scsh-0.6/doc/install.txt

160 lines
5.8 KiB
Plaintext

-*- Mode: Indented-text; -*-
Here are some remarks to complement what's in the INSTALL file.
-----
When running "make", don't worry if the ".notify" target fails. Its
only purpose is to send an email message to
scheme-48-notifications@martigny.ai.mit.edu, so that we can get a
rough idea of how much Scheme 48 is being used and by whom. We
promise not to use your name or email address for any commercial
purpose. If you don't want us to know, just do "make -t .notify"
first.
-----
Customizing the installation
1. If you don't believe in configure scripts, or don't have a
/bin/sh that can handle the configure script, you can make
sysdep.h and Makefile manually from sysdep.h.in and Makefile.in.
The technique is fairly obvious. For Makefile, just give
reasonable values for all of the variables at the top that are
defined as "foo = @foo@", e.g. srcdir=., CC=cc, LIBS=-lm,
INSTALL=cp, etc. For sysdep.h, read the comments. If your OS is
Posix compliant, you should be able to copy sysdep.h.in to
sysdep.h unmodified and everything should work.
2. If you definitely won't be installing Scheme 48, you should set
libdir to the distribution directory (e.g. "make libdir=`pwd`").
This will make the ,open and ,load-package commands work for the
library packages defined in more-packages.scm.
3. If desired, customize the contents of the development environment
heap image by editing the definitions of USUAL-COMMANDS and/or
USUAL-FEATURES in more-packages.scm; see below.
4. If you're using a DEC MIPS, and want to use the foreign function
interface, specify LDFLAGS=-N (with e.g. "make LDFLAGS=-N").
-----
Customizing scheme48.image
By default, the image consists of a core Scheme system (Revised^5
Scheme plus a very minimal read-eval-print loop) together with a
standard set of "options" (command processor, debugging commands,
inspector, disassembler, generic arithmetic). The set of options is
controlled by the definitions of USUAL-COMMANDS and USUAL-FEATURES in
more-packages.scm. If you make the (open ...) clause empty, then
"make scheme48.image" will create a Scheme system without any extras
(such as error recovery), and the image will be smaller. The files
are listed in approximate order of decreasing desirability; you'll
probably want at least these:
package-commands, build
- necessary for the scheme48.image script to work
debuginfo, disclosers
- necessary if you want error messages to be at all helpful
debugging
- defines important debugging commands such as ,preview and ,trace
After editing the definition of usual-features, simply
make scheme48.image
to rebuild the image.
-----
Deeper changes to the system -- for example, edits to most of the
files in the rts/ directory -- will require using the static linker to
make a new initial.image. After you have a working scheme48.image
(perhaps a previous version of Scheme 48), you can create a linker
image with
make linker
after which you can say
make image
to get the linker to build a new initial.image and initial.debug.
scheme48.image will then be built from those.
You might think that "make scheme48.image" ought to do this, but the
circular dependencies
scheme48.image on initial.image
initial.image on link/linker.image
link/linker.image on scheme48.image
needs to be broken somewhere, or else make will (justifiably) barf. I
chose to break the cycle by making scheme48.image not depend on
initial.image, since this is most robust for installation purposes.
-----
Editor support
We recommend interacting with the Scheme 48 command processor using the
emacs/scheme interface written by Olin Shivers at CMU. Copies of the
relevant .el files, together with a "cmuscheme48.el", are in the
emacs/ subdirectory of the release. Usage information is in
doc/user-guide.txt.
You will probably want to byte-compile the .el files to get .elc
files. Use M-x byte-compile-file to do this.
-----
Performance
If you don't have a C compiler that optimizes as well as gcc does,
then performance may suffer. Take a look at the automatically
generated code in scheme48vm.c to find out why. With a good register
allocator, all those variables (including some of the virtual
machine's virtual registers) get allocated to hardware registers, and
it really flies. Without one, performance can be pretty bad.
The configure script automatically sets the Makefile variable CFLAGS
to -O2 -g if gcc is available, or to -O if it isn't. This can be
overriden by specifying a different CFLAGS, e.g. "make CFLAGS=-g" for
no optimization.
Even if you do have a good compiler, you should be able to improve
overall performance even more, maybe about 6-10%, by removing the
range check from the interpreter's instruction dispatch. To do this,
use the -S flag to get assembly code for scheme48vm.c, then find the
instructions in scheme48vm.s corresponding to the big dispatch in
restart():
L19173: {
code_pointer_83X = arg1K0;
switch ((*((unsigned char *) code_pointer_83X))) {
... }
There will be one or two comparison instructions to see whether the
opcode is in range; just remove them. For the 68000 I use a "sed"
script
/cmpl #137,d0/ N
/cmpl #137,d0\n jhi L/ d
but of course the constant will probably have to change when a new
release comes along.
See the user's guide for information on the ,bench command, which
makes programs run faster.
-----
filenames.make is "include"d by the Makefile, but is automatically
generated from the module dependencies laid out in the various
configuration files (*-packages.scm). If you edit any of these .scm
files, you may want to do a "make filenames.make" before you do any
further "make"s in order to update the depedencies. This step isn't
necessary if you're using Gnu make, because Gnu make will make
included files automatically.