Internet Info 1994 March

home *** CD-ROM | disk | FTP | other *** search

/ Internet Info 1994 March / Internet Info CD-ROM (Walnut Creek) (March 1994).iso / answers / comp / unix-faq / shell / rc < prev next >

Wrap

Internet Message Format | 1994-04-09 | 43.6 KB

Path: bloom-beacon.mit.edu!hookup!swrinde!sdd.hp.com!saimiri.primate.wisc.edu!news.doit.wisc.edu!oldp.astro.wisc.edu!alan From: alan@oldp.astro.wisc.edu (Alan Watson) Newsgroups: comp.unix.shell,comp.answers,news.answers Subject: rc shell Frequently Asked Questions Followup-To: poster Date: 8 Apr 1994 20:13:11 GMT Organization: Department of Astronomy, University of Wisconsin -- Madison Lines: 1098 Approved: news-answers-request@MIT.Edu Expires: 04 May 1994 Message-ID: <2o4dsn$j3j@news.doit.wisc.edu> NNTP-Posting-Host: oldp.astro.wisc.edu Summary: Answers to Frequently Asked Questions about rc -- a clean, portable, and freely-available shell for UNIX systems. Originator: alan@oldp.astro.wisc.edu Xref: bloom-beacon.mit.edu comp.unix.shell:9508 comp.answers:4824 news.answers:17834 Archive-name: unix-faq/shell/rc Comp-unix-shell-archive-name: rc-FAQ Version: 2.16 Last-modified: 08 Apr 1994 This document attempts to answer some Frequently Asked Questions about rc, a clean, portable, and freely-available shell for UNIX systems (see 1.1 for a very brief overview of rc). It in no way replaces either the formal documentation for rc (see 1.2) or the mailing list archives (see 2.2). This document draws heavily on the collective wisdom of the rc mailing list; the quality of the correspondence is such that often all that was required was a summary of one or two messages. The rc mailing list archive is recommend reading for newcomers to rc (see 2.2). Contributors are acknowledged at the end of this document. Corrections, comments, and suggestions concerning this document are welcome (see 3.2). NOTE: Neither the editor nor any of the contributors to this document are responsible for any consequences resulting from the advice in this document. Table of Contents 1. Implementations of rc 1.1 What is rc? 1.2 Where is rc documented? 1.3 How do I obtain rc? 1.4 Are there any patches to rc? 1.5 Can I compile rc on a ... machine with ... operating system and ... compiler? 1.6 Should rc be dynamically or statically linked? 1.7 To whom do I report bugs? 1.8 Does anyone have an implementation of /dev/fd for my UNIX? 2. The rc mailing list 2.1 How do I subscribe to the rc mailing list? 2.2 How do I obtain the rc mailing list archive? 3. The rc FAQ list 3.1 How do I obtain the current version of the rc FAQ list? 3.2 To whom do I make suggestions about the rc FAQ list? 4. Does rc have ... ? 4.1 Does rc have job control? 4.2 Does rc have command-line editing? 4.3 Does rc have a history mechanism? 4.4 Does rc have file name completion? 4.5 Does rc have home directory expansion? 4.6 Does rc check for new mail? 4.7 Does rc have hierarchical lists? 4.8 Does rc have cartesian products? 4.9 Does rc cache the location of executables in the path? 4.10 Does rc have C-style \ escapes? 4.11 Does rc have ... built in? 4.12 Does rc have echo built in? 4.13 Does rc have test built in? 4.14 Does rc have true, false, and `:' built in? 5. Using rc as a login shell 5.1 How do I make rc my login shell? 5.2 How do I use rc in a heterogeneous cluster? 5.3 What problems occur with rc on Apollos? 5.4 What problems occur with rc on SYSV-derived systems? 5.5 What problems occur with rc on NeXTs? 5.6 How do I keep down the size of my .rcrc file? 5.7 How do I speed up the execution of my .rcrc file? 6. Interactions between rc and the rest of UNIX 6.1 How do I get each new interactive rc to do ...? 6.2 How do I get rc to do ... in each new xterm? 6.3 How do I get rc to read .rcrc when I use rsh? 6.4 How do I get at or batch to run an rc script? 7. Programming in rc 7.1 What makes rc such a good scripting language? 7.2 What are the trade-offs between functions and scripts? 7.3 What can I do if I have deficient sed, awk, test, ... ? 7.4 Where do I obtain the GNU tools and other software? 