By Doug Bates, Ed Kademan, Frank Ritter and David Smith
This file documents S.el, a GNU Emacs mode for running
Splus in a buffer.
This documentation relates to Version 4.8 of S.el; however much
of it is out-of-date and the rest of it isn't even complete.
Info Author: David M. Smith (D.M.Smith@lancaster.ac.uk), Department of Mathematics and Statistics, Lancaster University, UK.
Info version: 1.7
Please note: This manual is still under development and has not yet been updated for version 4.8. Revisions are very welcome!
@sp5
S-mode
version 4.8
Doug Bates, Ed Kademan, Frank Ritter and David Smith
@sp2
An GNU Emacs package
for interacting with the
S/Splus statistical software packages
@sp7
Documentation by David Smith
(D.M.Smith@lancaster.ac.uk)
Department of Mathematic and Statistics
Lancaster University, UK
@sp7
Documentation version: 1.7
Copyright (C) 1992, 1993, 1994, 1995 David M. Smith
Permission is granted to make and distribute verbatim copies of this manual provided the copyright notice and this permission notice are preserved on all copies.
Permission is granted to copy and distribute modified versions of this manual under the conditions for verbatim copying, provided that the entire resulting derived work is distributed under the terms of a permission notice identical to this one.
The S and Splus packages provide sophisticated statistical and graphical routines for manipulating data. The S-mode package provides useful routines for making the use of these packages much easier.
A bit of notation before we begin. I will refer to both the `new S'
package (as described in Becker, Chambers and Wilks, The New S
Language: A programming environment for data analysis and graphics) and
`Splus' (an enhanced version of new S from Statsci) simply by
"S". The interface which is used to run S under Emacs (which this manual
documents) will be referred to as "S-mode", which should not be
confused with the GNU Emacs major mode S-mode which is used for
editing S source.
For exclusively interactive users of S, S-mode provides a number of features to make life easier. There is an easy-to-use command history mechanism, including a quick prefix-search history. To reduce typing, command-line completion is provided for all S objects and "hot keys" are provided for common S function calls. Help files are easily accessible, and a paging mechanism is provided to view them. Finally, an incidental (but very useful) side-effect of S-mode is that a transcript of your session is kept for later saving or editing. No special knowledge of Emacs is necessary when using S interactively under S-mode.
For those that use S in the typical edit-test-revise cycle when programming S functions, S-mode provides for editing of S functions in Emacs edit buffers. Unlike the typical use of S where the editor is restarted every time an object is edited, S-mode uses the current Emacs session for editing. In practical terms, this means that you can edit more than one function at once, and that the S process is still available for use while editing. Error checking is performed on functions loaded back into S, and a mechanism to jump directly to the error is provided. S-mode also provides for maintaining text versions of your S functions in specified source directories.
S is a powerful system for manipulating and analysing data, but its user interface -- particularly on Unix platforms -- leaves something to be desired. S-mode is designed as a package that makes S easier to use.
S-mode provides several features which make it easier to interact with the S process (i.e. enter commands and view the output). These include:
tcsh's facility for filenames; here it also
applies to object names and list components. See section Completion of object names.
objects() and search(). See section Hot keys for common commands.
If, like me, you commonly create or modify S functions, you will have found the standard facilities for this (the `fix()' function, for example) severely limiting. Using S's standard features, you can only edit one function at a time, and you can't continue to use S while editing. S-mode corrects these problems by introducing the following features:
Finally, S-mode provides features for re-submitting commands from saved transcript files, including:
*** Not yet written ***
S-mode is based on Olin Shivers' excellent comint package (which is supplied
with version 19 of FSF GNU Emacs). The original version of S-mode was
written by Doug Bates (bates@stat.wisc.edu) and Ed Kademan
(kademan@stat.wisc.edu). Frank Ritter
(ritter@psy.cmu.edu) then merged this version with his own S
mode to form S.el version 2.1.
Version 2.1 of S.el was then updated and expanded by David Smith to form version 3.4. This was then updated for Emacs 19 to create version 4. Most bugs have now been fixed (and several new ones introduced) and many new features have been added. Thanks must go to the many people who have helped with the development of the present version of S-mode:
S-eval-line-and-next-line are by Rod Ball.
The latest version is always available via WWW from:
http://www.maths.lancs.ac.uk:2080/~maa036/elisp/S-mode/
Recent versions of S-mode are also available for anonymous FTP from the following sites:
/anonymous@wingra.stat.wisc.edu:pub/src/emacs-lisp
/anonymous@attunga.stats.adelaide.edu.au:pub/S-mode
Check the README file first to see which files you need. S-mode
is also available from the Emacs-Lisp archive on
archive.cis.ohio-state-edu -- retrieve
`pub/gnu/emacs/elisp-archive/README'
for information on the
archive. An older version is also available from Statlib by sending a
blank message with subject "send index from S" to
statlib@stat.cmu.edu, and following the directions from there.
Note that all new user-visible features to versions of S-mode are documented in the `NEWS' file; all changes are listed in the `ChangeLog' file.
If S-mode has already been installed on your system, the next chapter has details on how to get started using S under S-mode.
If you need to install S-mode, read section Installing S-mode on your system for details on what needs to be done before proceeding to the next chapter.
section Customizing S-mode provides details of user variables you can change to customize S-mode to your taste, but it is recommended that you defer this section until you are more familiar with S-mode.
Don't forget that this manual is not the only source of information about S-mode. In particular, the mode-based online help (obtained by pressing C-h m when in the process buffer, edit buffer or help buffer) is quite useful. However the best source of information is, as always, experience -- try it out!
To start an S session, simply type M-x S RET, i.e. press ESC, then x, then capital S and then the RETURN key.
S will then (by default) ask the question
S starting data directory?
