_LSB_TAG]: If the pointer subtraction yields a
negative number, don't allow sign bits to bleed into the encoded
value. Shift in zero bits instead.