7.5 How do I perform arithmetic in rc? 7.6 How do I find the first occurrence of an element in a list? 7.7 How do I use ranges in array subscripts in rc? 7.8 How do I perform the equivalent of read in rc? 1. Implementations of rc 1.1 What is rc? rc is a shell designed by Tom Duff of Bell Labs to replace the venerable Bourne sh in Plan 9. (Plan 9 is an experimental operating and networking system developed at Bell Labs.) In addition to Plan 9, Duff's rc runs on SunOS and v10 UNIX. The design of rc is extremely clean, in that precisely those features of Bourne's sh that have caused problems over the years (the slightly awkward flow control syntax, the Byzantine quoting rules, the failure to export all variables into the environment) were reworked, and the temptation to add superfluous features was strongly resisted. Byron Rakitzis has written a version of rc that is portable between many version on UNIX and is largely compatible with Duff's design. Rakitzis' implementation has been successfully used on UNIX platforms ranging from PCs running Linux to Cray Y-MPs running Unicos. Most of this document deals with issues that have arisen from use of Rakitzis' rc, and most of the example code will follow his minor modifications to rc. Henceforth, a plain `rc' refers to Rakitzis' rather than Duff's rc. 1.2 Where is rc documented? rc is documented in its man page, distributed with the source (see 1.3). The Plan 9 manual pages, including those on Duff's rc, are available as research.att.com:/dist/plan9man/manual.ps.Z by anonymous FTP. The full manual is almost 600 pages long, but the manual pages for Duff's rc are physical pages 142 to 147 (logical pages 116 to 121). An early version of Duff's rc is described in the paper `rc -- a Shell for Plan 9 and UNIX Systems', written for a UKUUG meeting. Duff has kindly allowed the paper to be distributed by anonymous FTP as viz.tamu.edu:/pub/rc/plan9.ps. It provides an excellent introduction and tutorial for Duff's rc, and describes the motivations for the differences between Duff's rc and Bourne's sh. When reading Duff's paper and the Plan 9 manual pages, bear in mind that Rakitzis' implementation of rc differs in detail from Duff's (these differences are documented in the rc man page). Another valuable source of information on rc is the rc mailing list archive (see 2.2). 1.3 How do I obtain rc? The sources are available by anonymous FTP in the directory viz.tamu.edu:/pub/rc. Be sure to get the most recent version. At the time this document was last updated, the most recent version of rc was v1.4 announced on 21 May 1992, available as viz.tamu.edu:/pub/rc/rc-1.4.tar.Z. (A beta release of v1.5 is on the horizon.) rc is distributed on very liberal terms, although it has not been placed in the public domain (see the file COPYRIGHT distributed with the sources). Duff's rc can be obtained by acquiring either a Plan 9 or a v10 UNIX system from Bell Labs. Plan 9 can be licensed by academic institutions, which should contact: Neera Kuckreja RM 2C557 ATT Bell Laboratories Murray Hill NJ 07974 USA +1 908 582-3855 neera@research.att.com 1.4 Are there any patches to rc? Although the current version of rc is remarkably mature, a small number of bugs have been found since its release and patches have been posted to the rc mailing list (see 2.2). Other patches posted to the mailing list include work-arounds for broken compilers or libraries and assorted features. An up-to-date set of patches extracted from the mailing list archive, together with a script to apply those patches, is available as oldp.astro.wisc.edu:/pub/rc/patches.tar.Z by anonymous FTP. 1.5 Can I compile rc on a ... machine with ... operating system and ... compiler? rc is written in portable ANSI C, but it can be compiled by many older C compilers after passing the sources through unproto (get ftp.win.tue.nl:/pub/unix/unproto4.shar.Z by anonymous FTP). If you use gcc, use v1.39 or later. rc has been successfully compiled on systems with the following combinations of architectures, operating systems, and compilers, although this list is probably not exhaustive: DEC VAX Ultrix v3.x gcc v1 DEC MIPS Ultrix v4.[23] gcc v2 DEC MIPS Ultrix v4.[23] cc [a] DEC AXP OSF/1 v1.2 cc [a][b] DEC AXP OSF/1 v1.2 cc -std1 [b] Sun SPARC Solaris v2 gcc v2 [b][c] Sun SPARC acc [b][c] Sun SPARC SunOS v4.1.[123] cc [a][b][c] Sun SPARC SunOS v4.1.[123] gcc v2 [b][c] Sun SPARC BSD [b][c] Sun 68K SunOS v4.1.1 gcc v2 [b][c] Sun 68K SunOS v3.5 gcc v1 [b][c] SGI Iris Irix v[45].x gcc v2 IBM RS/6000 AIX v3.1.5 cc IBM RS/6000 AIX v3.2.