Mercurial > hg > xemacs-beta
annotate dynodump/README @ 4677:8f1ee2d15784
Support full Common Lisp multiple values in C.
lisp/ChangeLog
2009-08-11 Aidan Kehoe <kehoea@parhasard.net>
* bytecomp.el :
Update this file to support full C-level multiple values. This
involves:
-- Four new bytecodes, and special compiler functions to compile
multiple-value-call, multiple-value-list-internal, values,
values-list, and, since it now needs to pass back multiple values
and is a special form, throw.
-- There's a new compiler variable, byte-compile-checks-on-load,
which is a list of forms that are evaluated at the very start of a
file, with an error thrown if any of them give nil.
-- The header is now inserted *after* compilation, giving a chance
for the compilation process to influence what those checks
are. There is still a check done before compilation for non-ASCII
characters, to try to turn off dynamic docstrings if appopriate,
in `byte-compile-maybe-reset-coding'.
Space is reserved for checks; comments describing the version of
the byte compiler generating the file are inserted if space
remains for them.
* bytecomp.el (byte-compile-version):
Update this, we're a newer version of the byte compiler.
* byte-optimize.el (byte-optimize-funcall):
Correct a comment.
* bytecomp.el (byte-compile-lapcode):
Discard the arg with byte-multiple-value-call.
* bytecomp.el (byte-compile-checks-and-comments-space):
New variable, describe how many octets to reserve for checks at
the start of byte-compiled files.
* cl-compat.el:
Remove the fake multiple-value implementation. Have the functions
that use it use the real multiple-value implementation instead.
* cl-macs.el (cl-block-wrapper, cl-block-throw):
Revise the byte-compile properties of these symbols to work now
we've made throw into a special form; keep the byte-compile
properties as anonymous lambdas, since we don't have docstrings
for them.
* cl-macs.el (multiple-value-bind, multiple-value-setq)
(multiple-value-list, nth-value):
Update these functions to work with the C support for multiple
values.
* cl-macs.el (values):
Modify the setf handler for this to call
#'multiple-value-list-internal appropriately.
* cl-macs.el (cl-setf-do-store):
If the store form is a cons, treat it specially as wrapping the
store value.
* cl.el (cl-block-wrapper):
Make this an alias of #'and, not #'identity, since it needs to
pass back multiple values.
* cl.el (multiple-value-apply):
We no longer support this, mark it obsolete.
* lisp-mode.el (eval-interactive-verbose):
Remove a useless space in the docstring.
* lisp-mode.el (eval-interactive):
Update this function and its docstring. It now passes back a list,
basically wrapping any eval calls with multiple-value-list. This
allows multiple values to be printed by default in *scratch*.
* lisp-mode.el (prin1-list-as-multiple-values):
New function, printing a list as multiple values in the manner of
Bruno Haible's clisp, separating each entry with " ;\n".
* lisp-mode.el (eval-last-sexp):
Call #'prin1-list-as-multiple-values on the return value of
#'eval-interactive.
* lisp-mode.el (eval-defun):
Call #'prin1-list-as-multiple-values on the return value of
#'eval-interactive.
* mouse.el (mouse-eval-sexp):
Deal with lists corresponding to multiple values from
#'eval-interactive. Call #'cl-prettyprint, which is always
available, instead of sometimes calling #'pprint and sometimes
falling back to prin1.
* obsolete.el (obsolete-throw):
New function, called from eval.c when #'funcall encounters an
attempt to call #'throw (now a special form) as a function. Only
needed for compatibility with 21.4 byte-code.
man/ChangeLog addition:
2009-08-11 Aidan Kehoe <kehoea@parhasard.net>
* cl.texi (Organization):
Remove references to the obsolete multiple-value emulating code.
src/ChangeLog addition:
2009-08-11 Aidan Kehoe <kehoea@parhasard.net>
* bytecode.c (enum Opcode /* Byte codes */):
Add four new bytecodes, to deal with multiple values.
(POP_WITH_MULTIPLE_VALUES): New macro.
(POP): Modify this macro to ignore multiple values.
(DISCARD_PRESERVING_MULTIPLE_VALUES): New macro.
(DISCARD): Modify this macro to ignore multiple values.
(TOP_WITH_MULTIPLE_VALUES): New macro.