Enter the name of the directory you wish to start S from (that is, the
directory you would have cd'd to before starting S from the
shell). This directory should have a `.Data' subdirectory.
You will then be popped into a buffer with name `*S*' which will be used for interacting with the S process, and you can start entering commands.
S-mode allows you to run more than one S process simultaneously in the same session. Each process has a name and a number; the initial process (process 1) is simply named `S'. You may start a new process by passing a numeric argument to M-x S. For example, typing ESC 2 M-x S starts up an S process with name `S2', in a buffer whose name is initially `*S2*'. The name of the process is shown in the mode line in square brackets (for example, `[S2]'); this is useful if the process buffer is renamed. Without a prefix argument, M-x S starts a new S process, using the first available process number.
You can switch to any active S process with the command C-c C-k
(S-request-a-process). Just enter the name of the process you
require; completion is provided over the names of all running S
processes. This is a good command to bind to a global key.
If you do not wish S-mode to prompt for a starting directory when
starting a new process, set the variable S-ask-for-S-directory to
nil. In this case, the value of the variable S-directory
is used as the starting directory. The default value for this variable
is your home directory. If S-ask-for-S-directory has a
non-nil value (as it does by default) then the value of
S-directory provides the default when prompting for the
starting directory. Incidentally, S-directory is an ideal
variable to set in S-pre-run-hook.
If you like to keep a records of your S sessions, set the variable
S-ask-about-transfile to t, and you will be asked for a
filename for the transcript before the S process starts.
nil, as for a file name in which to save the session
transcript.
Enter the name of a file in which to save the transcript at the prompt. If the file doesn't exist it will be created (and you should give it a file name ending in `.St'; if the file already exists the transcript will be appended to the file. (Note: if you don't set this variable but you still want to save the transcript, you can still do it later -- see section Keeping a record of your S session.)
Once these questions are answered (if they are asked at all) the
S process itself is started by calling the program name
specified in the variable inferior-S-program.
If you need to pass any arguments to this program, they may be specified
in the variable
inferior-S_program_name-args (e.g. if
inferior-S-program is "S+" then the variable to set is
inferior-S+-args.
It is not normally necessary to pass arguments to the S program; in
particular do not pass the `-e' option to Splus, since
S-mode provides its own command history mechanism.
The primary function of the S-mode package is to provide an easy-to-use front end to the S interpreter. This is achieved by running the S process from within an Emacs buffer, so that the Emacs editing commands are available to correct mistakes in commands, etc. The features of Inferior S mode are similiar to those provided by the standard Emacs shell mode (see section `Shell Mode' in The Gnu Emacs Reference Manual). Command-line completion of S objects and a number of `hot keys' for commonly-used S commands are also provided for ease of typing.
Sending a command to the S process is as simple as typing it in and pressing the RETURN key:
If you make a typing error before pressing RET all the usual Emacs editing commands are available to correct it (see section `Basic editing commands' in The GNU Emacs Reference Manual). Once the command has been corrected you can press RETURN (even if the cursor is not at the end of the line) to send the corrected command to the S process.
S-mode provides some other commands which are useful for fixing mistakes:
backward-kill-word) comint-kill-input) comint-bol) See section `Shell Mode' in The Gnu Emacs Reference Manual, for other commands relevant to entering input.
In the process buffer, the TAB key is for completion, similar to that provided by Shell Mode for filenames. In Inferior S mode, pressing the TAB key when the cursor is following the first few characters of an object name completes the object name; if the cursor is following a file name TAB completes the file name.
When the cursor is just after a partially-completed object name,
pressing TAB provides completion in a similar fashion to
tcsh
except that completion is performed over all known S object names
instead of file names. S-mode maintains a list of all objects known to
S at any given time, which basically consists of all objects (functions
and datasets) in every attached directory listed by the search()
command
along with the component objects of attached data frames
(if your version of S supports them).
For example, consider the three functions (available in Splus version
3.0) called binomplot(), binom.test() and
binomial(). Typing bin TAB after the S prompt will insert
the characters `om', completing the longest prefix (`binom')
which distinguishes these three commands. Pressing TAB once more
provides a list of the three commands which have this prefix, allowing
you to add more characters (say, `.') which specify the function
you desire. After entering more characters pressing TAB yet again
will complete the object name up to uniqueness, etc. If you just wish to
see what completions exist without adding any extra characters, type
M-?.
S-mode also provides completion over the components of named lists accessed using the `$' notation, to any level of nested lists. This feature is particularly useful for checking what components of a list object exist while partway through entering a command: simply type the object name and `$' and press TAB to see the names of existing list components for that object.
Completion is also provided over file names, which is particularly
useful when using S functions such as get() or scan()
which require fully expanded file names. Whenever the cursor is within
an S string, pressing TAB completes the file name before point,
and also expands any `~' or environment variable references.
If the cursor is not in a string and does not follow a (partial) object name, the TAB key has a third use: it expands history references. See section References to historical commands.
S-mode automatically keeps track of any objects added or deleted to the system (such as new objects created, or directories added to the search list) to make completion as accurate as possible. Whenever S-mode notices that search list has changed (1) when you attach a directory or data frame, the objects associated with it immediately become available for a completion; when it is detached completion is no longer available on those objects.
To maintain a list of accessible objects for completion, S-mode needs to
determine which objects are contained in each directory or data frame on
the search list. This is done at the start of each S session, by
running the objects() command on every element of the search
list. On some systems, however, this can be rather slow; it's doubly
frustrating when you consider that most of the directories on the search
list are the standard S libraries, which never change anyway! When
S-mode was installed, a database of the standard object names should
have been created which should speed up this process at the start of an
S session; if it has not been created you will get a warning like
`S-namedb.el does not exist'. See section Installing S-mode on your system, for information on
how to create this database.