[013] cc HP PA-RISC HP-UX v8.0 gcc v2 [d][e] HP PA-RISC HP-UX v9.0 gcc v2 [d] HP PA-RISC OSF/1 v1.0.2 cc i386/i486 SCO-ODT SVR3 gcc v2 [d][e] i386/i486 SCO-ODT SVR3 cc [d][e] i386 Minix v1.5 [d] i386/i486 NetBSD gcc v[12] i386/i486 Linux gcc v2 [d] i386/i486 386BSD r0.0 gcc [j] i386/i486 SCO Xenix SVR2.3x [f] NeXT NeXT OS v2.x gcc v[12] [g] NeXT NeXT OS v3.0 gcc v2 [g] Apollo 68K DomainOS v10.4 (BSD) cc [h] Apollo 10000 DomainOS v10.4 (BSD) cc [h] Sony MIPS NEWS/OS cc [a] Sony 68K NEWS/OS v4.0.1 cc [a] Convex C1 ConvexOS v10.1 cc Pyramid 9810 OSx 5.1a (BSD) cc -Xa [i] Cray Y-MP Unicos 6.1.7 scc [e][j] Sequent Dynix v3.2.0 (BSD) gcc v2 Apple 68K A/UX v3.0 cc Encore MIPS Umax v4.3 gcc v2 Acorn ARM iX v1.21a cc Motorola 88K System V R32V3 gcc v2 Motorola 88K System V R32V3 cc [a][k] Notes: [a] Sources pre-processed with unproto. [b] See 1.6. [c] Sun's yacc has a memory leak -- use another implementation such as GNU bison (see 7.4). [d] Requires intervention by hand to get mksignal to work, because of of the non-standard signal.h file. [e] See 4.13. [f] See 5.4. [g] See 5.5. [h] See 5.3 [i] This compiler has no stdarg.h, but vararg.h sufficed. [j] Minor modifications required to prototypes. [k] Greenhills cc, used without any optimization. 1.6 Should rc be dynamically or statically linked? As far as one can generalize, statically linked executables fork faster than dynamically linked executables. This suggests that shells, which fork a great deal, should probably be statically linked (i.e., you should not link them with shared libraries). rc makes so little use of libc functions that static linking results in only a small increase in the size of the executable on disc. Versions of UNIX which support shared libraries include SVR4, SunOS, Solaris, OSF/1, and Irix. 1.7 To whom do I report bugs? Byron Rakitzis (byron@netapp.com) or the rc mailing list (rc@hawkwind.utcs.toronto.edu). Please check the documentation before reporting bugs, and check the mailing list archive in case the bug is already known. 1.8 Does anyone have an implementation of /dev/fd for my UNIX? /dev/fd pseudo-devices allow a more reliable implementation of non-linear pipelines (i.e., <{...} and >{...}) than named pipes, but they are not present in all versions of UNIX. An implementation of Ultrix v4.2 and later was posted to the mailing list in March 1993 (see 2.2). An implementation for SunOS v4.1.1 is also available (although it may well work with v4.1.2 and v4.1.3) by anonymous FTP as ftp.cs.arizona.edu:/devfd.shar. An implementation for SunOS v4.1.2 is available by anonymous FTP as emx.cc.utexas.edu:/pub/mnt/source/sun/devfd.shar. (This is packaged as a loadable module, so you do not have to rebuild your kernel.) An implementation for NeXT OS 2.1 is available by anonymous FTP as ftp.sys.toronto.edu:/pub/rc/devfd-next.shar.gz. You will, of course, need to rebuild your kernel to install these pseudo-devices. 2. The rc mailing list 2.1 How do I subscribe to the rc mailing list? Send mail to rc-request@hawkwind.utcs.toronto.edu. Chris Siebenmann maintains the mailing list, and can be reached at the above address in the event of problems with the mailing list or to unsubscribe. The rc mailing list represents hundreds of user-years of experience with rc. Clearly, it is a courtesy to the list and to your own advantage to check the mailing list archive before posting to the list, to benefit from the experience and mistakes of those before you, and to prevent topics being continually rehashed. 2.2 How do I obtain the rc mailing list archive? The archive can be obtained by anonymous FTP. At the moment, it is split chronologically between the original and current maintainers of the mailing list. Period File Jun 91 to Dec 91 viz.tamu.edu:/pub/rc/rc-list.1991.Z Jan 92 to May 92 viz.tamu.edu:/pub/rc/rc-list May 92 to present ftp.sys.toronto.edu:/pub/rc/rc-list 3. The rc FAQ list 3.1 How do I obtain the current version of the rc FAQ list? The current version of the rc FAQ list is available by anonymous FTP as oldp.astro.wisc.edu:/pub/rc/rc-FAQ. This FAQ is also archived automatically and is available by anonymous FTP as rtfm.mit.edu:/pub/usenet/comp.unix.shell/rc-FAQ. 3.2 To whom do I make suggestions about the rc FAQ list? To its editor, Alan Watson (alan@oldp.astro.wisc.edu), or the rc mailing list if you think your suggestion warrants a wider audience. 4. Does rc have ... ? 4.1 Does rc have job control? No. This has been discussed at length in the rc mailing list. Now that personal workstations with window systems are common place, the case for job control is less compelling. However, if you have to use rc on a traditional terminal, you may wish to obtain the screens program from a GNU archive (see 7.4), the sm program that was posted to alt.sources in mid-1992 (see 7.4), or investigate shell escape mechanisms in vi and shell mode in emacs. 