view src/unexec.c @ 5353:38e24b8be4ea

Improve the lexical scoping in #'block, #'return-from. lisp/ChangeLog addition: 2011-02-07 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el: * bytecomp.el (byte-compile-initial-macro-environment): Shadow `block', `return-from' here, we implement them differently when byte-compiling. * bytecomp.el (byte-compile-active-blocks): New. * bytecomp.el (byte-compile-block-1): New. * bytecomp.el (byte-compile-return-from-1): New. * bytecomp.el (return-from-1): New. * bytecomp.el (block-1): New. These are two aliases that exist to have their own associated byte-compile functions, which functions implement `block' and `return-from'. * cl-extra.el (cl-macroexpand-all): Fix a bug here when macros in the environment have been compiled. * cl-macs.el (block): * cl-macs.el (return): * cl-macs.el (return-from): Be more careful about lexical scope in these macros. * cl.el: * cl.el ('cl-block-wrapper): Removed. * cl.el ('cl-block-throw): Removed. These aren't needed in code generated by this XEmacs. They shouldn't be needed in code generated by XEmacs 21.4, but if it turns out the packages do need them, we can put them back. 2011-01-30 Mike Sperber <mike@xemacs.org> * font-lock.el (font-lock-fontify-pending-extents): Don't fail if `font-lock-mode' is unset, which can happen in the middle of `revert-buffer'. 2011-01-23 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (delete): * cl-macs.el (delq): * cl-macs.el (remove): * cl-macs.el (remq): Don't use the compiler macro if these functions were given the wrong number of arguments, as happens in lisp-tests.el. * cl-seq.el (remove, remq): Removed. I added these to subr.el, and forgot to remove them from here. 2011-01-22 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-setq, byte-compile-set): Remove kludge allowing keywords' values to be set, all the code that does that is gone. * cl-compat.el (elt-satisfies-test-p): * faces.el (set-face-parent): * faces.el (face-doc-string): * gtk-font-menu.el: * gtk-font-menu.el (gtk-reset-device-font-menus): * msw-font-menu.el: * msw-font-menu.el (mswindows-reset-device-font-menus): * package-get.el (package-get-installedp): * select.el (select-convert-from-image-data): * sound.el: * sound.el (load-sound-file): * x-font-menu.el (x-reset-device-font-menus-core): Don't quote keywords, they're self-quoting, and the win from backward-compatibility is sufficiently small now that the style problem overrides it. 2011-01-22 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (block, return-from): Require that NAME be a symbol in these macros, as always documented in the #'block docstring and as required by Common Lisp. * descr-text.el (unidata-initialize-unihan-database): Correct the use of non-symbols in #'block and #'return-from in this function. 2011-01-15 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (concatenate): Accept more complicated TYPEs in this function, handing the sequences over to #'coerce if we don't understand them here. * cl-macs.el (inline): Don't proclaim #'concatenate as inline, its compiler macro is more useful than doing that. 2011-01-11 Aidan Kehoe <kehoea@parhasard.net> * subr.el (delete, delq, remove, remq): Move #'remove, #'remq here, they don't belong in cl-seq.el; move #'delete, #'delq here from fns.c, implement them in terms of #'delete*, allowing support for sequences generally. * update-elc.el (do-autoload-commands): Use #'delete*, not #'delq here, now the latter's no longer dumped. * cl-macs.el (delete, delq): Add compiler macros transforming #'delete and #'delq to #'delete* calls. 2011-01-10 Aidan Kehoe <kehoea@parhasard.net> * dialog.el (make-dialog-box): Correct a misplaced parenthesis here, thank you Mats Lidell in 87zkr9gqrh.fsf@mail.contactor.se ! 2011-01-02 Aidan Kehoe <kehoea@parhasard.net> * dialog.el (make-dialog-box): * list-mode.el (display-completion-list): These functions used to use cl-parsing-keywords; change them to use defun* instead, fixing the build. (Not sure what led to me not including this change in d1b17a33450b!) 2011-01-02 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (define-star-compiler-macros): Make sure the form has ITEM and LIST specified before attempting to change to calls with explicit tests; necessary for some tests in lisp-tests.el to compile correctly. (stable-union, stable-intersection): Add compiler macros for these functions, in the same way we do for most of the other functions in cl-seq.el. 2011-01-01 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (dolist, dotimes, do-symbols, macrolet) (symbol-macrolet): Define these macros with defmacro* instead of parsing the argument list by hand, for the sake of style and readability; use backquote where appropriate, instead of calling #'list and and friends, for the same reason. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * x-misc.el (device-x-display): Provide this function, documented in the Lispref for years, but not existing previously. Thank you Julian Bradfield, thank you Jeff Mincy. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * cl-seq.el: Move the heavy lifting from this file to C. Dump the cl-parsing-keywords macro, but don't use defun* for the functions we define that do take keywords, dynamic scope lossage makes that not practical. * subr.el (sort, fillarray): Move these aliases here. (map-plist): #'nsublis is now built-in, but at this point #'eql isn't necessarily available as a test; use #'eq. * obsolete.el (cl-delete-duplicates): Make this available for old compiler macros and old code. (memql): Document that this is equivalent to #'member*, and worse. * cl.el (adjoin, subst): Removed. These are in C. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * simple.el (assoc-ignore-case): Remove a duplicate definition of this function (it's already in subr.el). * iso8859-1.el (char-width): On non-Mule, make this function equivalent to that produced by (constantly 1), but preserve its docstring. * subr.el (subst-char-in-string): Define this in terms of #'substitute, #'nsubstitute. (string-width): Define this using #'reduce and #'char-width. (char-width): Give this a simpler definition, it makes far more sense to check for mule at load time and redefine, as we do in iso8859-1.el. (store-substring): Implement this in terms of #'replace, now #'replace is cheap. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * update-elc.el (lisp-files-needed-for-byte-compilation) (lisp-files-needing-early-byte-compilation): cl-macs belongs in the former, not the latter, it is as fundamental as bytecomp.el. 2010-12-30 Aidan Kehoe <kehoea@parhasard.net> * cl.el: Provde the Common Lisp program-error, type-error as error symbols. This doesn't nearly go far enough for anyone using the Common Lisp errors. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (delete-duplicates): If the form has an incorrect number of arguments, don't attempt a compiler macroexpansion. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (cl-safe-expr-p): Forms that start with the symbol lambda are also safe. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (= < > <= >=): For these functions' compiler macros, the optimisation is safe even if the first and the last arguments have side effects, since they're only used the once. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (inline-side-effect-free-compiler-macros): Unroll a loop here at macro-expansion time, so these compiler macros are compiled. Use #'eql instead of #'eq in a couple of places for better style. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (notany, notevery): Avoid some dynamic scope stupidity with local variable names in these functions, when they weren't prefixed with cl-; go into some more detail in the doc strings. 2010-12-29 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (side-effect-free-fns): #'remove, #'remq are free of side-effects. (side-effect-and-error-free-fns): Drop dot, dot-marker from the list. 2010-11-17 Aidan Kehoe <kehoea@parhasard.net> * cl-extra.el (coerce): In the argument list, name the first argument OBJECT, not X; the former name was always used in the doc string and is clearer. Handle vector type specifications which include the length of the target sequence, error if there's a mismatch. * cl-macs.el (cl-make-type-test): Handle type specifications starting with the symbol 'eql. 2010-11-14 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (eql): Don't remove the byte-compile property of this symbol. That was necessary to override a bug in bytecomp.el where #'eql was confused with #'eq, which bug we no longer have. If neither expression is constant, don't attempt to handle the expression in this compiler macro, leave it to byte-compile-eql, which produces better code anyway. * bytecomp.el (eq): #'eql is not the function associated with the byte-eq byte code. (byte-compile-eql): Add an explicit compile method for this function, for cases where the cl-macs compiler macro hasn't reduced it to #'eq or #'equal. 