xref: /titanic_51/usr/src/lib/libshell/common/sh.memo (revision 15d9d0b528387242011cdcc6190c9e598cfe3a07)
1.	 \" use troff -mm
2.nr C 3
3.nr N 2
4.SA 1  \"  right justified
5.ND "December 21, 1993"
6.TL "311466-6713" "61175"  \" charging case filing case
7Introduction to \f5ksh-93\fP
8.AU "David G. Korn" DGK MH 11267 7975 3C-526B "(research!dgk)"
9.TM  11267-931221-26  \"  technical memo + TM numbers
10.MT 1  \"  memo type
11.OK Shell "Command interpreter" Language UNIX  \" keyword
12.AS 2   \" abstract start for TM
13\f5ksh-93\fP is a
14major rewrite of \f5ksh\fP,
15a program that serves as a command language
16(shell) for the UNIX*
17.FS  *
18UNIX is a registered trademark of Novell.
19.FE
20operating system.
21As with \f5ksh\fP, \f5ksh-93\fP
22is essentially compatible with the System V version of the Bourne shell\*(Rf,
23.RS
24S. R. Bourne,
25.I "An Introduction to the UNIX
26Shell,"
27BSTJ - Vol. 57, No. 6 part 2, pages 1947-1972, 1978.
28.RF
29and compatible with previous versions of \f5ksh\fP.
30\f5ksh-93\fP is intended to comply with the IEEE POSIX 1003.2
31and ISO 9945-2\*(Rf
32.RS
33.I "POSIX \- Part 2: Shell and Utilities,"
34IEEE Std 1003.2-1992, ISO/IEC 9945-2, IEEE, 1993.
35.RF
36shell standard.
37In addition to changes in the language required
38by these standards, the primary focus of \f5ksh-93\fP
39is related to shell programming.
40\f5ksh-93\fP provides the programming power of several
41other interpretive languages such as \f5awk\fP\*(Rf,
42.RS
43Al Aho,
44Brian Kernighan,
45and
46Peter Weinberger,
47.I "The AWK Programming Language,"
48Addison Wesley, 1988.
49.RF
50\f5FIT\fP\*(Rf,
51.RS
52Lloyd H. Nakatani and Laurence W. Ruedisueli,
53.I "The FIT Programming Language Primer",
54TM 1126-920301-03, 1992.
55.RF
56\f5PERL\fP\*(Rf,
57.RS
58Larry Wall and Randal Schwartz,
59.I "Programming perl,"
60O'Reilly & Assoc, 1990.
61.RF
62and
63\f5tcl\fP\*(Rf.
64.RS
65John K. Ousterhout,
66.I "Tcl: An Embeddable Command Language",
67Proceedings of the Washington USENIX meeting, pp. 133-146, 1990.
68.RF
69.P
70This memo
71assumes that the reader is already familiar with the Bourne shell.
72It introduces most of the features of \f5ksh-93\fP
73relative to the Bourne shell; both
74as a command language and as a programming language.
75The Appendix contains
76a sample script written in \f5ksh-93\fP.
77.AE   \" abstract end
78.H 1 "INTRODUCTION"
79.P
80The term "shell" is used to describe a program that provides
81a command language
82interface.
83Because the UNIX*\
84.FS *
85UNIX is a registered trademark of USL
86.FE
87system shell is a user level program, and not part of
88the operating system itself,
89anyone can write a new shell or modify an existing one.
90This has caused an evolutionary progress
91in the design and implementation of shells,
92with the better ones surviving.
93The most widely available UNIX system shells are the Bourne shell\*(Rf,
94.RS
95S. R. Bourne,
96.IR "An Introduction to the UNIX Shell" ,
97Bell System Technical Journal,
98Vol. 57, No. 6, Part 2, pp. 1947-1972, July 1978.
99.RF
100written by Steve Bourne
101at AT&T Bell Laboratories,
102the C shell\*(Rf,
103.RS
104W. Joy,
105.IR "An Introduction to the C Shell" ,
106Unix Programmer's Manual, Berkeley Software Distribution,
107University of California, Berkeley, 1980.
108.RF
109written by Bill Joy at the University of California, Berkeley,
110and the KornShell language \*(Rf,
111.RS
112Morris Bolsky and David Korn,
113.IR "The KornShell Command and Programming Language" ,
114Prentice Hall, 1989.
115.RF
116written by David Korn
117at AT&T Bell Laboratories.
118The Bourne shell is available on almost all versions of the UNIX
119system.
120The C Shell is available with all Berkeley Software Distribution (BSD) UNIX systems and on many other systems.
121The KornShell
122is available on System V Release 4 systems.
123In addition, it is available on many other systems.
124The source for the KornShell language is available from the AT&T Toolchest,
125an electronic software distribution system.
126It runs on all known versions of the UNIX system and
127on many UNIX system look-alikes.
128.P
129There have been several articles comparing the UNIX system shells.
130Jason Levitt\*(Rf
131.RS
132Jason Levitt,
133.IR "The Korn Shell: An Emerging Standard" ,
134UNIX/World, pp. 74-81, September 1986.
135.RF
136highlights some of the new features
137introduced by the KornShell language.
138Rich Bilancia\*(Rf
139.RS
140Rich Bilancia,
141.IR "Proficiency and Power are Yours With the Korn Shell" ,
142UNIX/World, pp. 103-107, September 1987.
143.RF
144explains some of the advantages of using the KornShell language.
145John Sebes\*(Rf
146.RS
147John Sebes,
148.I "Comparing UNIX Shells,"
149UNIX Papers,
150Edited by the Waite Group, Howard W. Sams & Co., 1987.
151.RF
152provides a more detailed comparison of the three shells,
153both as a command language and as a programming language.
154.P
155The KornShell language is a superset of the
156Bourne shell. The KornShell language has many of the popular C shell features,
157plus additional features of its own.
158Its initial popularity stems primarily from its improvements as
159a command language.
160The primary interactive benefit of the KornShell command language
161is a visual command line editor that allows you to
162make corrections to your current command line
163or to earlier command lines,
164without having to retype them.
165.P
166However,
167in the long run,
168the power of the KornShell language as a high-level programming language,
169as described by Dolotta and Mashey\*(Rf,
170.RS
171T. A. Dolotta and J. R. Mashey,
172.I "Using the shell as a Primary Programming Tool,"
173Proc. 2nd. Int. Conf. on Software Engineering, 1976,
174pages 169-176.
175.RF
176may prove to be of greater significance.
177\f5ksh-93\fP provides the programming power of several
178other interpretive languages such as \f5awk\fP,
179\f5FIT\fP,
180\f5PERL\fP,
181and
182\f5tcl\fP.
183An application that was originally written in the C programming language
184was rewritten in the KornShell language.
185More than 20,000 lines of C code were replaced with KornShell scripts
186totaling fewer than 700 lines.
187In most instances there was no perceptible difference in performance
188between the two versions of the code.
189.P
190The KornShell language has been embedded into windowing systems
191allowing graphical user interfaces to be developed in shell
192rather than having to build applications that need to be
193compiled.
194The \f5wksh\fP program\*(Rf
195.RS
196J. S. Pendergrast,
197.IR "WKSH - Korn Shell with X-Windows Support",
198USL. 1991.
199.RF
200provides a method of developing OpenLook or Motif
201applications as \f5ksh\fP scripts.
202.P
203This memo is an introduction to \f5ksh-93\fP,
204the program that implements an enhanced version
205of the KornShell language.
206It is referred to as \f5ksh\fP in the rest of this memo.
207The memo describes the KornShell language based on the
208features of the 12/28/93 release of \f5ksh\fP.
209This memo is not a tutorial, only an introduction.
210The second edition of reference [9] gives
211a more complete treatment of the KornShell language.