Efficiency in completion is gained by maintaining a cache of objects currently known to S; when a new object becomes available or is deleted, only one component of the cache corresponding to the associated directory needs to be refreshed. If S-mode ever becomes confused about what objects are available for completion (such as when if refuses to complete an object you know is there), the command M-x S-resynch forces the entire cache to be refreshed, which should fix the problem.
Most of the time, the cursor spends most of its time at the bottom of the S process buffer, entering commands. However all the input and output from the current (and previous) S sessions is stored in the process buffer (we call this the transcript) and often we want to move back up through the buffer, to look at the output from previous commands for example.
Within the process buffer, a paragraph
is defined as the prompt, the command after the prompt, and the output
from the command. Thus M-{ and M-} move you backwards and
forwards, respectively, through commands in the transcript. A
particularly useful command is M-h (mark-paragraph) which
will allow you to mark a command and its entire output (for deletion,
perhaps). For more information about paragraph commands,
see section `Paragraphs' in The GNU Emacs Reference Manual.
If an S process finishes and you restart it in the same process buffer, the output from the new S process appears after the output from the first S process separated by a form-feed (`^L') character. Thus pages in the S process buffer correspond to S sessions. Thus, for example, you may use C-x [ and C-x ] to move backward and forwards through S sessions in a single S process buffer. For more information about page commands, see section `Pages' in The GNU Emacs Reference Manual.
Viewing the output of the command you have just entered is a common occurence and S-mode provides a number of facilities for doing this. Whenever a command produces a longish output, it is possible that the window will scroll, leaving the next prompt near the middle of the window. The first part of the command output may have scrolled off the top of the window, even though the entire output would fit in the window if the prompt were near the bottom of the window. If this happens, you can use the command
comint-show-maximum-output)
to make more of the last output visible. (To make this happen
automatically for all inputs, set the variable
comint-scroll-to-bottom-on-input to t; for information on
this and other options for handling process input and output
see section `Shell Mode Options' in The GNU Emacs Reference Manual.)
If the first part of the output is still obscured, use
comint-show-output) to view it. Finally, if you want to discard the last command output altogether, use
to delete it. Use this command judiciously to keep your transcript to a more manageable size.
If you want to view the output from more historic commands than the previous command, commands are also provided to move backwards and forwards through previously entered commands in the process buffer:
comint-previous-input) comint-next-input) Note that these two commands are analagous to C-p and C-n but apply to command lines rather than text lines. And just like C-p and C-n, passing a prefix arg to these commands means to move to the ARG'th next (or previous) command. (These commands are also discussed in section `Shell History Copying' in The GNU Emacs Reference Manual.)
There are also two similar commands (not bound to any keys by default) which move to preceding or succeeding commands, but which first prompt for a regular expression (see section `Syntax of Regular Expression' in The GNU Emacs Reference Manual), and then moves to the next (previous) command matching the pattern.
When moving through the transcript, you may wish to re-execute some of the commands you find there. S-mode provides three commands to do this; these commands may be used whenever the cursor is within a command line in the transcript (if the cursor is within some command output, an error is signalled). Note all three commands involve the RETURN key.
inferior-S-send-input) comint-copy-old-input) S-transcript-send-command-and-move) When the cursor is not after the current prompt, the RETURN key has a slightly different behaviour than usual. Pressing RET on any line containing a command that you entered (i.e. a line beginning with a prompt) sends that command to the S process once again. If you wish to edit the command before executing it, use C-c RET instead; it copies the command to the current propt but does not execute it, allowing you to edit it before submitting it.
These two commands leave the cursor at the new command line, allowing you to continue with interactive use of S. If you wish to resubmit a series of commands from the transcript, consider using M-RET instead, which leaves the cursor at the command line following the one you re-submitted. Thus by using M-RET repeatedly, you can re-submit a whole series of commands.
These commands work even if if the current line is a continuation line (i.e. the prompt is `+' instead of `>') -- in this case all the lines that form the multi-line command are concatenated together and the resulting command is sent to the S process (currently this is the only way to resubmit a multi-line command to the S process in one go). If the current line does not begin with a prompt, an error is signalled. This feature, coupled with the command-based motion commands described above, could be used as a primitive history mechanism. S-mode provides a more sophisticated mechanism, however, which is described in section Command History.
To keep a record of your S session in a disk file, use the Emacs command
C-x C-w (write-file) to attach a file to the S process
buffer. The name of the process buffer will (probably) change to the
name of the file, but this is not a problem. You can still use S as
usual; just remember to save the file before you quit Emacs with
C-x C-s. You can make S-mode prompt you for a filename in which
to save the transcript every time you start S by setting the variable
S-ask-about-transfile to t; see section Changing the startup actions.
We recommend you save your transcripts with filenames that end in
`.St'. There is a special mode (S transcript mode ---
see section Manipulating saved transcript files) for editing transcript files which is
automatically selected for files with this suffix.
S transcripts can get very large, so some judicious editing is appropriate if you are saving it in a file. Use C-c C-o whenever a command produces excessively long output (printing large arrays, for example). Delete erroneous commands (and the resulting error messages or other output) by moving to the command (or its output) and typing M-h C-w. Also, remember that C-c C-e (and other hot keys) may be used for commands whose output you do not wish to appear in the transcript. These suggestions are appropriate even if you are not saving your transcript to disk, since the larger the transcript, the more memory your Emacs process will use on the host machine.
Finally, if it is your intention to produce S source code (suitable for
using with source() or inclusion in an S function) from a
transcript, then the command M-x S-clean-region may be of use.
This command works in any Emacs buffer, and removes all prompts and
command output from an S transcript within the current region, leaving
only the commands. Don't forget to remove any erroneous commands first!