4.2 Does rc have command-line editing? If you wish. Hooks are provided for the readline or editline libraries which provide command-line editing. The default is no command-line editing. The readline library can be obtained from a GNU archive (see 7.4). The editline library is similar to the readline library, but is a fraction of the size and offers fewer features. The sources are available as ftp.sys.toronto.edu:/pub/rc/editline.shar by anonymous FTP). Alternatively, you can use fep, ile, atty, or Rk to implement editing on a per-terminal basis. These have been discussed in on the mailing list at one time or another, and modifications to ile have been posted there. 4.3 Does rc have a history mechanism? Yes. While, rc does not have a built-in csh-like history mechanism, rather it is able to write interactive commands to a file (see the history special variable in the man page), allowing a history mechanism similar to the v8 `=' command to be implemented as an external program (one such implementation is supplied with the source of rc). In Plan 9, `:' was the history program (the name was probably changed from the `=' used in v8 because `=' is a special character in rc, and needs quoting if it refers to a command). However, this form of history was apparently used so little that the ability to write commands to a file was removed from Duff's rc. The history program distributed with the rc source is invoked with `-'. Both editline and readline have command recall. 4.4 Does rc have file name completion? If you wish. Plain rc does not have file name completion (as it obtains its command input with read), but the readline library does. 4.5 Does rc have home directory expansion? No. This has been hotly debated on the rc mailing list, and a patch to implement this was posted to the rc mailing list in November 1992 (see 2.2 and 1.4). 4.6 Does rc check for new mail? No. However, you can make it do so with the following prompt function: fn prompt { mail -e && echo You have mail. } Alternatively, consider using biff or xbiff. 4.7 Does rc have hierarchical lists? No. Each element of an rc list must be a string, and cannot be another list. Whenever a list appears inside another list, it is expanded so that the outer list contains its elements. Similarly, whenever two lists are adjacent, they are concatenated to form a single list (with one exception). In rc terms, the arguments to a command form a list, even though they may not have explicit `(' and `)' delimiters around them. Hierarchical lists do not correspond well to the `char **argv' model for arguments to UNIX commands, although simple lists do. The single exception to these rules are the arguments to the `~' built-in command, which rc interprets as two lists -- a subject list and a pattern list. Normal expansion and concatenation is performed on the two lists separately. A consequence of this is that the following command: if ( ~ $1 () '' ) ... does not check if $1 is either the null list or the null string, as normal list concatenation occurs on the pattern list leading to a pattern list with one element -- the null string. 4.8 Does rc have cartesian products? No. If you really need them, they can be obtained in the following manner: ; x = (a b c) ; y = (1 2 3) ; product = () ; for ( elem in $x ) product = ( $product $elem^$y ) ; echo $product a1 a2 a3 b1 b2 b3 c1 c2 c3 rc's concatenation operator works in a distributive manner if one of the lists has only a single element and in a pairwise manner if both lists have the same number of elements: ; echo a^( x y z ) ax ay az ; echo ( a b c )^( x y z ) ax by cz 4.9 Does rc cache the location of executables in the path? No. This has been discussed on the mailing list. As far as one can generalize, searching the path is not a performance bottleneck on most modern UNIX systems. 4.10 Does rc have C-style \ escapes? No. One of the most attractive features of rc is its single quoting rule. If you need to embed control characters in your strings, investigate tr and printf (see 7.3 if your system lacks the latter). 