212.P
213A concerted effort has been made to achieve both System V Bourne shell
214compatibility and IEEE POSIX compatibility
215so that scripts written for either of these shells
216can run without modification with \f5ksh\fP.
217In addition, \f5ksh-93\fP attempts to
218be compatible with older versions of \f5ksh\fP.
219When there are conflicts between versions of the shell,
220\f5ksh-93\fP selects the behavior dictated by the IEEE POSIX
221standard.
222The description of features in this memo assumes
223that the reader is already familiar with the Bourne shell.
224.H 1 "COMMAND LANGUAGE"
225There is no separate command language.
226All features of the language, except job control,
227can be
228used both within a script and interactively from a terminal.
229However, features that are more likely to be used
230while running commands interactively from a terminal
231are presented here.
232.H 2 "Setting Options"
233By convention, UNIX commands
234consist of a command name followed by options and other arguments.
235Options are either of the form \f5-\fP\fIletter\fP,
236or \f5-\fP\fIletter value\fP.
237In the former case, several options may be grouped after a single \f5-\fP.
238The argument \f5--\fP signifies an end to the option list and is
239only required when the first non-option argument begins with
240a \f5-\fP.
241Most commands print an error message which
242shows which options are permitted
243when given incorrect arguments.
244In addition, the option sequence \f5-?\fP causes most commands
245to print a usage message which lists the valid options.
246.P
247Ordinarily, \f5ksh\fP executes a command by
248using the command name to locate a program to run
249and by running the program as a separate process.
250Some commands, referred to as
251.IR built-ins ,
252are carried out by \f5ksh\fP itself,
253without creating a separate process.
254The reasons that some commands are built-in are presented later.
255In nearly all cases the distinction
256between a command that is built-in and one that
257is not is invisible to the user.
258However, nearly
259all commands that are built-in follow command line conventions.
260.P
261\f5ksh\fP has several options that can be set by the user
262as command line arguments at invocation and as option arguments to the
263\f5set\fP command.
264Most other options can be set with a single letter option or as a name
265that follows the \f5-o\fP option.
266Use
267\f5set\ -o\fP
268to display the current option settings.
269Some of these options, such as
270.B interactive
271and
272.B monitor
273(see
274.I "Job Control"
275below),
276are enabled automatically by \f5ksh\fP
277when the shell is connected to a terminal device.
278Other options, such as
279.B noclobber
280and
281.BR ignoreeof ,
282are normally placed in a startup file.
283The
284.B noclobber
285option causes
286\f5ksh\fP
287to print an error message when you use
288.B >
289to redirect output to a file that already exists.
290If you want to redirect to an existing file, then
291you have to use
292.B >|
293to override
294the
295.B noclobber
296option.
297The
298.B ignoreeof
299option
300is used to prevent the
301.I end-of-file
302character, normally
303.B ^D
304(Control- d),
305from exiting the shell and possibly logging you out.
306You must type \f5exit\fP
307to log out.
308Most of the options are described in this memo as appropriate.
309.H 2 "Command Aliases"
310.P
311Command aliases provide a mechanism of associating a command name and
312arguments with a shorter name.
313Aliases are defined with the \f5alias\fP
314built-in.
315The form of an \f5alias\fP
316command definition is:
317.ce
318\f5alias\fP \fIname\fP\f5=\fP\fIvalue\fP
319As with most other shell assignments, no space is allowed before or after
320the \f5=\fP.
321The characters of an alias name cannot be characters that are
322special to the shell.
323The replacement string,
324.I value,
325can contain any valid shell script,
326including meta-characters such as pipe symbols and i/o-redirection
327provided that they are quoted.
328Unlike
329\f5csh\fP,
330aliases in
331\f5ksh\fP
332cannot take arguments.
333The equivalent functionality of aliases with arguments can
334be achieved with shell functions, described later.
335.P
336As a command is being read,
337the command name is checked against a list of
338.I alias
339names.
340If it is found,
341the name is replaced by the alias value associated with the
342.I alias
343and then rescanned.
344When rescanning the value for an alias, alias substitutions
345are performed except for an alias that is currently being processed.
346This prevents infinite loops in alias substitutions.
347For example with the aliases, \f5alias\ l=ls\ 'ls=ls\ -C'\fP,
348the command name \f5l\fP becomes \f5ls\fP, which becomes \f5ls\ -C\fP.
349Ordinarily, only the command name word is processed for alias substitution.
350However, if the value of an alias ends in a space,
351then the word following the alias is also checked for alias substitution.
352This makes it possible
353to define an alias whose first argument is the name of a command
354and have alias substitution performed on this argument,
355for example
356\f5nohup='nohup\ '\fP.
357.P
358Aliases can be used to redefine built-in commands so that
359the alias,
360.ce
361\f5alias test=./test\fP
362can be used to look for \f5test\fP
363in your current working directory rather than
364using the built-in \f5test\fP command.
365Reserved words such as
366\f5for\fP and \f5while\fP
367cannot be changed by aliasing.
368The command \f5alias\fP,
369without arguments, generates
370a list of aliases and corresponding alias values.
371The \f5unalias\fP command removes the name and text of an alias.
372.P
373Aliases are used to save typing and to improve readability of scripts.
374Several aliases are predefined by \f5ksh\fP.
375For example, the predefined alias
376.ce
377\f5alias integer='typeset -i'\fP
378allows the integer variables \f5i\fP and \f5j\fP
379to be declared and initialized with the command
380.ce
381\f5integer i=0 j=1\fP
382.P
383While aliases can be defined in scripts,
384it is not recommended.
385The location of an alias command can be important
386since aliases are only processed when a command is read.
387A \fB\s+2.\s-2\fP
388procedure (the shell equivalent of an include file)
389is read all at once (unlike
390start up files
391which are read a command at
392a time) so that any aliases defined there will not effect any commands
393within this script.
394Predefined aliases do not have this problem.
395.H 2 "Command Re-entry"
396.P
397When run interactively,
398\f5ksh\fP saves the
399commands you type at a terminal in a file.
400If the variable
401\fB\s-1HISTFILE\s+1\fP
402is set to the name of a file to which the user
403has write access,
404then the commands are stored in this
405.I history
406file.
407Otherwise the file
408\fB$\s-1HOME\s+1/.sh_history\fP
409is checked for write access and if this fails
410an unnamed file is used to hold the history lines.
411Commands are always appended to this file.
412Instances of \f5ksh\fP
413that run concurrently and use the same history file
414name, share access to the history file so that a command
415entered in one shell will be available for editing in another
416shell.
417The file may be truncated when \f5ksh\fP
418determines that no other shell is using the history file.
419The number of commands accessible to the user is determined by the value of the
420\fB\s-1HISTSIZE\s+1\fP
421variable at the time the shell is invoked.
422The default value is 256.
423Each command may consist of one or more lines since a compound
424command is considered one command.
425If the character
426.B !
427is placed within the
428.I "primary prompt"
429string,
430\fB\s-1PS1\s+1\fP,
431then it is replaced by the command number each time the prompt is given.
432.P
433A built-in command named \f5hist\fP
434is used to list and/or edit
435any of these saved commands.
436The option
437.B \-l
438is used to specify listing of previous commands.
439The command can always be specified with
440a range of one or more commands.
441The range can be specified by giving the command
442number, relative or absolute, or by giving
443the first character or characters of the command.
444When given without specifying the range,
445the last 16
446commands are listed, each
447preceded by the command number.
448.P
449If the listing option is not selected,
450then the range of commands specified,
451or the last command if no range is given,
452is passed to an editor program before
453being re-executed by \f5ksh\fP.
454The editor to be used may be specified
455with the option
456.B \-e
457and following it with the editor name.