S-mode provides easy-to-use facilities for re-executing or editing
previous commands. An input history of the last few commands is
maintained (by default the last 50 commands are stored, although this
can be changed by setting the variable comint-input-ring-size in
inferior-S-mode-hook.) The simplest history commands simply
select the next and previous commands in the input history:
comint-previous-input) comint-next-input) For example, pressing M-p once will re-enter the last command into the process buffer after the prompt but does not send it to the S process, thus allowing editing or correction of the command before the S process sees it. Once corrections have been made, press RET to send the edited command to the S process.
If you want to select a particular command from the history by matching it against a regular expression (see section `Syntax of Regular Expression' in The GNU Emacs Reference Manual), to search for a particular variable name for example, these commands are also available:
comint-previous-matching-input) comint-next-matching-input) A common type of search is to find the last command that began with a particular sequence of characters; the following two commands provide an easy way to do this:
comint-previous-matching-input-from-input) comint-next-matching-input-from-input)
Instead of prompting for a regular expression to match against, as they
instead select commands starting with those characters already entered.
For instance, if you wanted to re-execute the last attach()
command, you may only need to type att and then A-M-r and
RET. (Note: you may not have an ALT key on your keyboard,
in which case it may be a good idea to bind these commands to some other
keys.)
See section `Shell History Ring' in The GNU Emacs Reference Manual, for a more detailed discussion of the history mechanism.
Instead of searching through the command history using the command
described in the previous section, you can alternatively refer to a
historical command directly using a notation very similar to that used
in csh. History references are introduced by a `!' or
`^' character and have meanings as follows:
In addition, you may follow the reference with a word designator
to select particular words of the input. A word is defined as a
sequence of characters separated by whitespace. (You can modify this
definition by setting the value of comint-delimiter-argument-list
to a list of characters that are allowed to separate words and
themselves form words.) Words are numbered beginning with zero. The
word designator usually begins with a `:' (colon) character;
however it may be omitted if the word reference begins with a `^',
`$', `*' or `-'. If the word is to be selected from the
previous command, the second `!' character can be omitted from the
event specification. For instance, `!!:1' and `!:1' both
refer to the first word of the previous command, while `!!$' and
`!$' both refer to the last word in the previous command. The
format of word designators is as follows:
In addition, you may surround the entire reference except for the first `!' by braces to allow it to be followed by other (non-whitespace) characters (which will be appended to the expanded reference).
Finally, S-mode also provides quick substitution; a reference like `^old^new^' means "the last command, but with the first occurrence of the string `old' replaced with the string `new'" (the last `^' is optional). Similarly, `^old^' means "the last command, with the first occurrence of the string `old' deleted" (again, the last `^' is optional).
To convert a history reference as described above to an input suitable for S, you need to expand the history reference, using the TAB key. For this to work, the cursor must be preceeded by a space (otherwise it would try to complete an object name) and not be within a string (otherwise it would try to complete a filename). So to expand the history reference, type SPC TAB. This will convert the history reference into an S command from the history, which you can then edit or press RET to execute.
For example, to execute the last command that referenced the variable
data, type !?data SPC TAB RET.
S-mode provides a number of commands for executing the commonly used
functions. These commands below are basically information-gaining
commands (such as objects() or search()) which tend to
clutter up your transcript and for this reason some of the hot keys
display their output in a temporary buffer
instead of the process buffer by default. This behaviour is controlled
by the variable S-execute-in-process-buffer which, if
non-nil, means that these commands will produce their output in
the process buffer instead. In any case, passing a prefix argument to
the commands (with C-u) will reverse the meaning of
S-execute-in-process-buffer for that command, i.e. the output
will be displayed in the process buffer if it usually goes to a
temporary buffer, and vice-versa. These are the hot keys that behave in
this way:
S-execute-objects) objects()
command to the S process. A prefix argument specifies the position on
the search list (use a negative argument to toggle
S-execute-in-process-buffer as well).
A quick way to see what objects are in your working
directory.
S-execute-search) search()
command to the S process.
S-execute)
S-execute may seem pointless when you could just type the command
in anyway, but it proves useful for `spot' calculations which would
otherwise clutter your transcript, or for evaluating an expression while
partway through entering a command. You can also use this command to
generate new hot keys using the Emacs keyboard macro facilities;
see section `Keyboard Macros' in The GNU Emacs Reference Manual.
The following hot keys do not use S-execute-in-process-buffer to
decide where to display the output -- they either always display in
the process buffer or in a separate buffer, as indicated:
S-execute-attach) attach() command.
If a numeric prefix argument is given it is used as the position on the
search list to attach the directory; otherwise the S default of 2 is
used. The attach() command actually executed appears in the
process buffer.
S-load-file) source(). If
there is an error during loading, you can jump to the error in the file
with C-x ` (S-parse-errors).
See section Detecting errors in source files for more details.
S-display-help-on-object) S-quit) q()
command to the S process, and cleans up any temporary buffers (such as
help buffers or edit buffers) you may have created along the way. Use
this command when you have finished your S session instead of simply
typing q() yourself, otherwise you will need to issue the command
M-x S-cleanup
command explicitly to make sure that all the files that need to be saved
have been saved, and that all the temporary buffers have been killed.
The following commands are also provided in the process buffer:
comint-interrupt-subjob) S-abort) q() nor .Last
will be executed and device drivers will not finish cleanly. This
command is provided as a safety to comint-stop-subjob, which is
usually bound to C-c C-z. If you want to quit from S, use C-c
C-q (S-quit) instead.
S-dump-object-into-edit-buffer) Other commands available is Inferior S mode are discussed in section `Shell Mode' in The Gnu Emacs Reference Manual.
S-mode provides facilities for editing S objects within your Emacs session. Most editing is performed on S functions, although in theory you may edit datasets as well. Edit buffers are always associated with files, although you may choose to make these files temporary if you wish. Alternatively, you may make use of a simple yet powerful mechanism for maintaining backups of text representations of S functions. Error-checking is performed when S code is loaded into the S process.