2010-10-25 Aidan Kehoe <kehoea@parhasard.net> Add compiler macros and compilation sanity-checking for various functions that take keywords. * byte-optimize.el (side-effect-free-fns): #'symbol-value is side-effect free and not error free. * bytecomp.el (byte-compile-normal-call): Check keyword argument lists for sanity; store information about the positions where keyword arguments start using the new byte-compile-keyword-start property. * cl-macs.el (cl-const-expr-val): Take a new optional argument, cl-not-constant, defaulting to nil, in this function; return it if the expression is not constant. (cl-non-fixnum-number-p): Make this into a separate function, we want to pass it to #'every. (eql): Use it. (define-star-compiler-macros): Use the same code to generate the member*, assoc* and rassoc* compiler macros; special-case some code in #'add-to-list in subr.el. (remove, remq): Add compiler macros for these two functions, in preparation for #'remove being in C. (define-foo-if-compiler-macros): Transform (remove-if-not ...) calls to (remove ... :if-not) at compile time, which will be a real win once the latter is in C. (define-substitute-if-compiler-macros) (define-subst-if-compiler-macros): Similarly for these functions. (delete-duplicates): Change this compiler macro to use #'plists-equal; if we don't have information about the type of SEQUENCE at compile time, don't bother attempting to inline the call, the function will be in C soon enough. (equalp): Remove an old commented-out compiler macro for this, if we want to see it it's in version control. (subst-char-in-string): Transform this to a call to nsubstitute or nsubstitute, if that is appropriate. * cl.el (ldiff): Don't call setf here, this makes for a load-time dependency problem in cl-macs.el 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * term/vt100.el: Refer to XEmacs, not GNU Emacs, in permissions. * term/bg-mouse.el: * term/sup-mouse.el: Put copyright notice in canonical "Copyright DATE AUTHOR" form. Refer to XEmacs, not GNU Emacs, in permissions. * site-load.el: Add permission boilerplate. * mule/canna-leim.el: * alist.el: Refer to XEmacs, not APEL/this program, in permissions. * mule/canna-leim.el: Remove my copyright, I've assigned it to the FSF. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * gtk.el: * gtk-widget-accessors.el: * gtk-package.el: * gtk-marshal.el: * gtk-compose.el: * gnome.el: Add copyright notice based on internal evidence. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * easymenu.el: Add reference to COPYING to permission notice. * gutter.el: * gutter-items.el: * menubar-items.el: Fix typo "Xmacs" in permissions notice. 2010-06-14 Stephen J. Turnbull <stephen@xemacs.org> * auto-save.el: * font.el: * fontconfig.el: * mule/kinsoku.el: Add "part of XEmacs" text to permission notice. 2010-10-14 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (side-effect-free-fns): * cl-macs.el (remf, getf): * cl-extra.el (tailp, cl-set-getf, cl-do-remf): * cl.el (ldiff, endp): Tighten up Common Lisp compatibility for #'ldiff, #'endp, #'tailp; add circularity checking for the first two. #'cl-set-getf and #'cl-do-remf were Lisp implementations of #'plist-put and #'plist-remprop; change the names to aliases, changes the macros that use them to using #'plist-put and #'plist-remprop directly. 2010-10-12 Aidan Kehoe <kehoea@parhasard.net> * abbrev.el (fundamental-mode-abbrev-table, global-abbrev-table): Create both these abbrev tables using the usual #'define-abbrev-table calls, rather than attempting to special-case them. * cl-extra.el: Force cl-macs to be loaded here, if cl-extra.el is being loaded interpreted. Previously other, later files would redundantly call (load "cl-macs") when interpreted, it's more reasonable to do it here, once. * cmdloop.el (read-quoted-char-radix): Use defcustom here, we don't have any dump-order dependencies that would prevent that. * custom.el (eval-when-compile): Don't load cl-macs when interpreted or when byte-compiling, rely on cl-extra.el in the former case and the appropriate entry in bytecomp-load-hook in the latter. Get rid of custom-declare-variable-list, we have no dump-time dependencies that would require it. * faces.el (eval-when-compile): Don't load cl-macs when interpreted or when byte-compiling. * packages.el: Remove some inaccurate comments. * post-gc.el (cleanup-simple-finalizers): Use #'delete-if-not here, now the order of preloaded-file-list has been changed to make it available. * subr.el (custom-declare-variable-list): Remove. No need for it. Also remove a stub define-abbrev-table from this file, given the current order of preloaded-file-list there's no need for it. 2010-10-10 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-constp) Forms quoted with FUNCTION are also constant. (byte-compile-initial-macro-environment): In #'the, if FORM is constant and does not match TYPE, warn at byte-compile time. 2010-10-10 Aidan Kehoe <kehoea@parhasard.net> * backquote.el (bq-vector-contents, bq-list*): Remove; the former is equivalent to (append VECTOR nil), the latter to (list* ...). (bq-process-2): Use (append VECTOR nil) instead of using #'bq-vector-contents to convert to a list. (bq-process-1): Now we use list* instead of bq-list * subr.el (list*): Moved from cl.el, since it is now required to be available the first time a backquoted form is encountered. * cl.el (list*): Move to subr.el. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * test-harness.el (Check-Message): Add an omitted comma here, thank you the buildbot. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * hash-table.el (hash-table-key-list, hash-table-value-list) (hash-table-key-value-alist, hash-table-key-value-plist): Remove some useless #'nreverse calls in these files; our hash tables have no order, it's not helpful to pretend they do. * behavior.el (read-behavior): Do the same in this file, in some code evidently copied from hash-table.el. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * info.el (Info-insert-dir): * format.el (format-deannotate-region): * files.el (cd, save-buffers-kill-emacs): Use #'some, #'every and related functions for applying boolean operations to lists, instead of rolling our own ones that cons and don't short-circuit. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-initial-macro-environment): * cl-macs.el (the): Rephrase the docstring, make its implementation when compiling files a little nicer. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * descr-text.el (unidata-initialize-unicodedata-database) (unidata-initialize-unihan-database, describe-char-unicode-data) (describe-char-unicode-data): Wrap calls to the database functions with (with-fboundp ...), avoiding byte compile warnings on builds without support for the database functions. (describe-char): (reduce #'max ...), not (apply #'max ...), no need to cons needlessly. (describe-char): Remove a redundant lambda wrapping #'extent-properties. (describe-char-unicode-data): Call #'nsubst when replacing "" with nil in the result of #'split-string, instead of consing inside mapcar. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * x-faces.el (x-available-font-sizes): * specifier.el (let-specifier): * package-ui.el (pui-add-required-packages): * msw-faces.el (mswindows-available-font-sizes): * modeline.el (modeline-minor-mode-menu): * minibuf.el (minibuf-directory-files): Replace the O2N (delq nil (mapcar (lambda (W) (and X Y)) Z)) with the ON (mapcan (lambda (W) (and X (list Y))) Z) in these files. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * cl-macs.el (= < > <= >=): When these functions are handed more than two arguments, and those arguments have no side effects, transform to a series of two argument calls, avoiding funcall in the byte-compiled code. * mule/mule-cmds.el (finish-set-language-environment): Take advantage of this change in a function called 256 times at startup. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * bytecomp.el (byte-compile-function-form, byte-compile-quote) (byte-compile-quote-form): Warn at compile time, and error at runtime, if a (quote ...) or a (function ...) form attempts to quote more than one object. 2010-09-16 Aidan Kehoe <kehoea@parhasard.net> * byte-optimize.el (byte-optimize-apply): Transform (apply 'nconc (mapcar ...)) to (mapcan ...); warn about use of the first idiom. * update-elc.el (do-autoload-commands): * packages.el (packages-find-package-library-path): * frame.el (frame-list): * extents.el (extent-descendants): * etags.el (buffer-tag-table-files): * dumped-lisp.el (preloaded-file-list): * device.el (device-list): * bytecomp-runtime.el (proclaim-inline, proclaim-notinline) Use #'mapcan, not (apply #'nconc (mapcar ...) in all these files. * bytecomp-runtime.el (eval-when-compile, eval-and-compile): In passing, mention that these macros also evaluate the body when interpreted. tests/ChangeLog addition: 2011-02-07 Aidan Kehoe <kehoea@parhasard.net> * automated/lisp-tests.el: Test lexical scope for `block', `return-from'; add a Known-Bug-Expect-Failure for a contorted example that fails when byte-compiled.
author Aidan Kehoe <kehoea@parhasard.net>
date Mon, 07 Feb 2011 12:01:24 +0000
parents 304aebb79cd3
children 308d34e9f07d
line wrap: on
line source