458If this option is not specified, the
459value of the shell variable
460\fB\s-1HISTEDIT\s+1\fP
461is used as the name of the editor,
462providing that this variable has a non-null value.
463If this variable is not set, or is null,
464and the
465.B \-e
466option has not been selected,
467then
468\f5/bin/ed\fP
469is used.
470When editing has been complete,
471the edited text automatically becomes
472the input for \f5ksh\fP.
473As this text is read by \f5ksh\fP, it is echoed onto the terminal.
474.P
475The
476.B \-s
477option causes the editing to be bypassed
478and just re-executes the command.
479In this case only a single command can be specified as the range
480and an optional argument of the form
481\fIold\fP\fB=\fP\fInew\fP
482may be added which requests a simple string substitution
483prior to evaluation.
484A convenient alias,
485.ce
486\f5alias r='hist -s'\fP
487has been pre-defined so that
488the single key-stroke
489\f5r\fP
490can be used to re-execute the previous command
491and the key-stroke sequence,
492\f5r\ abc=def\ c\fP
493can be used to re-execute the last command that starts with
494the letter \f5c\fP
495with the first occurrence of the string \f5abc\fP
496replaced with the string \f5def\fP.
497Typing
498\f5r\ c\ >\ file\fP
499re-executes the most recent command starting with the letter \f5c\fP,
500with standard output redirected to
501.IR file .
502.H 2 "In-line editing"
503.P
504Lines typed from a terminal frequently need changes made
505before entering them.
506With the Bourne shell the only method to fix up commands
507is by backspacing or killing the whole line.
508\f5ksh\fP offers options that allow the user to edit parts of the
509current command line before submitting the command.
510The in-line edit options make the command line into a single
511line screen edit window.
512When the command is longer than the width of the terminal,
513only a portion of the command is visible.
514Moving within the line automatically makes that portion visible.
515Editing can be performed on this window until the
516.I return
517key is pressed.
518The editing modes have editing directives that access the history file
519in which previous commands are saved.
520A user can copy any of the most recent
521\fB\s-1HISTSIZE\s+1\fP
522commands from this file into the input edit window.
523You can locate commands by searching or by position.
524.P
525The in-line editing options do not use the
526.I termcap
527or
528.I terminfo
529databases.
530They work on most standard terminals.
531They only require that the backspace character moves the cursor left
532and the space character overwrites the current character on the screen
533and moves the cursor to the right.
534Very few terminals or terminal emulators do not have
535this behavior.
536.P
537There is a choice of editor options.
538The
539.BR emacs ,
540.BR gmacs ,
541or
542.B vi
543option is selected by turning on the
544corresponding
545option of the \f5set\fP
546command.
547If the value of the
548\fB\s-1EDITOR\s+1\fP
549or
550\fB\s-1VISUAL\s+1\fP
551variables ends with any of these suffixes
552the corresponding option is turned on.
553A large subset of each of these editors'
554features is available within the shell.  Additional
555functions, such as file name completion, have also been added.
556.P
557In the
558.B emacs
559or
560.B gmacs
561mode the user positions the cursor to the point
562needing correction and inserts, deletes, or replaces
563characters as needed.
564The only difference between these two modes is the
565meaning of the directive
566.BR ^T .
567Control keys and escape sequences are used for cursor
568positioning and control functions.
569The available editing functions are listed in the manual page.
570.P
571The
572.B vi
573editing mode
574starts in insert mode and enters control mode when the
575user types ESC ( 033 ).
576The
577.I return
578key, which submits the current command for processing,
579can be entered from either mode.
580The cursor can be anywhere on the line.
581A subset of commonly used
582.I vi
583editing directives are available.
584The
585.B k
586and
587.B j
588directives that normally move up and down by one
589.IR line ,
590move up and down one
591.I command
592in the history file,
593copying the command into the input edit window.
594For reasons of efficiency,
595the terminal is kept in canonical mode until an
596ESC
597is typed.
598On some terminals,
599and on earlier versions of the UNIX operating system,
600this doesn't work correctly.
601The
602.B viraw
603option,
604which always uses
605.I raw
606or
607.I cbreak
608mode,
609must be used in this case.
610.P
611Most of the code for the editing options does not rely on the
612\f5ksh\fP code and can be used in a stand-alone mode with most any command
613to add in-line edit capability.
614However,
615all versions of the in-line editors have some features that
616use some shell specific code.  For example,
617with all edit modes, the
618ESC-=
619directive applied to command words
620(the first word on the line,
621or the first word after a
622.BR ; ,
623.BR | ,
624.BR ( ,
625or
626.BR & )
627lists all aliases, functions, or commands
628that match the portion of the given current word.
629When applied to other words, this directive
630prints the names of files that match the current
631word.
632The ESC\fB-*\fP directive
633adds the expanded list of matching files to the command line.
634A trailing
635.B *
636is added to the word if it doesn't contain any file pattern matching
637characters before the expansion.
638In
639.B emacs
640and
641.B gmacs
642mode,
643ESC-ESC
644indicates command completion when applied to
645command names, otherwise it indicates pathname completion.
646With command or pathname completion,
647the list generated by the
648ESC-= directive is examined to find
649the longest common prefix.
650With command completion, only the last component of
651the pathname is used to compute the longest command prefix.
652If the longest common prefix is a complete match,
653then the word is replaced by the pathname, and a
654.B /
655is appended if
656pathname is a directory, otherwise a space is added.
657In
658.B vi
659mode,
660.B \e
661from control mode gives the same behavior.
662.H 2 "Key Binding"
663.P
664It is possible to intercept keys as they are entered and
665apply new meanings or bindings.
666A trap named
667\fB\s-1KEYBD\s+1\fP
668is evaluated each time
669\f5ksh\fP processes characters entered
670from the keyboard,
671other than those typed
672while entering a search string or an argument to an
673edit directive such as
674.B r
675in vi-mode.
676The action associated with this trap can change the value of
677the entered key to cause the key to perform a different
678operation.
679.P
680When the
681\fB\s-1KEYBD\s+1\fP
682trap is entered,
683the \fB.sh.edtext\fP
684variable contains the contents of the current input line
685and the \fB.sh.edcol\fP
686variable gives the current cursor position within this line.
687The \fB.sh.edmode\fP
688variable contains the
689.B ESC
690character when the trap is entered from
691.B vi
692insert mode.
693Otherwise, this value is null.
694The \fB.sh.edchar\fP
695variable contains the character or
696escape sequence that caused the trap.
697A key sequence is either a single character,
698.B ESC
699followed by a single character,
700or
701.B ESC[
702followed by a single character.
703In the \fBvi\fP edit mode,
704the characters after the
705.B ESC
706must be entered within half a second after the
707.BR ESC .
708The value of \fB.sh.edchar\fP
709at the end of the trap will be used as
710the input sequence.
711.P
712Using the associative array facility of \f5ksh\fP described later,
713and the function facility of \f5ksh\fP, it is easy to write
714a single trap so that keys can be bound dynamically.  For example,
715.sp
716.nf
717.in .5i
718.ta 4i
719\f5typeset -A Keytable
720trap 'eval "${Keytable[${.sh.edchar}]}"' KEYBD
721function keybind # key action
722{
723        typeset key=$(print -f "%q" "$2")
724        case $# in
725        2)      Keytable[$1]='.sh.edchar=${.sh.edmode}'"$key"
726                ;;
727        1)      unset Keytable[$1]
728                ;;
729        *)      print -u2 "Usage: $0 key [action]"
730                ;;
731        esac
732}\fP
733.ta
734.in
735.fi
736.sp
737.H 2 "Job Control"
738.P
739The job control mechanism
740is almost identical to the version introduced in \f5csh\fP
741of the Berkeley UNIX operating system,
742version 4.1 and later.