To edit an S object, type
from within the S process buffer (*S*).
You will then be prompted for an object to edit: you may either type in
the name of an existing object (for which completion is available using
the TAB key),
or you may enter the name of a new object.
A buffer will be created containing the text representation of the
requested object or, if you entered the name of a non-existent object at
the prompt and the variable S-insert-function-templates
is non-nil, you will be presented with a template defined by
S-function-template
which defaults to a skeleton function construct.
You may then edit the function as required.
The edit buffer generated by S-dump-object-into-edit-buffer is placed
in the S-mode major mode which provides a number of commands to
facilitate editing S source code. Commands are provided to intelligently
indent S code, evaluate portions of S code and to move around S code
constructs.
Note: when you dump a file with C-c C-d, S-mode first
checks to see whether there already exists an edit buffer containing
that object and, if so, pops you directly to that buffer. If not, S-mode
next checks whether there is a file in the appropriate place with the
appropriate name (See section Maintaining S source files) and if so, reads in that file.
You can use this facility to return to an object you were editing in a
previous session (and which possibly was never loaded to the S session).
Finally, if both these tests fail, the S process is consulted and a
dump() command issued.
If you want to force S-mode to ask the S process for the object's
definition (say, to reformat an unmodified buffer or to revert back to
S's idea of the object's definition) pass a prefix argument to
S-dump-object-into-edit-buffer by typing C-u C-c C-d.
The best way to get information -- particularly function definitions
--- into S is to load them in as source file, using S's source
function. You have already seen how to create source files using
C-c C-d; S-mode provides a complementary command for loading
source files (even files not created with S-mode!) into the S process:
After typing C-c C-l you will prompted for the name of the file to load into S; usually this is the current buffer's file which is the default value (selected by simply pressing RET at the prompt). You will be asked to save the buffer first if it has been modified (this happens automatically if the buffer was generated with C-c C-d). The file will then be loaded, and if it loads successfully you will be returned to the S process.
If any errors occur when loading a file with C-c C-l, S-mode will
inform you of this fact. In this case,
you can jump directly to the line in the source file which caused the
error by typing C-x ` (S-parse-errors).
You will be returned to the offending file (loading it into a buffer if
necessary) with point at the line S reported as containing the error.
You may then correct the error, and reload the file. Note that none of
the commands in an S source file will take effect if any part of the
file contains errors.
Sometimes the error is not caused by a syntax error (loading a
non-existent file for example). In this case typing C-x ` will
simply display a buffer containing S's error message. You can force this
behaviour (and avoid jumping to the file when there is a syntax
error) by passing a prefix argument to S-parse-errors with
C-u C-x `.
Other commands are also available for evaluating portions of code in the
S process. These commands cause the selected code to be evaluated
directly by the S process as if you had typed them in at the command
line; the source() function is not used. You may choose whether
both the commands and their output appear in the process buffer (as if
you had typed in the commands yourself) or if the output alone is
echoed. The behaviour is controlled by the variable
S-eval-visibly-p whose default is nil
(display output only). Passing a prefix argument (C-u) to any of
the following commands, however, reverses the meaning of
S-eval-visibly-p for that command only -- for example C-u
C-c C-j echoes the current line of S-code in the S process buffer,
followed by its output. This method of evaluation is an alternative to
S's source() function
when you want the input as well as the output to be displayed. (You can
sort of do this with source() when the option echo=T is
set, except that prompts do not get displayed. S-mode puts prompts in
the right places.) The commands for evaluating code are:
S-eval-line) S-eval-line-and-go) S-eval-function) S-eval-function-and-go) S-eval-region) S-eval-region-and-go) S-eval-buffer) S-eval-buffer-and-go) S-eval-line-and-next-line)
It should be stressed once again that these S-eval- commands
should only be used for evaluating small portions of code for debugging
purposes, or for generating transcripts from source files. When editing
S functions, C-c C-l is the command to use to update the
function's value. In particular, S-eval-buffer is now largely
obsolete.
One final command is provided for spot-evaluations of S code:
S-execute-in-tb) This is useful for quick calculations, etc.
All the above commands are useful for evaluating small amounts of code
and observing the results in the process buffer. A useful way to work
is to divide the frame into two windows; one containing the source code
and the other containing the process buffer. If you wish to make the
process buffer scroll automatically when the output reaches the bottom
of the window, you will need to set the variable
comint-scroll-to-bottom-on-output to others or t.
*** Maybe a link to customisation section here ***
S-mode now provides a sophisticated mechanism for indenting S source
code (thanks to Ken'ichi Shibayama). Compound statements (delimited by
`{' and `}') are indented relative to their enclosing block.
In addition, the braces have been electrified to automatically indent to
the correct position when inserted, and optionally insert a newline at
the appropriate place as well. Lines which continue an incomplete
expression are indented relative to the first line of the expression.
Function definitions, if statements, calls to expression()
and loop constructs are all recognised and indented appropriately. User
variables are provided to control the amount if indentation in each
case, and there are also a number of predefined indentation styles to
choose from. See section Variables controlling indentation.
Comments are also handled specially by S-mode, using an idea borrowed
from the Emacs-Lisp indentation style. Comments beginning with `###'
are aligned to the beginning of the line. Comments beginning with
`##' are aligned to the current level of indentation for the block
containing the comment. Finally, comments beginning with `#' are
aligned to a column on the right (the 40th column by default, but this
value is controlled by the variable comment-column,)
or just after the expression on the line containing the comment if it
extends beyond the indentation column.
The indentation commands provided by S-mode are:
S-indent-command) newline-and-indent) S-indent-exp) S-electric-brace) indent-for-comment) GNU, BSD, K&R and C++. The
DEFAULT style uses the default values for the indenting variables
(unless they have been modified in your `.emacs' file.)
This command causes all of the formatting variables to be buffer-local.