/* Copyright (C) 1985, 1986, 1987, 1988, 1992, 1993, 1994
/* Copyright (C) 2001 Ben Wing.
   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.  */

/* Synched up with: FSF 19.31. */

/*
 * unexec.c - Convert a running program into an a.out file.
 *
 * Author:	Spencer W. Thomas
 * 		Computer Science Dept.
 * 		University of Utah
 * Date:	Tue Mar  2 1982
 * Modified heavily since then.
 *
 * Synopsis:
 *	unexec (new_name, a_name, data_start, bss_start, entry_address)
 *	char *new_name, *a_name;
 *	unsigned data_start, bss_start, entry_address;
 *
 * Takes a snapshot of the program and makes an a.out format file in the
 * file named by the string argument new_name.
 * If a_name is non-NULL, the symbol table will be taken from the given file.
 * On some machines, an existing a_name file is required.
 *
 * The boundaries within the a.out file may be adjusted with the data_start
 * and bss_start arguments.  Either or both may be given as 0 for defaults.
 *
 * Data_start gives the boundary between the text segment and the data
 * segment of the program.  The text segment can contain shared, read-only
 * program code and literal data, while the data segment is always unshared
 * and unprotected.  Data_start gives the lowest unprotected address.
 * The value you specify may be rounded down to a suitable boundary
 * as required by the machine you are using.
 *
 * Specifying zero for data_start means the boundary between text and data
 * should not be the same as when the program was loaded.
 * If NO_REMAP is defined, the argument data_start is ignored and the
 * segment boundaries are never changed.
 *
 * Bss_start indicates how much of the data segment is to be saved in the
 * a.out file and restored when the program is executed.  It gives the lowest
 * unsaved address, and is rounded up to a page boundary.  The default when 0
 * is given assumes that the entire data segment is to be stored, including
 * the previous data and bss as well as any additional storage allocated with
 * break (2).
 *
 * The new file is set up to start at entry_address.
 *
 * If you make improvements I'd like to get them too.
 * harpo!utah-cs!thomas, thomas@Utah-20
 *
 */

/* Modified to support SysVr3 shared libraries by James Van Artsdalen
 * of Dell Computer Corporation.  james@bigtex.cactus.org.
 */

/* There are several compilation parameters affecting unexec:

* COFF

Define this if your system uses COFF for executables.