743The job control feature allows the user to stop and
744restart programs, and to move programs to and from the
745foreground and the background.
746It will only work on systems that provide support for
747these features.
748However,
749even systems without job control have a
750.B monitor
751option which, when enabled, will report the progress
752of background jobs and enable the user to \f5kill\fP
753jobs by job number or job name.
754.P
755An interactive shell associates a
756.I job
757with each pipeline typed in from the terminal
758and assigns it a small integer number
759called the job number.
760If the job is run asynchronously,
761the job number is printed at the terminal.
762At any given time, only one job owns the terminal,
763i.e., keyboard signals are only sent to the processes in one job.
764When \f5ksh\fP creates a foreground job,
765it gives it ownership of the terminal.
766If you are running a job and wish to stop
767it you hit the key
768.B ^Z
769(control-\fBZ\fP)
770which sends a
771\fB\s-1STOP\s+1\fP
772signal to all processes in the current job.
773The shell receives notification that the processes
774have stopped and takes back control of the terminal.
775.P
776There are commands to continue programs in the foreground
777and background.
778There are several ways to refer to jobs.
779The character
780.B %
781introduces a job name.
782You can refer to jobs by name or number as described in the manual page.
783The built-in command \f5bg\fP
784allows you to continue a job in the background,
785while the built-in command \f5fg\fP
786allows you to continue a job in the foreground even
787though you may have started it in the background.
788.P
789A job being run in the background will stop if it tries
790to read from the terminal.
791It is also possible to stop background jobs that try to write on
792the terminal by setting the terminal options
793appropriately.
794.P
795There is a built-in command \f5jobs\fP
796that lists the status of all running and stopped jobs.
797In addition,
798you are informed of the change of state (running or stopped)
799of any background
800jobs just before each prompt.
801If you want to be notified about background job completions
802as soon as they occur without waiting for a prompt, then use the
803.B notify
804option.
805When you try to exit the shell while jobs are stopped or running,
806you will receive a message from \f5ksh\fP.
807If you ignore this message and try to exit again,
808all stopped processes will be terminated.
809In addition, for login shells, the
810\fB\s-1HUP\s+1\fP
811signal will be sent to
812all background jobs
813unless the job has been disowned with the
814.B disown
815command.
816.P
817A built-in version of \f5kill\fP
818makes it possible to use
819.I job
820numbers as targets for signals.
821Signals can be selected by number or name.
822The name of the signal is the name found in the
823.I include
824file
825.B /usr/include/sys/signal.h
826with the prefix
827.B \s-1SIG\s+1
828removed.
829The
830.B \-l
831option of \f5kill\fP
832provides a means to map individual signal names to and from
833signal number.
834In addition, if no signal name or number is given,
835\f5kill\ -l\fP
836generates a list of valid signal names.
837.H 2 "Changing Directories"
838By default,
839\f5ksh\fP
840maintains a logical view of the file system hierarchy
841which makes symbolic links transparent.
842For systems that have symbolic links,
843this means that if \f5/bin\fP is a symbolic link to \f5/usr/bin\fP
844and you change directory to \f5/bin\fP, \f5pwd\fP will indicate
845that you are in \f5/bin\fP, not \f5/usr/bin\fP.
846\f5pwd\ -P\fP
847generates the physical pathname of the present working
848directory by resolving all the symbolic links.
849By default,
850the \f5cd\fP
851command will take you where you expect to go even if you cross
852symbolic links.
853A subsequent \f5cd\ ..\fP in the example above
854will place you in \f5/\fP, not \f5/usr\fP.
855On systems with symbolic links,
856\f5cd\ -P\fP
857causes
858.B ..
859to be treated physically.
860.P
861\f5ksh\fP remembers your last directory
862in the variable
863\fB\s-1OLDPWD\s+1\fP.
864The \f5cd\fP
865built-in can be given with argument
866.B \-
867to return to the previous directory
868and print the name of the directory.
869Note that \f5cd\ -\fP
870done twice returns you to the starting directory,
871not the second previous directory.
872A directory
873.I stack
874manager has been written as shell
875.I functions
876to
877.I push
878and
879.I pop
880directories from the stack.
881.H 2 "Prompts"
882.P
883When \f5ksh\fP
884reads commands from a terminal,
885it issues a prompt whenever it is ready
886to accept more input and then
887waits for the user to respond.
888The
889\fB\s-1TMOUT\s+1\fP
890variable
891can be set to be the number of seconds that the shell will wait for
892input before terminating.
893A 60 second warning message is printed
894before terminating.
895.P
896The shell uses two prompts.
897The primary prompt,
898defined by the value of the
899\fB\s-1PS1\s+1\fP
900variable,
901is issued at the start of each command.
902The secondary prompt,
903defined by the value of the
904\fB\s-1PS2\s+1\fP
905variable,
906is issued when more input is needed to complete a command.
907.P
908\f5ksh\fP allows the user to specify a list of files or directories
909to check before issuing the
910\fB\s-1PS1\s+1\fP
911prompt.
912The variable
913\fB\s-1MAILPATH\s+1\fP
914is a colon (
915.B :
916) separated list of file names to be checked for changes
917periodically. The user is notified
918before the next prompt.
919Each of the names in this list can be followed by a
920.B ?
921and a message to be given when a change has been detected in the file.
922The prompt will be evaluated for parameter expansion, command
923substitution and arithmetic expansion which are described later.
924The parameter
925.B $_
926within a mail message will evaluate to the name of the file that
927has changed.
928The parameter
929\fB\s-1MAILCHECK\s+1\fP
930is used to specify the minimal interval in seconds before
931new mail is checked for.
932.P
933In addition to replacing each
934.B !
935in the prompt with the command number,
936\f5ksh\fP expands
937the value of the
938.B \s-1PS1\s+1
939variable
940for parameter expansions, arithmetic expansions,
941and command substitutions as described below
942to generate the prompt.
943The expansion characters that are to be applied when
944the prompt is issued must be quoted to prevent the
945expansions from occurring when assigning the value to
946.B \s-1PS1\s+1.
947For example,
948\f3\s-1PS1\s+1="$\s-1PWD\s+1"\fP
949causes
950.B \s-1PS1\s+1
951to be set to the value of
952.B \s-1PWD\s+1
953at the time of the assignment whereas
954.B \s-1PS1\s+1='$\s-1PWD\s+1'
955causes
956.B \s-1PWD\s+1
957to be expanded at the time the prompt is issued.
958.P
959Command substitution may require a separate process
960to execute and cause the prompt display to be somewhat
961slow, especially
962when the return key is pressed several times in a row.
963Therefore, its use
964within
965.B \s-1PS1\s+1
966is discouraged.
967Some variables are maintained by \f5ksh\fP
968so that their values can be used with
969.B \s-1PS1\s+1.
970The
971.B \s-1PWD\s+1
972variable stores the pathname of the current working directory.
973The value of
974.B \s-1SECONDS\s+1
975variable
976is the value of the most
977recent assignment plus the elapsed time.
978By default, the time is measured in milli-seconds,
979but since
980.B \s-1SECONDS\s+1
981is a floating point variable, the
982number of places after the decimal point in the expanded
983value can be
984specified with
985\f5typeset\ -F\fP\fIplaces\fP\f5\ SECONDS\fP.
986In a roundabout way, this variable
987can be used to generate a time stamp into the
988.B \s-1PS1\s+1
989prompt without creating a process at each prompt.
990The following code explains how you can do this on
991System V.  On BSD, you need a different command to initialize
992the
993.B \s-1SECONDS\s+1
994variable.