A number of commands are provided to move across function definitions in the edit buffer:
S-beginning-of-function) S-end-of-function) S-mark-function) Don't forget the usual Emacs commands for moving over balanced expressions and parentheses: See section `Lists and Sexps' in The GNU Emacs Reference Manual.
Completion is provided in the edit buffer in a similar fashion to the process buffer: M-TAB completes file names and M-? lists file completions. Since TAB is used for indentation in the edit buffer, object completion is now performed with C-c TAB. Note however that completion is only provided over globally known S objects (such as system functions) -- it will not work for arguments to functions or other variables local to the function you are editing.
Finally, two commands are provided for returning to the S process buffer:
S-switch-to-end-of-S) S-switch-to-S) In addition some commands available in the process buffer are also available in the edit buffer. You can still read help files with C-c C-v, edit another function with C-c C-d and of course C-c C-l can be used to load a source file into S. See section Other commands provided by inferior-S-mode for more details on these commands.
Every edit buffer in S-mode is associated with a dump file on
disk. Dump files are created whenever you type C-c C-d
(S-dump-object-into-edit-buffer), and may either be deleted
after use, or kept as a backup file or as a means of keeping
several versions of an S function.
nil, dump files created with C-c C-d are deleted
immediately after they are created by the S-process.
Since immediately after S dumps an object's definition to a disk file
the source code on disk corresponds exactly to S's idea of the object's
definition, the disk file isn't really needed; deleting it now has the
advantage that if you don't modify the file (say, because you
just wanted to look at the definition of one of the standard S
functions) the source dump file won't be left around when you kill the
buffer. Note that this variable only applies to files generated with
S's dump function; it doesn't apply to source files which already
exist. The default value is t.
nil (always
delete), ask (always ask whether to delete), check (delete
files generated with C-c C-d in this Emacs session, otherwise ask
--- this is the default) and t (never delete). This variable is
buffer-local.
After an object has been successfully (i.e. without error) been loaded back into S with C-c C-l, the disk file again corresponds exactly (well, almost -- see below) to S's record of the object's definition, and so some people prefer to delete the disk file rather than unnecessarily use up space. This option allows you to do just that.
If the value of S-keep-dump-files is t, dump files are
never deleted after they are loaded. Thus you can maintain a complete
text record of the functions you have edited within S-mode. Backup
files kept as usual, and so by using the Emacs numbered backup facility
--- see section `Single or Numbered Backups' in The Gnu Emacs Reference Manual, you can keep a historic
record of function definitions. Another possibility is to maintain the
files with a version-control system such as RCS See section `Version Control' in The Gnu Emacs Reference Manual. As long as a dump file exists in the appropriate place for a
particular object, editing that object with C-c C-d finds that
file for editing (unless a prefix argument is given) -- the S process
is not consulted. Thus you can keep comments outside the
function definition as a means of documentation that does not clutter
the S object itself. Another useful feature is that you may format the
code in any fashion you please without S re-indenting the code every
time you edit it. These features are particularly useful for
project-based work.
If the value of S-keep-dump-files is nil, the dump file is always
silently deleted after a successful load with C-c C-l. While this
is useful for files that were created with C-c C-d it also applies
to any other file you load (say, a source file of function
definitions), and so can be dangerous to use unless you are careful.
Note that since S-keep-dump-files is buffer-local, you can make
sure particular files are not deleted by setting it to t in the
Local Variables section of the file See section `Local Variables in Files' in The Gnu Emacs Reference Manual.
A safer option is to set S-keep-dump-files to ask; this
means that S-mode will always ask for confirmation before deleting the
file. Since this can get annoying if you always want to delete dump
files created with C-c C-d, but not any other files, setting
S-keep-dump-files to check (the default value) will
silently delete dump files created with C-c C-d in thie current
Emacs session, but query for any other file. Note that in any case you
will only be asked for confirmation once per file, and your answer
is remembered for the rest of the Emacs session.
Note that in all cases, if an error (such as a syntax error) is detected while loading the file with C-c C-l, the dump file is never deleted. This is so that you can edit the file in a new Emacs session if you happen to quit Emacs before correcting the error.
Dump buffers are always autosaved, regardless of the value of
S-keep-dump-files.
Every dump file should be given a unique file name, usually the dumped object name with some additions.
%s
is replaced by the object name.
By default, dump file names are the user name, followed by `.' and
the object and ending with `.S'. Thus if user joe dumps the
object myfun the dump file will have name `joe.myfun.S'. The
username part is included to avoid clashes when dumping into a
publicly-writable directory, such as `/tmp'; you may wish to remove
this part if you are dumping into a directory owned by you.
You may also specify the directory in which dump files are written:
By default, dump files are always written to `/tmp', which is fine
when S-keep-dump-files is nil. If you are keeping dump
files, then you will probably want to keep them somewhere in your home
directory, say `~/S-source'. This could be achieved by including
the following line in your `.emacs' file:
(setq S-source-directory (expand-file-name "~/S-source/"))
If you would prefer to keep your dump files in separate directories
depending on the value of some variable, S-mode provides a facility for
this also. By setting S-source-directory to a lambda expression
which evaluates to a directory name, you have a great deal of
flexibility in selecting the directory for a particular source file to
appear in. The lambda expression is evaluated with the process buffer
as the current buffer and so you can use the variables local to that
buffer to make your choice. For example, the following expression
causes source files to be saved in the subdirectory `Src' of the
directory the S process was run in.
(setq S-source-directory
(lambda ()
(concat S-directory "Src/")))
(S-directory is a buffer-local variable in process buffers which
records the directory the S process was run from.)