(TOP_ADDRESS): New macro.
(TOP): Modify this macro to ignore multiple values.
(TOP_LVALUE): New macro.
(Bcall): Ignore multiple values where appropriate.
(Breturn): Pass back multiple values.
(Bdup): Preserve multiple values.
Use TOP_LVALUE with most bytecodes that assign anything to
anything.
(Bbind_multiple_value_limits, Bmultiple_value_call,
Bmultiple_value_list_internal, Bthrow): Implement the new
bytecodes.
(Bgotoifnilelsepop, Bgotoifnonnilelsepop, BRgotoifnilelsepop,
BRgotoifnonnilelsepop):
Discard any multiple values.
* callint.c (Fcall_interactively):
Ignore multiple values when calling #'eval, in two places.
* device-x.c (x_IO_error_handler):
* macros.c (pop_kbd_macro_event):
* eval.c (Fsignal):
* eval.c (flagged_a_squirmer):
Call throw_or_bomb_out, not Fthrow, now that the latter is a
special form.
* eval.c:
Make Qthrow, Qobsolete_throw available as symbols.
Provide multiple_value_current_limit, multiple-values-limit (the
latter as specified by Common Lisp.
* eval.c (For):
Ignore multiple values when comparing with Qnil, but pass any
multiple values back for the last arg.
* eval.c (Fand):
Ditto.
* eval.c (Fif):
Ignore multiple values when examining the result of the
condition.
* eval.c (Fcond):
Ignore multiple values when comparing what the clauses give, but
pass them back if a clause gave non-nil.
* eval.c (Fprog2):
Never pass back multiple values.
* eval.c (FletX, Flet):
Ignore multiple when evaluating what exactly symbols should be
bound to.
* eval.c (Fwhile):
Ignore multiple values when evaluating the test.
* eval.c (Fsetq, Fdefvar, Fdefconst):
Ignore multiple values.
* eval.c (Fthrow):
Declare this as a special form; ignore multiple values for TAG,
preserve them for VALUE.
* eval.c (throw_or_bomb_out):
Make this available to other files, now Fthrow is a special form.
* eval.c (Feval):
Ignore multiple values when calling a compiled function, a
non-special-form subr, or a lambda expression.
* eval.c (Ffuncall):
If we attempt to call #'throw (now a special form) as a function,
don't error, call #'obsolete-throw instead.
* eval.c (make_multiple_value, multiple_value_aset)
(multiple_value_aref, print_multiple_value, mark_multiple_value)
(size_multiple_value):
Implement the multiple_value type. Add a long comment describing
our implementation.
* eval.c (bind_multiple_value_limits):
New function, used by the bytecode and by #'multiple-value-call,
#'multiple-value-list-internal.
* eval.c (multiple_value_call):
New function, used by the bytecode and #'multiple-value-call.
* eval.c (Fmultiple_value_call):
New special form.
* eval.c (multiple_value_list_internal):
New function, used by the byte code and
#'multiple-value-list-internal.
* eval.c (Fmultiple_value_list_internal, Fmultiple_value_prog1):
New special forms.
* eval.c (Fvalues, Fvalues_list):
New Lisp functions.
* eval.c (values2):
New function, for C code returning multiple values.
* eval.c (syms_of_eval):
Make our new Lisp functions and symbols available.
* eval.c (multiple-values-limit):
Make this available to Lisp.
* event-msw.c (dde_eval_string):
* event-stream.c (execute_help_form):
* glade.c (connector):
* glyphs-widget.c (glyph_instantiator_to_glyph):
* glyphs.c (evaluate_xpm_color_symbols):
* gui-x.c (wv_set_evalable_slot, button_item_to_widget_value):
* gui.c (gui_item_value, gui_item_display_flush_left):
* lread.c (check_if_suppressed):
* menubar-gtk.c (menu_convert, menu_descriptor_to_widget_1):
* menubar-msw.c (populate_menu_add_item):
* print.c (Fwith_output_to_temp_buffer):
* symbols.c (Fsetq_default):
Ignore multiple values when calling Feval.
* symeval.h:
Add the header declarations necessary for the multiple-values
implementation.
* inline.c:
#include symeval.h, now that it has some inline functions.
* lisp.h:
Update Fthrow's declaration. Make throw_or_bomb_out available to
all files.