995\f5
996.sp
997.nf
998.in .5i
999# . this script and use $TIME as part of your PS1 string to
1000# get the time of day in your prompt
1001typeset -RZ2  _x1 _x2 _x3
1002(( SECONDS=$(date  '+3600*%H+60*%M+%S') ))
1003_s='_x1=(SECONDS/3600)%24,_x2=(SECONDS/60)%60,_x3=SECONDS%60,0'
1004TIME='"${_d[_s]}$_x1:$_x2:$_x3"'
1005# PS1=${TIME}whatever
1006.fi
1007.ta
1008.in
1009.sp
1010\fP
1011.H 2 "Tilde substitution"
1012.P
1013The character
1014.B \(ap
1015at the beginning of a word has special meaning to \f5ksh\fP.
1016If the characters after the
1017.B \(ap
1018up to a
1019.B /
1020match a user login name in the password database, then the
1021.B \(ap
1022and the name are replaced by
1023that user's login directory.
1024If no match is found, the original word
1025is unchanged.
1026A
1027.B \(ap
1028by itself, or in front of a
1029.BR / ,
1030is replaced by the value of the
1031\fB\s-1HOME\s+1\fP
1032parameter.
1033A
1034.B \(ap
1035followed by a
1036.B +
1037or
1038.B \-
1039is replaced by the value of
1040.B $\s-1PWD\s+1
1041or
1042.B $\s-1OLDPWD\s+1
1043respectively.
1044.H 2 "Output formats"
1045The output of built-in commands and traces have values quoted so that they
1046can be re-input to the shell.
1047This makes it easy to cut and paste shell output on systems
1048which use a pointing device such as a mouse.
1049In addition, output can be saved in a file for reuse.
1050.P
1051.H 2 "The \fB\s-1ENV\s+1\fP file"
1052When an interactive \f5ksh\fP starts, it evaluates the
1053.B $\s-1ENV\s+1
1054variable to arrive at a file name.
1055If this value is not null,
1056\f5ksh\fP attempts to read and process
1057commands in a file by this name.
1058Earlier versions of \f5ksh\fP read the \fB\s-1ENV\s+1\fP file
1059for all invocations of the shell primarily to allow
1060function definitions to be available for all shell
1061invocations.
1062The function search path, \fB\s-1FPATH\s+1\fP, described later,
1063eliminated the primary need for this capability and it was
1064removed because the high performance cost was no longer
1065deemed acceptable.
1066.H 1 "PROGRAMMING LANGUAGE"
1067The KornShell vastly extends the set of applications that
1068can be implemented efficiently at the shell level.
1069It does this by providing simple yet powerful mechanisms
1070to perform arithmetic, pattern matching,
1071substring generation,
1072and arrays.
1073Users can write applications as separate functions that can
1074be defined in the same file or in a library of functions
1075stored in a directory and loaded on demand.
1076.H 2 "String Processing"
1077The shell is primarily a string processing language.
1078By default, variables hold variable length strings.
1079There are no limits to the length of strings.  Storage
1080management is handled by the shell automatically.
1081Declarations are not required.
1082With most programming languages, string constants are designated
1083by enclosing characters in single quotes or double quotes.
1084Since most of the words in the language are strings, the shell
1085requires quotes only when a string contains characters that
1086are normally processed specially by the shell, but their
1087literal meaning is intended.
1088However, since the shell is a string processing language,
1089and some characters can occur as literals and as language metacharacters,
1090quoting is an important part of the language.
1091.P
1092There are four quoting mechanisms in \f5ksh\fP.
1093The simplest is to enclose a sequence of characters inside single quotes.
1094All characters between a pair of single quotes have their literal meaning;
1095the single quote itself cannot appear.
1096A
1097.B $
1098immediately preceding
1099a single quoted string
1100causes all the characters until the matching single quote
1101to be interpreted as an ANSI-C language string.
1102Thus, \f5'\en'\fP represents characters \f5\e\fP and
1103\f5n\fP, whereas, \f5$'\en'\fP
1104represents the new-line character.
1105Double quoted strings remove the special meaning of all characters
1106except
1107.BR $ ,
1108.BR \(ga ,
1109and
1110.BR \e ,
1111so that parameter expansion and command substitution (defined below)
1112are performed.
1113The final mechanism for quoting a character is by preceding it with the
1114escape character
1115.BR \e\^ .
1116This mechanism works outside of quoted strings and for the characters
1117.BR $ ,
1118.BR \(ga ,
1119\fB"\fP,
1120and
1121.B \e
1122in double quoted strings.
1123.P
1124Variables are designated by
1125one or more
1126strings of alphanumeric
1127characters beginning with an alphabetic character
1128separated by a \fB\s+2.\s-2\fP.
1129Upper and lower case characters are distinct, so that the variable
1130.B A
1131and
1132.B a
1133are names of different variables.
1134There is no
1135limit to the length of the name of a variable.
1136You do not have to declare variables.
1137You can assign a value to a variable by writing the name of the
1138variable, followed by an equal sign, followed by a character string
1139that represents its value.
1140To create a variable whose name
1141contains a \fB\s+2.\s-2\fP,
1142the variable whose name consists of
1143the characters before the last \fB\s+2.\s-2\fP
1144must already exist.
1145You reference a variable by
1146putting the name inside curly braces and
1147preceding the braces with a dollar sign.
1148The braces may be omitted when the name
1149is alphanumeric.
1150If \f5x\fP and \f5y\fP
1151are two shell variables, then
1152to define a new variable,
1153\f5z\fP,
1154whose value is
1155the concatenation of the values of
1156\f5x\fP and \f5y\fP,
1157you just say
1158\f5z=$x$y\fP.
1159It is that easy.
1160.P
1161The
1162.B $
1163can be thought of as meaning
1164"value of."
1165You can also capture the output of any command with the notation
1166.BI $( command ) .
1167This is referred to as command substitution.
1168For example,
1169\f5x=$(date)\fP
1170assigns the output from the \f5date\fP
1171command to the variable \f5x\fP.
1172Command substitution in the
1173Bourne shell is denoted by enclosing the command between
1174backquotes,
1175(\fB\(ga\^\(ga\fP).
1176This notation
1177suffers from some
1178complicated quoting rules.
1179Thus, it is hard to write \f5sed\fP
1180patterns which contains back slashes within command substitution.
1181Putting the pattern in single quotes
1182is of little help.
1183\f5ksh\fP accepts the Bourne shell command substitution syntax
1184for backward compatibility.
1185The
1186.BI $( command )
1187notation allows
1188the \fIcommand\fP itself to contain quoted strings even if the substitution
1189occurs within double quotes. Nesting is legal.
1190.P
1191The special command substitution of the form
1192\f5$(cat\ file)\fP
1193can be replaced by
1194\f5$(<\ file)\fP,
1195which is faster because
1196the \f5cat\fP
1197command doesn't have to run.
1198.H 2 "Shell Parameters and Variables"
1199.P
1200There are three types of parameters used by \f5ksh\fP,
1201special parameters, positional parameters, and named
1202parameters which are called variables.
1203\f5ksh\fP defines the same special parameters,
1204.BR 0 ,
1205.BR * ,
1206.BR @ ,
1207.BR # ,
1208.BR ? ,
1209.BR $ ,
1210.BR ! ,
1211and
1212.BR \- ,
1213as in the Bourne shell.
1214.P
1215Positional parameters are set when the shell is invoked,
1216as arguments to the \f5set\fP built-in,
1217and by calls to functions (see below) and \fB\s+2.\s-2\fP
1218procedures.
1219They are named by numbers starting at 1.
1220.P
1221The third type of parameter is a variable.
1222As mentioned earlier,
1223\f5ksh\fP uses variables whose names
1224consist of one or more
1225alpha-numeric strings separated by a \fB\s+2.\s-2\fP.
1226There is no need to specify the
1227.I type
1228of a variable in the shell because, by default,
1229variables store strings of arbitrary length
1230and values will automatically be converted to numbers
1231when used in an arithmetic context.