* NO_REMAP

Define this if you do not want to try to save Emacs's pure data areas
as part of the text segment.

Saving them as text is good because it allows users to share more.

However, on machines that locate the text area far from the data area,
the boundary cannot feasibly be moved.  Such machines require
NO_REMAP.

Also, remapping can cause trouble with the built-in startup routine
/lib/crt0.o, which defines `environ' as an initialized variable.
Dumping `environ' as pure does not work!  So, to use remapping,
you must write a startup routine for your machine in Emacs's crt0.c.
If NO_REMAP is defined, Emacs uses the system's crt0.o.

* SEGMENT_MASK

Some machines require that the beginnings and ends of segments
*in core* be on certain boundaries.  For most machines, a page
boundary is sufficient.  That is the default.  When a larger
boundary is needed, define SEGMENT_MASK to a mask of
the bits that must be zero on such a boundary.

* A_TEXT_OFFSET(HDR)

Some machines count the a.out header as part of the size of the text
segment (a_text); they may actually load the header into core as the
first data in the text segment.  Some have additional padding between
the header and the real text of the program that is counted in a_text.

For these machines, define A_TEXT_OFFSET(HDR) to examine the header
structure HDR and return the number of bytes to add to `a_text'
before writing it (above and beyond the number of bytes of actual
program text).  HDR's standard fields are already correct, except that
this adjustment to the `a_text' field has not yet been made;
thus, the amount of offset can depend on the data in the file.

* A_TEXT_SEEK(HDR)

If defined, this macro specifies the number of bytes to seek into the
a.out file before starting to write the text segment.

* EXEC_MAGIC

For machines using COFF, this macro, if defined, is a value stored
into the magic number field of the output file.

* ADJUST_EXEC_HEADER

This macro can be used to generate statements to adjust or
initialize nonstandard fields in the file header

* ADDR_CORRECT(ADDR)

Macro to correct an int which is the bit pattern of a pointer to a byte
into an int which is the number of a byte.

This macro has a default definition which is usually right.
This default definition is a no-op on most machines (where a
pointer looks like an int) but not on all machines.

*/

#ifndef emacs
#define PERROR(arg) perror (arg); return -1
#else
#define IN_UNEXEC
#include <config.h>
#define PERROR(file) report_error (file, new_)
#endif

#if __STDC__ || defined(STDC_HEADERS)

/* I don't know how correct this attempt to get more prototypes is... */
# if defined(sun) && defined(_POSIX_SOURCE)
#  undef _POSIX_SOURCE
# endif

# include <stddef.h>
# include <stdlib.h>
# include <unistd.h>
# include <string.h>
# include <stddef.h>
# include <errno.h>

#endif

/* I don't understand this, but it's necessary to get some slots in struct exec
   from /usr/include/sys/exec.h when running LCC in strict ANSI mode.  We don't
   need this in K&R mode...
 */
#if defined(__lucid) && defined(__sparc) && !defined(sun)
# define sun 1
#endif

#include <a.out.h>

/* Define getpagesize if the system does not.
   Note that this may depend on symbols defined in a.out.h.  */
#include "getpagesize.h"

#ifndef makedev			/* Try to detect types.h already loaded */
#include <sys/types.h>
#endif /* makedev */
#include <stdio.h>
#include <sys/stat.h>
#include <errno.h>

#include <sys/file.h>	/* Must be after sys/types.h for USG */

#ifdef USG5
#include <fcntl.h>
#endif

#include "compiler.h"

#ifndef O_RDONLY
#define O_RDONLY 0
#endif
#ifndef O_RDWR
#define O_RDWR 2
#endif

extern char *start_of_text ();		/* Start of text */

extern void *start_of_data ();		/* Start of initialized data */

#ifdef COFF
static long block_copy_start;		/* Old executable start point */
static struct filehdr f_hdr;		/* File header */
static struct aouthdr f_ohdr;		/* Optional file header (a.out) */
long bias;			/* Bias to add for growth */
long lnnoptr;			/* Pointer to line-number info within file */
#define SYMS_START block_copy_start

static long text_scnptr;
static long data_scnptr;

#else /* not COFF */

#ifdef __STDC__
#ifndef __sys_stdtypes_h
#if !defined(_PTRDIFF_T) && !defined(_BSD_PTRDIFF_T_)
typedef long ptrdiff_t;
#endif
#endif
#ifndef HPUX
/* not sure where this for NetBSD should really go
   and it probably applies to other systems */
#if !defined(__NetBSD__) && !defined(__bsdi__) && !defined(__OpenBSD__)
extern void *sbrk (ptrdiff_t);
#else
extern char *sbrk ();
#endif /* __NetBSD__ or __OpenBSD__ */
#endif /* HPUX */
#else
extern void *sbrk ();
#endif

#define SYMS_START ((long) N_SYMOFF (ohdr))

#ifdef HPUX
#include <model.h>
#define MY_ID MYSYS
static MAGIC OLDMAGIC = {MY_ID, SHARE_MAGIC};
static MAGIC NEWMAGIC = {MY_ID, DEMAND_MAGIC};
#define N_TXTOFF(x) TEXT_OFFSET(x)
#define N_SYMOFF(x) LESYM_OFFSET(x)
static struct exec hdr, ohdr;