4.11 Does rc have ... built in? It is not by accident that rc has fewer built-in commands than most shells other than the original Bourne shell. The underlying philosophy, eloquently summarized by Boyd Roberts, is that a shell is there to run other programs, not to have other programs built into it. By having a well-defined role, rc has remained small and simple, properties that lead to better performance, fewer bugs, and a more thorough understanding of the shell by its users. Most of the rc built-in commands perform functions that are essential for a shell and cannot be performed by external programs (e.g., cd, eval, limit, umask, wait). rc does include hooks for users to add new built-in commands, and these have been used to add test, access, kill, read, and even sync and haltsys commands (see the mailing list and the examples file for more details). Such additions are unsupported and potentially unportable. 4.12 Does rc have echo built in? If you wish. While most of rc's built-in commands could not easily be replaced by external commands, echo is the one exception. The justification is that echo is used so much that providing it as a built-in command results in a significant gain in performance, although this opinion is by no means universally held. By default, echo is built in. It has also been noted that rc's echo is more portable that /bin/echo, as the v7 and SYSV implementations of echo differ in the method that must be used to suppress the final newline. (However, rc's echo differs from both, as it uses `--' to indicate that all other arguments are to be echoed literally.) People who dislike the notion of a built-in echo command have the option of excluding it from the compilation. Alternatively, the built-in echo can be subverted with the following function, although this will not defeat an explicit call to `builtin echo'. fn echo { /bin/echo $* } See also 7.3. 4.13 Does rc have test built in? No. You can use the built-in `~' command to match strings, and fall back on /bin/expr or /bin/test for numeric comparisons and /bin/test for querying the file system. If you are using rc on a system that lacks a stand-alone /bin/test and /bin/[ and has test built into sh, the following serves as an interim replacement: #! /bin/sh `basename $0` ${1+"$@"} # end-of-file This applies to Unicos and HP-UX v8. Note that trip.rc, the script provided with the rc sources to check basic functionality after compilation, will fail unless a suitable test command exists in the path. SCO UNIX implements /bin/test in precisely this way. You can almost certainly get better performance by using the GNU implementation of /bin/test (see 7.4), although this is obviously more important on an i386 machine than on a Cray. It is possible to get at the file system obliquely through the globbing mechanism and the builtin cd command, and in some circumstances these can be of use (see, for example, 5.7). Such speed hacks need to be carefully justified. See also 7.3. 4.14 Does rc have true, false, and `:' built in? No. The following functions are just as fast: fn true { return 0 } fn false { return 1 } fn : { return 0 } Note that an indefinite loop in rc does not require the use of an explicit `true' command: while () { ... } 5. Using rc as a login shell 5.1 How do I make rc my login shell? On most systems, you or the system administrator will have to add a line to /etc/shells containing the full path to rc. This will need to be done on each machine on which you wish to use rc. Next, use the chsh command to specify your new preference. If your machine is running Yellow Pages (YP), you will have to do this on the YP server, otherwise you will have to do this on each machine individually. (To determine if your machine is using YP, type `ypwhich -m passwd' that will either print the name of your YP server or give an error if your machine is not running YP.) Often, you will also need to change your password (using the yppasswd command) on the YP server to propagate the new /etc/passwd entry contain your selected shell; changing your password to itself will normally suffice. An alternative method is to invoke rc from your .profile file or its equivalent, but this has so many drawbacks that I cannot recommend it. 5.2 How do I use rc in a heterogeneous cluster? There are two issues: finding the correct executable when you login, and getting the correct path in scripts run using the #! kernel hack. If you have root privileges on every machine on the cluster, by far the simplest solution is the place a copy of rc in /bin and add /bin/rc to /etc/shells on each machine. Use `#! /bin/rc' in shell scripts. Otherwise, use a sh wrapper around rc. Create a shell script called, say, rc-login, and make this your login