1232However, \f5ksh\fP variables
1233can have one or more
1234.I attributes
1235that control the internal representation of the variable,
1236the way the variable is printed, and its access or
1237scope.
1238In addition,
1239\f5ksh\fP
1240allows variables to represent arrays of values
1241and references to other variables.
1242The \f5typeset\fP
1243built-in command of \f5ksh\fP
1244assigns attributes to variables.
1245Two of the attributes,
1246.I readonly
1247and
1248.IR export ,
1249are available in the Bourne shell.
1250Most of the remaining attributes are discussed here.
1251The complete list of attributes appears in the manual.
1252The \f5unset\fP
1253built-in of \f5ksh\fP removes
1254values and attributes of variables.
1255When a variable is exported, certain of its attributes are also exported.
1256.P
1257Whenever a value is assigned to a variable,
1258the value is transformed according to the attributes of the variable.
1259Changing the attribute of a variable can change its value.
1260The attributes
1261.B \-L
1262and
1263.B \-R
1264are for left and right field justification respectively.
1265They are useful for aligning columns in a report.
1266For each of these attributes, a width can be defined explicitly or else
1267it is defined the first time an assignment is made to the variable.
1268Each assignment causes justification of the field, truncating
1269if necessary.
1270Assignment to fixed sized variables
1271provides one way to generate a substring consisting of
1272a fixed number of characters from
1273the beginning or end of a string.
1274Other methods are discussed later.
1275.P
1276The attributes
1277.B \-u
1278and
1279.B \-l
1280are used for upper case and lower case
1281formatting, respectively.
1282Since it makes no sense to have both attributes on simultaneously,
1283turning on either of these attributes turns the other off.
1284The following script,
1285using \f5read\fP and \f5print\fP which are described later,
1286provides an example of the use of shell variables
1287with attributes.
1288This script reads a file of lines each consisting of five fields separated by
1289.B :
1290and prints fields 4 and 2 in upper case in columns 1-15, left justified,
1291and columns 20-25 right-justified respectively.
1292.sp
1293.nf
1294.in .5i
1295.ta 3.4i
1296\f5typeset -uL15 f4                # 15 character left justified
1297typeset -uR6 f2                 # 6 character right justified
1298IFS=:                           # set field separator to :
1299while   read -r f1 f2 f3 f4 f5  # read line, split into fields
1300do      print -r -- "$f4  $f2"  # print fields 4 and 2
1301done\fP
1302.fi
1303.ta
1304.in
1305.sp
1306.P
1307The
1308.BR \-i ,
1309.BR \-E ,
1310and
1311.BR \-F ,
1312attributes are used to represent numbers.
1313Each can be followed by a decimal number.
1314The
1315.B \-i
1316attribute causes the value to be represented as an integer and it
1317can be followed by a number representing the numeric base when expanding
1318its value.
1319Whenever a value is assigned to an integer variable, it is evaluated
1320as an arithmetic expression
1321and then truncated to an integer.
1322.P
1323The
1324.B \-E
1325attribute causes the value to be represented in scientific
1326notation whenever its value is expanded.  The number following the
1327.B \-E
1328determines the number of significant figures, and defaults to 6.
1329The
1330.B \-F
1331attribute causes the value to be represented with a fixed number
1332of places after the decimal point.
1333Assignments to variables with the
1334.B \-E
1335or
1336.B \-F
1337attributes cause the evaluation of the right hand side of the assignment.
1338.P
1339\f5ksh\fP allows one-dimensional
1340.I arrays
1341in addition to simple variables.
1342There are two types of arrays; associative arrays
1343and indexed arrays.
1344The subscript for an associative array is an arbitrary
1345string, whereas the subscript for an indexed array is
1346an arithmetic expression that is evaluated to yield an integer
1347index.
1348Any variable can become an indexed array
1349by referring to it with
1350an integer
1351.IR subscript .
1352All elements of an array need not exist.
1353Subscripts for arrays
1354must evaluate to an
1355integer between 0 and some maximum value, otherwise
1356an error results.
1357The maximum value may vary from one machine to another but
1358is at least 4095.
1359Evaluation of subscripts is described in
1360the next section.
1361Attributes apply to the whole array.
1362.P
1363Assignments to array variables can be made to individual elements
1364via parameter
1365assignment commands or the
1366.B typeset
1367built-in.
1368Additionally, values can be assigned sequentially with
1369compound assignment as described below, or by the
1370.B \-A
1371.I name
1372option of the \f5set\fP command.
1373Referencing of subscripted variables requires the character
1374.BR $ ,
1375but also requires braces around the array element name.
1376The braces are needed to avoid conflicts with the
1377file name generation mechanism.
1378The form of any array element reference is:
1379.ce
1380.BI ${ name [ subscript ]}
1381Subscript values of
1382.B *
1383and
1384.B @
1385can be used to generate all elements of an array,
1386as they are used for expansion of positional parameters.
1387The list of currently defined subscripts for a given
1388variable can be generated with
1389.BI ${! name [@]} ,
1390or
1391.BI ${! name [*]} .
1392.P
1393The
1394.B \-n
1395or
1396.I nameref
1397attribute causes the variable to be treated
1398as a reference to the variable defined by its value.
1399Once this attribute is set, all references to this variable
1400become references to the variable named by the value
1401of this variable.
1402For example, if \f5foo=bar\fP, then setting the reference
1403attribute on \f5foo\fP will cause all subsequent references
1404to \f5foo\fP to behave as the variable whose name is \f5$foo\fP
1405was referenced, which in this case is the variable \f5bar\fP.
1406Unsetting this attribute breaks the association.
1407Reference variables are usually used inside functions whose
1408arguments are the names of shell variables.
1409The names for reference variables cannot contain a \fB\s+2.\s-2\fP.
1410Whenever a shell variable is referenced, the portion of the
1411variable up to the first \fB\s+2.\s-2\fP
1412is checked to see whether it matches the name of a reference
1413variable.
1414If it does, then the name of the variable actually used
1415consists of the concatenation of the name of the variable
1416defined by the reference plus the remaining portion of the
1417original variable name.
1418For example, using the predefined alias, \f5alias\ nameref='typeset\ -n'\fP,
1419.sp
1420.nf
1421.in .5i
1422.ta 3.4i
1423\f5\^.bar.home.bam="hello world"
1424nameref foo=.bar.home
1425print ${foo.bam}
1426\fBhello world\fP\fP
1427.fi
1428.ta
1429.in
1430.sp
1431.H 2 "Compound Assignment"
1432Compound assignments are used to assign values to arrays
1433and compound data structures.
1434The syntax for a compound assignment is
1435.IB name =( assignment-list )
1436where
1437\fIname\fP
1438is the name of the variable to which you want to assign values.
1439No space is permitted between the variable name and the \fB=\fP
1440but can appear between the \fB=\fP and the open parenthesis.
1441New-lines can appear between the parentheses.
1442.P
1443The \fIassignment-list\fP can be in several different forms
1444yielding different results.
1445If \fIassignment-list\fP is simply a list of words, then
1446the words are processed as they are with the \f5for\fP command
1447and assigned sequentially as an indexed array.
1448For example,
1449.ce
1450\f5foo=( * )\fP
1451creates an indexed array \f5foo\fP and assigns the
1452file names in the current directory to each index starting
1453at zero.
1454.P
1455The second form for \fIassignment-list\fP is a list of assignments
1456of the special form \fB[\fP\fIword\fP\fB]=\fP\fIword\fP.
1457No space is permitted before or after the \fB=\fP.
1458In this case, the variable given by \fIname\fP becomes
1459an associative array with the given arguments as subscripts.