#else /* not HPUX */

#if defined (USG) && !defined (LINUX)
static struct bhdr hdr, ohdr;
#define a_magic fmagic
#define a_text tsize
#define a_data dsize
#define a_bss bsize
#define a_syms ssize
#define a_trsize rtsize
#define a_drsize rdsize
#define a_entry entry
#define	N_BADMAG(x) \
    (((x).fmagic)!=OMAGIC && ((x).fmagic)!=NMAGIC &&\
     ((x).fmagic)!=FMAGIC && ((x).fmagic)!=IMAGIC)
#define NEWMAGIC FMAGIC
#else /* !USG or LINUX */
static struct exec hdr, ohdr;
#define NEWMAGIC ZMAGIC
#endif /* !USG or LINUX */
#endif /* not HPUX */

static int unexec_text_start;
static int unexec_data_start;

#endif /* not COFF */

static int pagemask;

/* Correct an int which is the bit pattern of a pointer to a byte
   into an int which is the number of a byte.
   This is a no-op on ordinary machines, but not on all.  */

#ifndef ADDR_CORRECT   /* Let m-*.h files override this definition */
#define ADDR_CORRECT(x) ((char *)(x) - (char*)0)
#endif

#ifdef emacs

#include "lisp.h"

static void
report_error (const char *file, int fd)
{
  if (fd)
    close (fd);
  report_error_with_errno (Qio_error, "Cannot unexec",
			   build_extstring (file, Qfile_name));
}
#endif /* emacs */

#define ERROR0(msg) report_error_1 (new_, msg, 0, 0); return -1
#define ERROR1(msg,x) report_error_1 (new_, msg, x, 0); return -1
#define ERROR2(msg,x,y) report_error_1 (new_, msg, x, y); return -1

static void
report_error_1 (fd, msg, a1, a2)
     int fd;
     const char *msg;
     int a1, a2;
{
  close (fd);
#ifdef emacs
  signal_ferror (Qio_error, msg, a1, a2);
#else
  fprintf (stderr, msg, a1, a2);
  fprintf (stderr, "\n");
#endif
}

static int make_hdr (int new_, int a_out, unsigned data_start,
		     unsigned bss_start, unsigned entry_address,
		     char *a_name, char *new_name);
static int copy_text_and_data (int new_, int a_out);
static int copy_sym (int new_, int a_out, char *a_name, char *new_name);
static void mark_x (char *name);

/* ****************************************************************
 * unexec
 *
 * driving logic.
 */
int
unexec (new_name, a_name, data_start, bss_start, entry_address)
     char *new_name, *a_name;
     unsigned data_start, bss_start, entry_address;
{
  int new_, a_out = -1;

  if (a_name && (a_out = open (a_name, O_RDONLY)) < 0)
    {
      PERROR (a_name);
    }
  if ((new_ = creat (new_name, 0666)) < 0)
    {
      PERROR (new_name);
    }

  if (make_hdr (new_, a_out, data_start, bss_start, entry_address, a_name, new_name) < 0
      || copy_text_and_data (new_, a_out) < 0
      || copy_sym (new_, a_out, a_name, new_name) < 0
#ifdef COFF
      || adjust_lnnoptrs (new_, a_out, new_name) < 0
#endif
      )
    {
      close (new_);
      /* unlink (new_name);	    	/ * Failed, unlink new a.out */
      return -1;
    }

  close (new_);
  if (a_out >= 0)
    close (a_out);
  mark_x (new_name);
  return 0;
}

/* ****************************************************************
 * make_hdr
 *
 * Make the header in the new a.out from the header in core.
 * Modify the text and data sizes.
 */
static int
make_hdr (int new_, int a_out, unsigned data_start, unsigned bss_start,
	  unsigned entry_address, char *a_name, char *new_name)
{
#ifdef COFF
  auto struct scnhdr f_thdr;		/* Text section header */
  auto struct scnhdr f_dhdr;		/* Data section header */
  auto struct scnhdr f_bhdr;		/* Bss section header */
  auto struct scnhdr scntemp;		/* Temporary section header */
  int scns;
#endif /* COFF */
#ifdef USG_SHARED_LIBRARIES
  extern unsigned int bss_end;
#else
  unsigned int bss_end;
#endif

  pagemask = getpagesize () - 1;

  /* Adjust text/data boundary. */
#ifdef NO_REMAP
  data_start = (int) start_of_data ();
#else /* not NO_REMAP */
  if (!data_start)
    data_start = (int) start_of_data ();
#endif /* not NO_REMAP */
  data_start = ADDR_CORRECT (data_start);

#ifdef SEGMENT_MASK
  data_start = data_start & ~SEGMENT_MASK; /* (Down) to segment boundary. */
#else
  data_start = data_start & ~pagemask; /* (Down) to page boundary. */
#endif

  bss_end = ADDR_CORRECT (sbrk (0)) + pagemask;
  bss_end &= ~ pagemask;

  /* Adjust data/bss boundary. */
  if (bss_start != 0)
    {
      bss_start = (ADDR_CORRECT (bss_start) + pagemask);
      /* (Up) to page bdry. */
      bss_start &= ~ pagemask;
      if (bss_start > bss_end)
	{
	  ERROR1 ("unexec: Specified bss_start (%u) is past end of program",
		  bss_start);
	}
    }
  else
    bss_start = bss_end;

  if (data_start > bss_start)	/* Can't have negative data size. */
    {
      ERROR2 ("unexec: data_start (%u) can't be greater than bss_start (%u)",
	      data_start, bss_start);
    }