This is useful if you keep your dump files and you often edit objects
with the same name in different S processes. Alternatively, if you
often change your S working directory during an S session, you may like
to keep dump files in some subdirectory of the directory pointed to by
the first element of the current search list. This way you can edit
objects of the same name in different directories during the one S
session:
(setq S-source-directory
(lambda ()
(file-name-as-directory
(expand-file-name (concat
(car S-search-list)
"/.Src")))))
If the directory generated by the lambda function does not exist but can be created, you will be asked whether you wish to create the directory. If you choose not to, or the directory cannot be created, you will not be able to edit functions.
S-mode provides an easy-to-use facility for reading S help files from
within Emacs. From within the S process buffer or any S-mode edit
buffer, typing C-c C-v (S-display-help-on-object)
will prompt you for the name of an object for which you would like
documentation. Completion is
provided over all objects which have help files.
If the requested object has documentation, you will be popped into a
buffer (named *help(obj-name)*) containing the help file.
This buffer is placed in a special `S Help' mode which disables the
usual editing commands but which provides a number
of keys for paging through the help file:
S-describe-help-mode) S-display-help-on-object) scroll-down) scroll-up) beginning-of-buffer) and < (end-of-buffer) S-skip-to-next-section) and p
(S-skip-to-previous-section) S-eval-region) isearch-forward) S-switch-to-end-of-S) kill-buffer) S-kill-buffer-and-go) In addition, all of the S-mode commands available in the edit buffers are also available in S help mode (See section Creating or modifying S objects). Of course, the usual (non-editing) Emacs commands are available, and for convenience the digits and - act as prefix arguments.
If a help buffer already exists for an object for which help is
requested, that buffer is popped to immediately; the S process is not
consulted at all. If the contents of the help file have changed, you
either need to kill the help buffer first, or pass a prefix argument
(with C-u) to S-display-help-on-object.
Help buffers are marked as temporary buffers in S-mode, and are deleted
when S-quit or S-cleanup are called.
Inferior S mode records the transcript (the list of all commands executed, and their output) in the process buffer, which can be saved as a transcript file, which should normally have the suffix `.St'. The most obvious use for a transcript file is as a static record of the actions you have performed in a particular S session. Sometimes, however, you may wish to re-execute commands recorded in the transcript file by submitting them to a running S process. This is what Transcript Mode is for.
If you load file a with the suffix `.St' into Emacs, it is placed in S Transcript Mode. Transcript Mode is similar to Inferior S mode (see section Interacting with the S process): paragraphs are defined as a command and its output, and you can move though commands either with the paragraph commands or with C-c C-p and C-c C-n.
Three commands are provided to re-submit command lines from the transcript file to a running S process. They are:
S-transcript-send-command) S-transcript-copy-command) S-transcript-send-command-and-move) Note that these commands are similar to those on the same keys in Inferior S Mode. In all three cases, the commands should be executed when the cursor is on a command line in the transcript; the prompt is automatically removed before the command is submitted.
Yet another use for transcript files is to exatract the command lines for inclusion in an S source file or function. Transcript mode provides one command which does just this:
S-transcript-clean-region) The remaining command lines may then be copied to a source file or edit buffer for inclusion in a function definition, or may be evaluated directly (see section Sending code to the S process) using the code evaluation commands from S mode, also available in S Transcript Mode.
S-mode has a few miscellaneous features, which didn't fit anywhere else.
S-mode provides Font-Lock (see section `Using Multiple Typefaces' in The Gnu Emacs Reference Manual) patterns for Inferior S Mode, S Mode, and S Transcript Mode buffers.
To activate the highlighting, you need to turn on Font Lock mode in the
appropriate buffers. This can be done on a per-buffer basis with
M-x font-lock-mode, or may be done by adding
turn-on-font-lock to inferior-S-mode-hook,
S-mode-hook and S-transcript-mode-hook (see section Customizing S-mode with hooks).
Your systems administrator may have done this for you in
`S-site.el' (see section Customizing S-mode).
The font-lock patterns are defined in three variables, which you may modify if desired:
TRUE.
One of the main features of the S package is its ability to
generate high-resolution graphics plots, and S-mode provides a number of
features for dealing with such plots.
printer() driver
This is the simplest (and least desirable) method of using graphics
within S-mode. S's printer() device driver produces crude
character based plots which can be contained within the S process buffer
itself. To start using character graphics, issue the S command
printer(width=79)
(the width=79 argument prevents Emacs line-wrapping at column
80 on an 80-column terminal. Use a different value for a terminal with
a different number of columns.) Plotting commands do not generate
graphics immediately, but are stored until the show() command
is issued, which displays the current figure.
Of course, the ideal way to use graphics with S-mode is to use a
windowing system. Under X windows, this requires that the DISPLAY
environment variable be appropriately set, which may not always be the
case within your Emacs process. S-mode provides a facility for setting
the value of DISPLAY before the S process is started if the variable
S-ask-about-display
is non-nil. See section Customizing S-mode for details of this variable,
and see section Starting the S process for information on how to set the value of
DISPLAY when beginning an S session.
S-display-help-on-object and list completion cannot
be used while the user is entering a multi-line command. The only real
fix in this situation is to use another S process.
S-eval- commands can leave point in the S process buffer in
the wrong place when point is at the same position as the last process
output. This proves difficult to fix, in general, as we need to consider
all windows with window-point at the right place.
S-keep-dump-files is nil.
S-execute causes S-mode to
hang.
You can send a Bug report by typing M-x S-submit-bug-report, or by
sending E-mail to D.M.Smith@lancaster.ac.uk.
Comments, suggestions, words of praise and large
cash donations are also more than welcome.
The following section details those steps necessary to get S-mode running on your system.