shell. The purpose of this script is to set SHELL and PATH to point appropriately according to the machine, and then to exec rc. The following script is suitable for adaptation: #! /bin/sh # determine OS TMP=/tmp/$$ ( exec >$TMP ; echo OS=`uname -m` ) . $TMP rm -f $TMP # determine location of rc executable case $OS in RISC) PATH=$HOME/bin/mips:$PATH SHELL=$HOME/bin/mips/rc ;; sun4c) PATH=$HOME/bin/sparc:$PATH SHELL=$HOME/bin/sparc/rc ;; alpha) PATH=$HOME/bin/alpha:$PATH SHELL=$HOME/bin/alpha/rc ;; *) echo 1>&2 "WARNING: unknown architecture" SHELL=/bin/sh ;; esac # determine if this is a login shell case $SHELL in */rc) case $0 in -*) FLAG=-l ;; esac ;; esac export PATH export SHELL exec $SHELL $FLAG ${1+"$@"} # end-of-file The dance to capture the output of uname is necessary because of a bug whereby redirection does not work in the original v7 sh if stdout is closed. (This bug lives on in the Ultrix /bin/sh.) The easiest way to manage shell scripts is to keep a master copy in a central directory, say $home/bin/share, and have slave scripts in machine-specific directories which source the master script. For example, for a command called `foo' create the following in each machine-specific directory: #! /full/path/to/machine/specific/rc . $home/bin/share/foo $* # end-of-file See 5.3 if your cluster contains Apollos. 5.3 What problems occur with rc on Apollos? Apollos require the login shell to be a binary, so you cannot use a sh script wrapper around rc. 5.4 What problems occur with rc on SYSV-derived systems? Many SYSV-derived sh and ksh are `restricted' if the final component of SHELL contains the letter `r'. A restricted shell will not allow you to cd, perform output redirection, set PATH, or specify absolute paths to executables, meaning that many shell scripts written in sh or ksh will fail. This is also applies to BSD-derived systems that also ship with a SYSV- derived shell (for example, /bin/sh5 on Ultrix). One solution is to link rc to a file that does not contain the letter `r', and use that file as your SHELL. You can achieve this automatically by placing the following commands in your .rcrc: rcdir = `{ dirname $SHELL } { test -f $rcdir/RC || ln $rcdir/rc $rcdir/RC SHELL = $rcdir/RC } Make sure when you install a new version of rc that you remove the previous RC. This behaviour is also a potential problem if you use a sh wrapper around rc as your login shell (see 5.2 and 6.3). Make sure that the name of the wrapper does not contain the letter `r'. 5.5 What problems occur with rc on NeXTs? The NeXT window system assumes that all shells are `job control' shells and will place themselves in a new process group. Contrary to this assumption, rc is not a job control shell and so does not do this. The result is that a signal to one (such as an interrupt) will be sent to all. The `newpgrp' builtin-in command was provided to deal with this problem. You can use it to ensure that each new interactive rc belongs to a new process group by using a prompt function to run code in each new interactive instance of rc (see 6.1 and 6.2). Other more elegant solutions (such as a wrapper around rc) are precluded by the uncooperative and inflexible nature of the NeXT window system. See also 6.1 and 6.2. 5.6 How do I keep down the size of my .rcrc file? Given the nature of rc, the natural place to define functions is in the .rcrc file. However, as the number of functions grows, it becomes more awkward to maintain them all in a single file. One solution is to break out the functions into separate files and automatically load them within .rcrc: fnlib = $home/fnlib fn load { ~ $#* 1 && builtin . $fnlib/$1 } builtin cd $fnlib name = () for ( name in * ) load $name builtin cd See also 5.7. 5.7 How do I speed up the execution of my .rcrc file? If you have lots of functions, you may wish to use lazy function loading: fnlib = $home/fnlib fn load { ~ $#* 1 && builtin . $fnlib/$1 } builtin cd $fnlib name = () for ( name in * ) fn $name { load $0 && $0 $* } builtin cd Often, though, a more substantial amount of time is spent testing directories before they are added to a path: rc does not have a built-in test command, so each test implies a fork and an exec. In some circumstances, you can replace the test with code similar to the following: path = () dir = () for ( dir in ... ) builtin cd $dir >[1=] >[2=] && path = ( $path $dir ) builtin cd See also 5.6 and 4.13. 