#ifdef COFF
  /* Salvage as much info from the existing file as possible */
  if (a_out >= 0)
    {
      if (read (a_out, &f_hdr, sizeof (f_hdr)) != sizeof (f_hdr))
	{
	  PERROR (a_name);
	}
      block_copy_start += sizeof (f_hdr);
      if (f_hdr.f_opthdr > 0)
	{
	  if (read (a_out, &f_ohdr, sizeof (f_ohdr)) != sizeof (f_ohdr))
	    {
	      PERROR (a_name);
	    }
	  block_copy_start += sizeof (f_ohdr);
	}
      /* Loop through section headers, copying them in */
      lseek (a_out, sizeof (f_hdr) + f_hdr.f_opthdr, 0);
      for (scns = f_hdr.f_nscns; scns > 0; scns--) {
	if (read (a_out, &scntemp, sizeof (scntemp)) != sizeof (scntemp))
	  {
	    PERROR (a_name);
	  }
	if (scntemp.s_scnptr > 0L)
	  {
            if (block_copy_start < scntemp.s_scnptr + scntemp.s_size)
	      block_copy_start = scntemp.s_scnptr + scntemp.s_size;
	  }
	if (strcmp (scntemp.s_name, ".text") == 0)
	  {
	    f_thdr = scntemp;
	  }
	else if (strcmp (scntemp.s_name, ".data") == 0)
	  {
	    f_dhdr = scntemp;
	  }
	else if (strcmp (scntemp.s_name, ".bss") == 0)
	  {
	    f_bhdr = scntemp;
	  }
      }
    }
  else
    {
      ERROR0 ("can't build a COFF file from scratch yet");
    }

  /* Now we alter the contents of all the f_*hdr variables
     to correspond to what we want to dump.  */

#ifdef USG_SHARED_LIBRARIES

  /* The amount of data we're adding to the file is distance from the
   * end of the original .data space to the current end of the .data
   * space.
   */

  bias = bss_start - (f_ohdr.data_start + f_dhdr.s_size);

#endif

  f_hdr.f_flags |= (F_RELFLG | F_EXEC);
#ifdef EXEC_MAGIC
  f_ohdr.magic = EXEC_MAGIC;
#endif
#ifndef NO_REMAP
  f_ohdr.text_start = (long) start_of_text ();
  f_ohdr.tsize = data_start - f_ohdr.text_start;
  f_ohdr.data_start = data_start;
#endif /* NO_REMAP */
  f_ohdr.dsize = bss_start - f_ohdr.data_start;
  f_ohdr.bsize = bss_end - bss_start;
  /* On some machines, the old values are right.
     ??? Maybe on all machines with NO_REMAP.  */
  f_thdr.s_size = f_ohdr.tsize;
  f_thdr.s_scnptr = sizeof (f_hdr) + sizeof (f_ohdr);
  f_thdr.s_scnptr += (f_hdr.f_nscns) * (sizeof (f_thdr));
  lnnoptr = f_thdr.s_lnnoptr;
  text_scnptr = f_thdr.s_scnptr;
  f_dhdr.s_paddr = f_ohdr.data_start;
  f_dhdr.s_vaddr = f_ohdr.data_start;
  f_dhdr.s_size = f_ohdr.dsize;
  f_dhdr.s_scnptr = f_thdr.s_scnptr + f_thdr.s_size;
  data_scnptr = f_dhdr.s_scnptr;
  f_bhdr.s_paddr = f_ohdr.data_start + f_ohdr.dsize;
  f_bhdr.s_vaddr = f_ohdr.data_start + f_ohdr.dsize;
  f_bhdr.s_size = f_ohdr.bsize;
  f_bhdr.s_scnptr = 0L;
#ifndef USG_SHARED_LIBRARIES
  bias = f_dhdr.s_scnptr + f_dhdr.s_size - block_copy_start;
#endif

  if (f_hdr.f_symptr > 0L)
    {
      f_hdr.f_symptr += bias;
    }

  if (f_thdr.s_lnnoptr > 0L)
    {
      f_thdr.s_lnnoptr += bias;
    }

#ifdef ADJUST_EXEC_HEADER
  ADJUST_EXEC_HEADER;
#endif /* ADJUST_EXEC_HEADER */

  if (write (new_, &f_hdr, sizeof (f_hdr)) != sizeof (f_hdr))
    {
      PERROR (new_name);
    }

  if (write (new_, &f_ohdr, sizeof (f_ohdr)) != sizeof (f_ohdr))
    {
      PERROR (new_name);
    }

#ifndef USG_SHARED_LIBRARIES

  if (write (new_, &f_thdr, sizeof (f_thdr)) != sizeof (f_thdr))
    {
      PERROR (new_name);
    }

  if (write (new_, &f_dhdr, sizeof (f_dhdr)) != sizeof (f_dhdr))
    {
      PERROR (new_name);
    }

  if (write (new_, &f_bhdr, sizeof (f_bhdr)) != sizeof (f_bhdr))
    {
      PERROR (new_name);
    }

#else /* USG_SHARED_LIBRARIES */

  /* The purpose of this code is to write out the new file's section
   * header table.
   *
   * Scan through the original file's sections.  If the encountered
   * section is one we know (.text, .data or .bss), write out the
   * correct header.  If it is a section we do not know (such as
   * .lib), adjust the address of where the section data is in the
   * file, and write out the header.
   *
   * If any section precedes .text or .data in the file, this code
   * will not adjust the file pointer for that section correctly.
   */

  /* This used to use sizeof (f_ohdr) instead of .f_opthdr.
     .f_opthdr is said to be right when there is no optional header.  */
  lseek (a_out, sizeof (f_hdr) + f_hdr.f_opthdr, 0);

  for (scns = f_hdr.f_nscns; scns > 0; scns--)
    {
      if (read (a_out, &scntemp, sizeof (scntemp)) != sizeof (scntemp))
	PERROR (a_name);

      if (!strcmp (scntemp.s_name, f_thdr.s_name))	/* .text */
	{
	  if (write (new_, &f_thdr, sizeof (f_thdr)) != sizeof (f_thdr))
	    PERROR (new_name);
	}
      else if (!strcmp (scntemp.s_name, f_dhdr.s_name))	/* .data */
	{
	  if (write (new_, &f_dhdr, sizeof (f_dhdr)) != sizeof (f_dhdr))
	    PERROR (new_name);
	}
      else if (!strcmp (scntemp.s_name, f_bhdr.s_name))	/* .bss */
	{
	  if (write (new_, &f_bhdr, sizeof (f_bhdr)) != sizeof (f_bhdr))
	    PERROR (new_name);
	}
      else
	{
	  if (scntemp.s_scnptr)
	    scntemp.s_scnptr += bias;
	  if (write (new_, &scntemp, sizeof (scntemp)) != sizeof (scntemp))
	    PERROR (new_name);
	}
    }
#endif /* USG_SHARED_LIBRARIES */

  return (0);