First of all, you need to create a directory (say, `~/elisp') to place the Emacs-Lisp files. Copy `S.el', `S-tek.el', `comint.el', `comint-isearch.el' and `comint-extra.el' to that directory, and add the lines
(setq load-path (cons (expand-file-name "~/elisp") load-path)) (autoload 'S "S" "Run an inferior S process" t) (autoload 's-mode "S" "Mode for editing S source" t)
This will be enough to get S-mode running on most systems -- see section Starting the S process for details on starting S-mode. If it does not work, see section Other variables you may need to change for other variables you may need to change. See section Customizing S-mode for other variables you may wish to set in your `.emacs' file, but it is suggested you defer this section until you are more familiar with S-mode.
It is recommended that the .el
files all be byte-compiled
with M-x byte-compile-file
for efficiency.
If you run the S program (from the shell) with a command other than
`Splus' you will need to set the variable inferior-S-program
to the name of the appropriate program by including a line such as
(setq inferior-S-program "S+")
in your `.emacs' file (substituting `S+' for the name of your S program.)
If you need to call this program with any arguments, the variable you
need to set is dependent on the value of inferior-S-program; for
example if it is "Splus", set the variable
inferior-Splus-args
to a string of arguments to the Splus program. If
inferior-S-program has some other value, substitute the
Splus part of inferior-Splus-args with the appropriate
program name. There aren't many instances where you need to call S with
arguments, however: in particular do not call the S program with the
`-e' command-line editor argument since S-mode provides this
feature for you.
If you are running an older version of S, you may need to set the
variable S-version-running
to reflect this fact. The default is "3.0" which indicates the
August '91 revision; any other value indicates an older version.
This variable is effective only when S-mode is loaded; setting it
during an S session has no effect.
If you are running Splus (the enhanced version of S from Statsci) you
may also need to set the variable S-plus to t. If your
value of inferior-S-program is "S+" or Splus this
will not be necessary, however; S-plus defaults to t in
this case.
Finally, if you use a non-standard prompt within S, you will need to set the
variable inferior-S-prompt
to a regular expression which will match both the primary prompt ("> "
by default) and the continuing prompt (default of "+ ".) The
default value of this variable matches S's default prompts. For example,
if you use ("$ ") as your primary prompt (you have
options(prompt="$ ") in your .First function), add the
following line to your `.emacs':
(setq inferior-S-prompt "^\\(\\+\\|[^\\$]*\\$\\) *")
You will also need to set the variable inferior-S-primary-prompt
to a regular expression which matches the primary prompt only. Do not
anchor the regexp to the beginning of the line with `^'. Once
again, the default value matches S's default prompt; in the example
above the appropriate value would be "[^\\$]*\\$ *".
Once these variables are set appropriately, S-mode should work on any system.
S-mode can be easily customised to your taste simply by including the appropriate lines in your `.emacs' file. There are numerous variables which affect the behaviour of S-mode in certain situations which can be modified to your liking. Keybindings may be set or changed to your preferences, and for per-buffer customisations hooks are also available.
S-mode is easily customisable by means of setting variables in your `.emacs' file. In most cases, you can change defaults by including lines of the form
(setq variable-name value)
in your `.emacs'.
In what follows, variable names will be listed along with their descriptions and default values. Just substitute the variable name and the new value into the template above.
t nil value, then every time S-mode is
run with M-x S
you will be prompted for a directory to use as the working directory for
your S session; this directory should have a `.Data' subdirectory.
If the value of S-ask-for-S-directory is nil, the value of
S-directory
is used as the working directory.
S-ask-for-S-directory
is nil, and the default when prompting for a directory if it is
not. For example, you may wish to set this to the value of the current
buffer's working directory before starting S by adding the following
line to your `.emacs' file (See section Customizing S-mode with hooks)
(setq S-pre-run-hook '((lambda () (setq S-directory default-directory))))
nil nil value, then every time S-mode is
run with M-x S
you will be asked for a value for the DISPLAY environment
variable
to be used in the current S session. If this variable is not set
correctly, S will not be able to create any windows under the X
windowing environment.
Completion is provided over the X-displays-list variable; the
default is the current value of DISPLAY. This feature is useful
is you often run S on a different display than that of the machine you
are running S from. If
S-ask-about-display is nil, the current value of
DISPLAY is used.
'(":0.0") DISPLAY environment variable,
provided for completion when prompting for such a value.
t nil value, then dumping a non-existent
object will result in the edit buffer containing a skeleton function
definition, ready for editing.
"/tmp/"
nil
.object_name.S
nil
t nil, then TAB in the edit buffer always indents the
current line, regardless of the position of point in the line.
Otherwise, indentation is only performed if point is in the lines
indentation, and a tab character is inserted is point is after the first
nonblank character.
nil nil means automatically newline before and after braces
inserted in S code.
The following variables control amounts of indentation. These variables automatically become buffer-local in any S-mode buffer, and so setting any of these variables has effect in the current buffer only.
S-continued-statement-offset.
foo when it is called as
the value of an argument to another function in
arg=foo(...) form. If not number, the statements are indented at
open-parenthesis following foo.
expression() specified in
obj <- expression(...)
form. If not a number, the statements are indented at open-parenthesis following `expression'.
else clause with respect to the
corresponding if.
In addition, a number of default styles are defined for you (in
S-style-alist):
DEFAULT
nil nil, then the S-execute- commands (see section Other commands provided by inferior-S-mode)
output to a temporary buffer. Otherwise, the
output goes to the S process.
nil nil, then the S-eval- commands (see section Creating or modifying S objects) echo the S commands in the process buffer by default. In any
case, passing a prefix argument to the eval command reverses the meaning
of this variable.
S-mode provides five hooks, as follows:
S-mode is run, i.e. every time an edit buffer
is generated.
S-eval-visibly, say.
S-mode provides a separate keymap variable for the S process buffer, for edit buffers and for help buffers.
comint-mode-map are automatically inherited.
S-help-sec-map is
the keymap for the `s' prefix key. Keys defined in
S-help-sec-keys-alist are automatically inserted into this keymap
when S-mode is loaded.