6. Interactions between rc and the rest of UNIX 6.1 How do I get each new interactive rc to do ...? Rather than set the `-l' flag and place the code in .rcrc, you can use a prompt function like the following: fn prompt { if ( ! ~ $ipid $pid ) { ipid=$pid ... } } This will run ... at the start of each new interactive instance of rc, but not in non-interactive instances of rc. See also 6.2 and 5.5. 6.2 How do I get rc to do ... in each new xterm? Since xterm puts WINDOWID into the environment, you can use a prompt function like the following: fn prompt { if ( ! ~ $wid $WINDOWID && ~ $TERM xterm* ) { wid=$WINDOWID ... } } See also 6.1. 6.3 How do I get rc to read .rcrc when I use rsh? rshd does not place a `-' in the first character of argv[0] when it execs a shell, so .rcrc is not run and this can cause problems. One solution is to have a false login shell called login-shell that explicitly supplies a -l flag to rc, i.e., login-shell contains: #! /path/to/rc exec /path/to/rc -l $* # end-of-file See also 5.2. 6.4 How do I get at or batch to run an rc script? at and batch only accept sh or csh scripts, so you need to package your rc script in a sh wrapper. The following two functions accomplish this. Note that they do not completely emulate the standard commands, as they will not accept a file name on the command line, but rather only read commands from stdin. A simple further enhancement might be to mail the stdout and stderr of the job back to the user. fn at { if ( ~ $#* 0 ) { echo >[2=1] \ 'usage: at [ time [day] | -r job ... | -l [job ...] ]' return 1 } switch ( $1 ) { case -r builtin at $* case -l builtin at $* >[2=1] case * { echo 'SHELL='^$SHELL echo 'export SHELL' echo 'sed ''s/^-//'' <<''+'' | '^$SHELL sed 's/^/-/' echo '+' } | SHELL = /bin/sh builtin at $* } } fn batch { if ( ! ~ $#* 0 ) { echo >[1=2] 'usage: batch' return 1 } { echo 'SHELL='^$SHELL echo 'export SHELL' echo 'sed ''s/^-//'' <<''+'' | '^$SHELL sed 's/^/-/' echo '+' } | SHELL = /bin/sh builtin batch } 7. Programming in rc 7.1 What makes rc such a good scripting language? Brian Kernighan and Rob Pike wrote of the original v7 Bourne shell: The UNIX shell isn't typical of command interpreters: although it lets you run commands in the usual way, because it is a programming language it can accomplish much more. Their comments apply equally to rc, perhaps more so. The two aspects of rc that contribute most to its utility as a scripting language are its simple quoting rule coupled with the absence of rescanning and the fact that the language is small, simple, and well defined. The former allows one to write scripts with complete control over the lexical aspects of the process: when and if globbing, pattern matching, and list formation take place. Tom Duff wrote that rc is `not a macro processor'. The lack or rescanning, together with improvements to backquote substitutions, allowed him to collapse Bourne's four types of quoting down to one, eliminating all of their complex interactions. The latter means that the whole language can be readily understood by the user. Its grammar is well defined, in sharp contrast to sh. (Indeed, a skeleton yacc grammar is included as an appendix to rc's manual page.) The language is small and well defined; there are, to quote John Mackin, no `dark corners'. Although the language is smaller and simpler, it retains all the power and flexibility of sh, and improves and extends on it in several areas. 7.2 What are the trade-offs between functions and scripts? Functions are run in the process that invokes them, so they can make persistent changes to the environment of the interpreter. They are essential for some actions (reading a variable, for example). Functions make excellent wrappers around standard commands, as one can use the `builtin' command to invoke the standard command from within the function. Invoking a standard command from within a script of the same name involves either hard-wiring the name into the script or using a more elaborate method, such as the `pathos' script posted to the mailing list in June 1992. There is less overhead in invoking a function, as the kernel need not exec rc once again. However, functions take up space in the environment. There is often a hard limit on the size of the environment. Although on some systems this limit is very large, on others it is not. In addition, very large environments can degrade the performance of fork. Functions cannot be exec-ed directly by a C program, but must be invoked through rc, although it is probably a good idea to do this anyway, as responsibility for globbing and variable substitution properly resides with the shell. 