#else /* if not COFF */

  /* Get symbol table info from header of a.out file if given one. */
  if (a_out >= 0)
    {
      if (read (a_out, (char *) &ohdr, sizeof (hdr)) != sizeof (hdr))
	{
	  PERROR (a_name);
	}

      if (N_BADMAG (ohdr))
	{
	  ERROR1 ("invalid magic number in %s", a_name);
	}
      hdr = ohdr;
    }
  else
    {
      memset ((void *)&hdr, 0, sizeof (hdr));
    }

  unexec_text_start = (long) start_of_text ();
  unexec_data_start = data_start;

  /* Machine-dependent fixup for header, or maybe for unexec_text_start */
#ifdef ADJUST_EXEC_HEADER
  ADJUST_EXEC_HEADER;
#endif /* ADJUST_EXEC_HEADER */

  hdr.a_trsize = 0;
  hdr.a_drsize = 0;
  if (entry_address != 0)
    hdr.a_entry = entry_address;

  hdr.a_bss = bss_end - bss_start;
  hdr.a_data = bss_start - data_start;
#ifdef NO_REMAP
  hdr.a_text = ohdr.a_text;
#else /* not NO_REMAP */
  hdr.a_text = data_start - unexec_text_start;

#ifdef A_TEXT_OFFSET
  hdr.a_text += A_TEXT_OFFSET (ohdr);
#endif

#endif /* not NO_REMAP */

  if (write (new_, (char *) &hdr, sizeof (hdr)) != sizeof (hdr))
    {
      PERROR (new_name);
    }

#if 0 /* This #ifndef caused a bug on Linux when using QMAGIC.  */
  /* This adjustment was done above only #ifndef NO_REMAP,
     so only undo it now #ifndef NO_REMAP.  */
  /* #ifndef NO_REMAP  */
#endif
#ifdef A_TEXT_OFFSET
  hdr.a_text -= A_TEXT_OFFSET (ohdr);
#endif

  return 0;

#endif /* not COFF */
}

static void write_segment (int, char *, char *);

/* ****************************************************************
 * copy_text_and_data
 *
 * Copy the text and data segments from memory to the new a.out
 */
static int
copy_text_and_data (int new_, 
#if defined (COFF) && defined (USG_SHARED_LIBRARIES)
		    int a_out
#else
		    int UNUSED (a_out)
#endif
		    )
{
  char *end;
  char *ptr;

#ifdef COFF

#ifdef USG_SHARED_LIBRARIES

  int scns;
  struct scnhdr scntemp;		/* Temporary section header */

  /* The purpose of this code is to write out the new file's section
   * contents.
   *
   * Step through the section table.  If we know the section (.text,
   * .data) do the appropriate thing.  Otherwise, if the section has
   * no allocated space in the file (.bss), do nothing.  Otherwise,
   * the section has space allocated in the file, and is not a section
   * we know.  So just copy it.
   */

  lseek (a_out, sizeof (struct filehdr) + sizeof (struct aouthdr), 0);

  for (scns = f_hdr.f_nscns; scns > 0; scns--)
    {
      if (read (a_out, &scntemp, sizeof (scntemp)) != sizeof (scntemp))
	PERROR ("temacs");

      if (!strcmp (scntemp.s_name, ".text"))
	{
	  lseek (new_, (long) text_scnptr, 0);
	  ptr = (char *) f_ohdr.text_start;
	  end = ptr + f_ohdr.tsize;
	  write_segment (new_, ptr, end);
	}
      else if (!strcmp (scntemp.s_name, ".data"))
	{
	  lseek (new_, (long) data_scnptr, 0);
	  ptr = (char *) f_ohdr.data_start;
	  end = ptr + f_ohdr.dsize;
	  write_segment (new_, ptr, end);
	}
      else if (!scntemp.s_scnptr)
	; /* do nothing - no data for this section */
      else
	{
	  char page[BUFSIZ];
	  int size, n;
	  long old_a_out_ptr = lseek (a_out, 0, 1);

	  lseek (a_out, scntemp.s_scnptr, 0);
	  for (size = scntemp.s_size; size > 0; size -= sizeof (page))
	    {
	      n = size > sizeof (page) ? sizeof (page) : size;
	      if (read (a_out, page, n) != n || write (new_, page, n) != n)
		PERROR ("emacs");
	    }
	  lseek (a_out, old_a_out_ptr, 0);
	}
    }

#else /* COFF, but not USG_SHARED_LIBRARIES */

  lseek (new_, (long) text_scnptr, 0);
  ptr = (char *) f_ohdr.text_start;
  end = ptr + f_ohdr.tsize;
  write_segment (new_, ptr, end);

  lseek (new_, (long) data_scnptr, 0);
  ptr = (char *) f_ohdr.data_start;
  end = ptr + f_ohdr.dsize;
  write_segment (new_, ptr, end);

#endif /* USG_SHARED_LIBRARIES */

#else /* if not COFF */

/* Some machines count the header as part of the text segment.
   That is to say, the header appears in core
   just before the address that start_of_text returns.
   For them, N_TXTOFF is the place where the header goes.
   We must adjust the seek to the place after the header.
   Note that at this point hdr.a_text does *not* count
   the extra A_TEXT_OFFSET bytes, only the actual bytes of code.  */