1460For example,
1461.ce
1462\f5bar=( [color]=red [shape]=box )\fP
1463creates an associate array named \f5bar\fP whose
1464subscripts are \f5color\fP and \f5shape\fP.
1465.P
1466The third form for \fIassignment-list\fP is a list of
1467normal assignments, including compound assignments.
1468These assignments cause sub-variables to be assigned
1469corresponding to the given assignments.
1470In addition to assignments, the \fIassignment-list\fP
1471can contain \f5typeset\fP commands.
1472In addition to creating sub-variables,
1473the effect of a compound assignment is to make
1474the value of the original variable be a parenthesized
1475assignment list of its components.
1476For example, the assignment
1477.sp
1478.nf
1479.in .5i
1480.ta 3.4i
1481\f5foo=(
1482        left=bar
1483        typeset -i count=3
1484        point=(
1485                x=50
1486                y=60
1487        )
1488        colors=( red green yellow )
1489        right=bam
1490) \fP
1491.ta
1492.in
1493.fi
1494.sp
1495is equivalent to the assignments
1496.sp
1497.nf
1498.in .5i
1499.ta 3.4i
1500\f5foo.left=bar
1501foo.count=3
1502foo.point.x=50
1503foo.point.y=60
1504foo.colors=( red green yellow )
1505foo.right=bam\fP
1506.ta
1507.in
1508.fi
1509.sp
1510In addition, the value of \f5"$foo"\fP is
1511.sp
1512.nf
1513.in .5i
1514.ta 3.4i
1515\f5(
1516        colors=( red green yellow )
1517        left=bar
1518        typeset -i count=3
1519        point=(
1520                y=60
1521                x=50
1522        )
1523        right=bam
1524)\fP
1525.ta
1526.in
1527.fi
1528.sp
1529.H 2 "Substring Generation"
1530The expansion of a variable or parameter can be modified so that
1531only a portion of the value results.
1532It is often necessary to extract a portion of a shell variable or
1533a portion of an array.
1534There are several parameter expansion operators that can do this.
1535One method to generate a substring is with an expansion of
1536the form \fB${\fP\fIname\fP\fB:\fP\fIoffset\fP\fB:\fP\fIlength\fP\fB}\fP
1537where \fIoffset\^\fP is an arithmetic expression that defines the
1538offset of the first character starting from 0, and
1539\fIlength\^\fP is an arithmetic expression that defines the
1540length of the substring.
1541If
1542.BI : length\^
1543is omitted,
1544the length of the value of
1545.I name\^
1546starting at
1547.I offset\^
1548is used.
1549The
1550.BI : offset : length
1551operators can also be applied to array expansions and to parameters
1552.B *
1553and
1554.B @
1555to generate portions of an array.
1556For example, the expansion, \fB${\fP\fIname\fP\fB[@]:\fP\fIoffset\fP\fB:\fP\fIlength\fP\fB}\fP, yields up to \fIlength\fP elements of the array \fIname\fP
1557starting at the element \fIoffset\fP.
1558.P
1559The other parameter expansion modifiers use shell patterns
1560to describe portions of the string to modify and delete.
1561A description of shell patterns is contained below.
1562When these
1563modifiers are applied to special parameters
1564.B @
1565and
1566.B *
1567or to array parameters given as
1568\fIname\fP\fB[@]\fP or \fIname\fP\fB[*]\fP,
1569the operation is performed on each element.
1570There are four parameter expansion modifiers that
1571strip off leading and trailing substrings
1572during parameter expansion
1573by removing the characters matching a given pattern.
1574An expansion of
1575the form \fB${\fP\fIname\fP\fB#\fP\fIpattern\fP\fB}\fP
1576causes the smallest matching prefix of the value of
1577.I name\^
1578to be removed.
1579The largest prefix matching
1580.I pattern\^
1581is removed by using
1582.B ##
1583instead of
1584.BR # .
1585Similarly,
1586an expansion of
1587the form \fB${\fP\fIname\fP\fB%\fP\fIpattern\fP\fB}\fP
1588causes the smallest matching substring at the end of
1589.I name\^
1590to be removed.
1591Again, using
1592.B %%
1593instead of
1594.BR % ,
1595causes the largest matching trailing substring to be deleted.
1596For example, if the shell variable
1597.B file
1598has value
1599.BR foo.c ,
1600then the expression
1601.B ${file%.c}.o
1602has value
1603.BR foo.o .
1604.P
1605The value of an expansion can be changed by
1606specifying a pattern that matches the part that needs to be changed
1607after the
1608the parameter expansion modifier
1609.BR / .
1610An expansion of the form
1611\fB${\fP\fIname\fP\fB/\fP\fIpattern\fP\fB/\fP\fIstring\fP\fB}\fP
1612replaces the first match of \fIpattern\fP with
1613the value of variable \fIname\fP to \fIstring\fP.
1614The second
1615.B /
1616is not necessary when \fIstring\fP is null.
1617The expansion
1618\fB${\fP\fIname\fP\fB//\fP\fIpattern\fP\fB/\fP\fIstring\fP\fB}\fP
1619changes all occurrences of the \fIpattern\fP into \fIstring\fP.
1620The parameter expansion modifiers
1621.B /#
1622and
1623.B /%
1624cause the matching pattern to be anchored to the beginning and
1625end respectively.
1626.P
1627Finally, there are parameter expansion modifiers that yield
1628the name of the variable, the string length of the value, or the number
1629of elements of an array.
1630\fB${!\fP\fIname\fP\fB}\fP
1631yields the name of the variable which will be \fIname\fP itself
1632except when \fIname\fP is a reference variable.  In this case
1633it will yield the name of the variable it refers to.
1634When applied to an array variable,
1635\fB${!\fP\fIname\fP\fB[@]}\fP and
1636\fB${!\fP\fIname\fP\fB[*]}\fP
1637generate the names of all subscripts.
1638\fB${#\fP\fIname\fP\fB}\fP
1639will be the length in bytes of
1640\fB$\fP\fIname\fP.
1641For an array variable
1642\fB${#\fP\fIname\fP\fB[*]}\fP
1643gives the number of elements in the array.
1644.H 2 "Arithmetic Evaluation"
1645.P
1646For the most part, the shell is a string processing
1647language.  However, the need for arithmetic has
1648long been obvious.
1649Many of the characters that are special to the
1650Bourne shell are needed as arithmetic operators.
1651To make arithmetic easy to use, and to maintain
1652compatibility with the Bourne shell, \f5ksh\fP uses matching
1653.B ((
1654and
1655.B ))
1656to delineate arithmetic expressions.
1657While single parentheses might have been
1658more desirable, these already mean
1659.I subshell\^
1660so that another notation was required.
1661The arithmetic expression
1662inside the double parentheses
1663follows the same syntax, associativity and precedence
1664as the ANSI-C\*(Rf
1665.RS
1666American National Standard for Information Systems \- Programming
1667Language \- C, ANSI X3.159-1989.
1668.RF
1669programming language.
1670The characters between the matching double parentheses
1671are processed with the same rules used for double
1672quotes so that spaces can be used to aid readability
1673without additional quoting.
1674.P
1675All arithmetic evaluations are performed using
1676double precision floating point arithmetic.
1677Floating point constants follow the same rules as
1678the ANSI-C programming language.
1679Integer arithmetic constants are written as
1680.ce
1681.IB base # number,
1682where
1683.I base\^
1684is a decimal integer between
1685two and sixty-four and
1686.I number\^
1687is any non-negative number.
1688Base ten is used
1689when no base is specified.
1690The digits are represented by the characters
1691.BR 0-9a-zA-Z_@ .
1692For bases less than or equal to 36,
1693upper and lower case characters can
1694be used interchangeably to represent the digits
1695from 10 thru 35.