* lrecord.h (enum lrecord_type):
Add the multiple_value type here.
author | Aidan Kehoe <kehoea@parhasard.net> |
---|---|
date | Sun, 16 Aug 2009 20:55:49 +0100 |
parents | 376386a54a3c |
children |
rev | line source |
---|---|
0 | 1 DYNODUMP |
2 -------- | |
3 | |
4 Dynodump, not to be confused with DinoTurd (as seen in Jurassic Park), | |
5 is a shared object that provides one function: | |
6 | |
7 int dynodump(char *new_file); | |
8 | |
9 dynodump(), called from a running program will write a new executable | |
10 in new_file a la unexec() in GNU Emacs. The difference lies in the | |
11 relocations. | |
12 | |
13 dynodump() will create an image with any relocations (which were | |
14 performed by the run-time dynamic linker) undone. This allows the new | |
15 image to be run in a different environment. There is, however, one | |
16 potentially major caveat. If a symbol reference gets updated during | |
17 the running of the calling program, its updated value will be lost. | |
18 An example (with additional blank lines for legibility): | |
19 | |
20 $ cat lib.c | |
21 char _foo[] = "hello"; | |
22 char _bar[] = "world"; | |
23 | |
24 $ cc -G -o lib.so lib.c | |
25 | |
26 $ cat prog.c | |
27 extern char _foo, _bar; | |
28 | |
29 int beenhere = 0; | |
30 char * foo = &_foo; | |
31 char * bar = &_bar; | |
32 | |
33 int | |
34 main(void) | |
35 { | |
36 (void) printf("%d: foo = %x\n", beenhere, foo); | |
37 (void) printf("%d: bar = %x, ", beenhere, bar); | |
38 | |
39 if (!beenhere) { | |
40 beenhere = 1; | |
41 bar++; | |
42 dynodump("newfile"); | |
43 } | |
44 (void) printf("%x\n", bar); | |
45 } | |
46 | |
47 $ cc -o prog prog.c -R. lib.so dynodump.so | |
48 | |
49 $ ./prog | |
50 0: foo = ef7503cc | |
51 0: bar = ef7503d2, ef7503d3 | |
52 | |
53 $ ./newfile | |
54 1: foo = ef7503cc | |
55 1: bar = ef7503d2, ef7503d2 | |
56 | |
57 Notice that in the run of newfile, bar points at "world" instead of | |
58 the perhaps expected "orld". | |
59 | |
60 Dynodump supports sparc, intel, and power pc architectures. | |
61 | |
62 Dynodump is buildable with GNU make and gcc. If it works for you | |
63 with these tools, let me know. | |
64 | |
65 unexec() | |
66 -------- | |
67 | |
68 Also supplied is an unexsol2.c which belongs in the XEmacs src | |
69 directory with the other unex*.c files. The src/s/sol2.h should be | |
70 have the following added to it: | |
71 | |
72 #ifdef UNEXEC | |
73 #undef UNEXEC | |
74 #endif | |
75 #define UNEXEC unexsol2.o | |
76 | |
77 This unexec() will attempt to dlopen() the dynodump.so to find the | |
78 dynodump function. The default is "../dynodump/dynodump.so" which | |
79 should be appropriate for the typical XEmacs build (unless you used | |
80 configure to set up separate build and source trees). You may change | |
81 it by setting the DYNODUMP environment variable to the full path of | |
82 the dynodump shared object. | |
83 | |
84 Other notes: | |
85 | |
86 If you're using the 4.0 C compiler from Sun, you should arrange to | |
87 shut off the incremental linker. This is done with the -xildoff | |
88 option to the compiler. The common.mk and testsuite/Makefile files | |
89 have commentary and ready made lines for uncommenting for this | |
90 purpose. | |
91 | |
92 If you're interested in playing with the UltraSPARC specific options | |
93 to the aforementioned compiler, then the same makefiles have some | |
94 commentary and flags settings to point you in that direction. | |
95 | |
96 Questions: | |
97 | |
98 dynodump() was developed by Rod.Evans@Eng.Sun.COM and | |
99 Georg.Nikodym@Canada.Sun.COM. If you have questions, feel free to ask | |
100 them but be aware that Rod, "don't know jack about emacs." | |
101 |