#ifdef A_TEXT_SEEK
  lseek (new_, (long) A_TEXT_SEEK (hdr), 0);
#else
  lseek (new_, (long) N_TXTOFF (hdr), 0);
#endif /* no A_TEXT_SEEK */

  ptr = (char *) unexec_text_start;
  end = ptr + hdr.a_text;
  write_segment (new_, ptr, end);

  ptr = (char *) unexec_data_start;
  end = ptr + hdr.a_data;
/*  This lseek is certainly incorrect when A_TEXT_OFFSET
    and I believe it is a no-op otherwise.
    Let's see if its absence ever fails.  */
/*  lseek (new_, (long) N_TXTOFF (hdr) + hdr.a_text, 0); */
  write_segment (new_, ptr, end);

#endif /* not COFF */

  return 0;
}

static void
write_segment (new_, ptr, end)
     int new_;
     char *ptr, *end;
{
  int i, nwrite, ret;
#if 0
  char buf[80];
#endif
  /* This is the normal amount to write at once.
     It is the size of block that NFS uses.  */
  int writesize = 1 << 13;
  int pagesize = getpagesize ();
  char zeros[1 << 13];

  memset (zeros, 0, sizeof (zeros));

  for (i = 0; ptr < end;)
    {
      /* Distance to next multiple of writesize.  */
      nwrite = (((int) ptr + writesize) & -writesize) - (int) ptr;
      /* But not beyond specified end.  */
      if (nwrite > end - ptr) nwrite = end - ptr;
      ret = write (new_, ptr, nwrite);
      /* If write gets a page fault, it means we reached
	 a gap between the old text segment and the old data segment.
	 This gap has probably been remapped into part of the text segment.
	 So write zeros for it.  */
      if (ret == -1
#ifdef EFAULT
	  && errno == EFAULT
#endif
	  )
	{
	  /* Write only a page of zeros at once,
	     so that we don't overshoot the start
	     of the valid memory in the old data segment.  */
	  if (nwrite > pagesize)
	    nwrite = pagesize;
	  write (new_, zeros, nwrite);
	}
#if 0 /* Now that we have can ask `write' to write more than a page,
	 it is legit for write do less than the whole amount specified.  */
      else if (nwrite != ret)
	{
	  sprintf (buf,
		   "unexec write failure: addr 0x%lx, fileno %d, size 0x%x, wrote 0x%x, errno %d",
		   (unsigned long) ptr, new_, nwrite, ret, errno);
	  PERROR (buf);
	}
#endif
      i += nwrite;
      ptr += nwrite;
    }
}

/* ****************************************************************
 * copy_sym
 *
 * Copy the relocation information and symbol table from the a.out to the new
 */
static int
copy_sym (int new_, int a_out, char *a_name, char *new_name)
{
  char page[1024];
  int n;

  if (a_out < 0)
    return 0;

#ifdef COFF
  if (SYMS_START == 0L)
    return 0;
#endif  /* COFF */

#ifdef COFF
  if (lnnoptr)			/* if there is line number info */
    lseek (a_out, lnnoptr, 0);	/* start copying from there */
  else
#endif /* COFF */
    lseek (a_out, SYMS_START, 0);	/* Position a.out to symtab. */

  while ((n = read (a_out, page, sizeof (page))) > 0)
    {
      if (write (new_, page, n) != n)
	{
	  PERROR (new_name);
	}
    }
  if (n < 0)
    {
      PERROR (a_name);
    }
  return 0;
}

/* ****************************************************************
 * mark_x
 *
 * After successfully building the new a.out, mark it executable
 */
static void
mark_x (char *name)
{
  struct stat sbuf;
  int um;
  int new_ = 0;  /* for PERROR */

  um = umask (777);
  umask (um);
  if (stat (name, &sbuf) == -1)
    {
      PERROR (name);
    }
  sbuf.st_mode |= 0111 & ~um;
  if (chmod (name, sbuf.st_mode) == -1)
    PERROR (name);
}

#ifdef COFF
/*
 *	If the COFF file contains a symbol table and a line number section,
 *	then any auxiliary entries that have values for x_lnnoptr must
 *	be adjusted by the amount that the line number section has moved
 *	in the file (bias computed in make_hdr).  The #@$%&* designers of
 *	the auxiliary entry structures used the absolute file offsets for
 *	the line number entry rather than an offset from the start of the
 *	line number section!
 *
 *	When I figure out how to scan through the symbol table and pick out
 *	the auxiliary entries that need adjustment, this routine will
 *	be fixed.  As it is now, all such entries are wrong and sdb
 *	will complain.   Fred Fish, UniSoft Systems Inc.
 */

/* This function is probably very slow.  Instead of reopening the new
   file for input and output it should copy from the old to the new
   using the two descriptors already open (WRITEDESC and READDESC).
   Instead of reading one small structure at a time it should use
   a reasonable size buffer.  But I don't have time to work on such
   things, so I am installing it as submitted to me.  -- RMS.  */

int
adjust_lnnoptrs (writedesc, readdesc, new_name)
     int UNUSED (writedesc);
     int UNUSED (readdesc);
     char *new_name;
{
  int nsyms;
  int new_;
#if defined defined (pfa)
  SYMENT symentry;
  AUXENT auxentry;
#else
  struct syment symentry;
  union auxent auxentry;
#endif

  if (!lnnoptr || !f_hdr.f_symptr)
    return 0;

  if ((new_ = open (new_name, O_RDWR)) < 0)
    {
      PERROR (new_name);
      return -1;
    }

  lseek (new_, f_hdr.f_symptr, 0);
  for (nsyms = 0; nsyms < f_hdr.f_nsyms; nsyms++)
    {
      read (new_, &symentry, SYMESZ);
      if (symentry.n_numaux)
	{
	  read (new_, &auxentry, AUXESZ);
	  nsyms++;
	  if (ISFCN (symentry.n_type) || symentry.n_type == 0x2400)
	    {
	      auxentry.x_sym.x_fcnary.x_fcn.x_lnnoptr += bias;
	      lseek (new_, -AUXESZ, 1);
	      write (new_, &auxentry, AUXESZ);
	    }
	}
    }
  close (new_);
  return 0;
}

#endif /* COFF */