7.3 What can I do if I have deficient sed, awk, test, ... ? Shell programming on UNIX rests on two foundations: the shell itself and the toolkit of programs provided in /bin and /usr/bin. Unfortunately, many vendors provide out-of-date tools and there are many incompatibilities between BSD- and SYSV- derived systems, so writing portable scripts is a problem. This problem can be especially acute in heterogeneous clusters. There are two approaches to this problem: one is to use only those tools and options that were present in v7, and the other is to use only those tools and options that are present in POSIX.2, and to replace the vendor-supplied tools with those from GNU if the former do not conform (see 7.4). 7.4 Where do I obtain the GNU tools and other software? Get prep.ai.mit.edu:/pub/gnu/GNUinfo/FTP by anonymous FTP for a list of sites that archive GNU software. The article `How to find sources' lists sites that archive software posted to Usenet source newgroups. This article is regularly cross-posted to comp.sources.wanted, comp.answers, and news.answers, and is also available via anonymous FTP as rtfm.mit.edu:/pub/usenet/comp.answers/finding-sources. 7.5 How do I perform arithmetic in rc? You can use /bin/expr, /bin/awk, or unary arithmetic with lists (see 7.6 for an example). 7.6 How do I find the first occurrence of an element in a list? Use an `index' function, as in: ; x = ( a b a b ) ; index b $x 2 This implementation of index uses a list to count up to the index of the element: fn index { switch ( $#* ) { case 0 1 echo >[1=2] 'usage: index pattern list ...' return 1 case * elem = $1 prefix = () { shift if ( ! ~ $elem $* ) { echo 0 return 1 } while ( prefix = ( $prefix $1 ) && ! ~ $elem $1 ) shift echo $#prefix return 0 } } } 7.7 How do I use ranges in array subscripts in rc? Use a `seq' function, as in: ; x = ( a b c d e f g h ) ; echo $x( `{seq 2 5} ) b c d e There are many implementations of seq, most of which use awk, like this one: fn seq { switch ( $#* ) { case 1 * = ( 1 $1 ) case 2 case * echo >[1=2] usage: seq [start] stop return 1 } echo $1 $2 | awk '{ if ( $1 <= $2 ) for ( i = $1; i <= $2; i++ ) print i; else for ( i = $1; i >= $2; i-- ) print i; }' } 7.8 How do I perform the equivalent of read in rc? Use a `read' function, to read a single line from stdin, remove the trailing newline, and assign it to a variable: ; read foo Do the funky gibbon! ; echo $foo Do the funky gibbon! ; echo $#foo 1 The following implementation of read returns 1 if EOF is seen, like the read in the v7 sh: fn read { if ( ! ~ $#* 1 ) { echo >[1=2] 'usage: read variable' return 1 } if ( ~ $1 '*' () *'='* ) { echo >[1=2] 'read: bad variable name' return 1 } lines = () nl = () { ifs = () { nl = `{ echo } } ifs = $nl { lines = `{ line ; echo $status } } if ( ~ $#lines 2 ) { $1 = $lines(1) return 0 } else { $1 = '' return $lines(1) } } } If your UNIX does not have a line command, the following is an straightforward substitute. #include <unistd.h> int main () { unsigned char c; while (read (0, &c, 1) == 1 && c != '\n') write (1, &c, 1); write (1, "\n", 1); return c != '\n'; } Attempts to use standard UNIX utilities (e.g., "sed 1q", "head -1", and "awk '{ print; exit; }'") to replace the line command fail because line does not buffer its input. This was discussed extensively in the mailing list in September 1991 (see 2.2). Alternatively, a patch to add a builtin read was posted to the mailing list in September 1993 (see 1.4). Acknowledgements Many people have contributed ideas, code, and information to this document, either directly or indirectly, through the rc mailing list and its influence on my understanding of rc. People who have directly contributed to this document include Vincent Broman, Raymond Chen, Tom Culliton, Stefan Dalibor, Jim Davis, Matt Day, Matthew Farwell, David Fiander, Tim Goodwin, Paul Haahr, Charles Hannum, David Hogan, Noel Hunt, John LoVerso, Hamish Macdonald, John Mackin, Byron Rakitzis, Boyd Roberts, Rich Salz, Chris Siebenmann, Emin Gun Sirer, Icarus Sparry, Gerry Tomlinson, Malte Uhl, Christopher Vance, and Victor Zandy. My apologies to anyone I have missed. As a user of rc, I am indebted to Tom Duff for its design, to Steve Bourne for its foundation, and to Byron Rakitzis for its implementation. Alan Watson (alan@oldp.astro.wisc.edu)