1696.P
1697Arithmetic expressions are made from constants,
1698variables, and operators.
1699Parentheses may be used for grouping.
1700The contents inside the double parentheses
1701are processed with the same expansions as occurs in a double quoted string,
1702so that all
1703.B $
1704expansions are performed before the expression is evaluated.
1705However, there is usually no need to use the
1706.B $
1707to get the value of a variable
1708because the arithmetic evaluator replaces the name of the variable
1709by its value within an arithmetic expression.
1710The
1711.B $
1712cannot be used when the variable is the subject of assignment
1713or an increment operation.
1714As a rule it is better not to use
1715.B $
1716in front of variables in an arithmetic expression.
1717.P
1718An arithmetic command of the form
1719.B
1720(( ... ))
1721.R
1722is a command that evaluates the enclosed arithmetic expression.
1723For example, the command
1724.ce
1725\f5(( x++ ))\fP
1726can be used to
1727increment the variable \f5x\fP,
1728assuming that \f5x\fP contains some numerical value.
1729The arithmetic command is true (return value 0), when the resulting
1730expression is non-zero, and false (return value 1) when the
1731expression evaluates to zero.
1732This makes the command easy to use with the \f5if\fP and \f5while\fP
1733compound commands.
1734.P
1735The \f5for\fP compound command
1736has been extended for use in arithmetic contexts.
1737The syntax,
1738.ce
1739\f5for\fP \fB((\fP \fIexpr1\fP\fB;\fP \fIexpr2\fP \fB;\fP \fIexpr3 \fP\fB))\fP
1740can be used as the first line of a \f5for\fP loop with the same semantics
1741as the \f5for\fP statement in the ANSI-C programming language.
1742.P
1743Arithmetic evaluations can also be performed as part of the evaluation
1744of a command line.
1745The syntax
1746.B
1747$((\ ...\ ))
1748.R
1749expands to the value of the enclosed arithmetic expression.
1750This expansion can occur wherever parameter expansion is performed.
1751For example using the \f5ksh\fP command \f5print\fP (described
1752later)
1753.ce
1754\f5print $((2+2))\fP
1755prints the number 4.
1756.P
1757The following script prints the first
1758.I n
1759lines of its standard input onto its standard output,
1760where
1761.I n
1762can be supplied as an optional argument whose default value is 20.
1763.sp
1764.nf
1765.in .5i
1766.ta 4i
1767\f5integer n=${1-20}                       # set n
1768while   (( n-- >=0 )) && read -r line   # at most n lines
1769do      print -r -- "$line"
1770done\fP
1771.fi
1772.ta
1773.in
1774.sp
1775.H 2 "Shell Expansions"
1776.P
1777The commands you enter from the terminal or from a script
1778are divided into words and each word undergoes several
1779expansions to generate the command name and its arguments.
1780This is done in two phases.
1781The first phase recognizes reserved words, spaces and operators
1782to decide where command boundaries lie.
1783Alias substitutions take place during this phase.
1784The second phase performs expansions in the following order:
1785.BL
1786.LI
1787Tilde substitution,
1788parameter expansion,
1789arithmetic expansion,
1790and command substitution
1791are performed from left to right.
1792The option
1793.B \-u
1794or
1795.BR nounset ,
1796will cause an error to occur when any variable
1797that is not set is expanded.
1798.LI
1799The characters that result from parameter expansion and
1800command substitution above are checked with the characters
1801in the
1802\fB\s-1IFS\s+1\fP variable
1803for possible
1804field splitting.
1805(See a description of \f5read\fP below to see how
1806\fB\s-1IFS\s+1\fP is used.)
1807Setting
1808\fB\s-1IFS\s+1\fP to a null
1809value causes field splitting to be skipped.
1810.LI
1811Pathname generation (as described below)
1812is performed on each of the fields.
1813Any field that doesn't match a pathname is left alone.
1814The option,
1815.B \-f
1816or
1817.BR noglob ,
1818is used to disable pathname generation.
1819.LE
1820.H 2 "Pattern Matching"
1821The shell is primarily a string processing language and uses
1822patterns for matching file names as well as for matching strings.
1823The characters
1824.BR ? ,
1825.BR * ,
1826and
1827.B [
1828are processed specially
1829by the shell when not quoted.
1830These characters are used to form patterns that
1831match strings.
1832Patterns are used by the shell to match pathnames,
1833to specify substrings,
1834and for
1835.B case
1836commands.
1837The character
1838.B ?
1839matches any one character.
1840The character
1841.B *
1842matches zero or more characters.
1843The character sequence
1844.BR [ ... ]
1845defines a character class
1846that matches any character contained within
1847.BR [\^] .
1848A range of characters can be specified by putting a
1849.B \-
1850between the first and last character of the range.
1851An exclamation mark,
1852.BR ! ,
1853immediately after the
1854.BR [ ,
1855means match all characters except the characters specified.
1856For example, the pattern
1857\f5a?c*.[!a-z]\fP
1858matches any string beginning with an
1859.BR a ,
1860whose third character is a
1861.BR c ,
1862and that ends in
1863.B .
1864(dot) followed by any character except the lower case letters,
1865.BR a\-z .
1866The sequence \f5[:alpha:]\fP
1867inside a character class, matches any set of characters in
1868the ANSI-C
1869.B alpha
1870class.
1871Similarly, \f5[:\fP\fIclass\fP\f5:]\fP matches
1872each of the characters in the given \fIclass\fP
1873for all the ANSI-C character classes.
1874For example, \f5[[:alnum:]_]\fP
1875matches any alpha-numeric character or the character
1876.BR _ .
1877.P
1878\f5ksh\fP
1879treats
1880strings of the form
1881.BI ( pattern-list
1882.BR ) ,
1883where
1884.I pattern-list
1885is a list of one or more patterns separated by a
1886.BR \(bv ,
1887specially when preceded by
1888.BR * ,
1889.BR ? ,
1890.BR + ,
1891.BR @ ,
1892or
1893.BR ! .
1894A
1895.B ?
1896preceding
1897.BI ( pattern-list )
1898means that the pattern list enclosed in
1899.B (\^)
1900is optional.
1901An
1902.BI @( pattern-list )
1903matches any pattern in the list of patterns enclosed in
1904.BR () .
1905A
1906.BI *( pattern-list )
1907matches any string that contains zero or more of each of the enclosed
1908patterns,
1909whereas
1910.BI +( pattern-list )
1911requires a match of one or more of any of the given patterns.
1912For instance, the pattern
1913.B +([0\-9])?(.)
1914matches one or more digits optionally followed by a
1915.BR . (dot).
1916A
1917.BI !( pattern-list )
1918matches anything except any of the given patterns.
1919For example,
1920\f5print\ !(*.o)\fP
1921displays all file names in the current directory that do not end in
1922.BR .o .
1923.P
1924When patterns are used to generate pathnames when expanding
1925commands several other rules apply.
1926A separate match is made
1927for each file name component of the pathname.
1928Read permission is required for
1929any portion of the pathname that contains any special
1930pattern character.
1931Search permission is required for every component except
1932possibly the last.
1933.P
1934By default,
1935file names in each directory that begin with \fB\s+2.\s-2\fP
1936are skipped when performing a match.
1937If the pattern to be matched starts with a leading \fB\s+2.\s-2\fP,
1938then only files beginning with a \fB\s+2.\s-2\fP,
1939are examined when reading each directory to find matching files.
1940If the
1941\fB\s-1FIGNORE\s+1\fP variable
1942is set,
1943then only files that do not match this pattern
1944are considered.
1945This overrides the special meaning of \fB\s+2.\s-2\fP
1946in a pattern and in a file name.
1947.P
1948If the
1949.B markdirs
1950option is set,
1951each matching pathname that is the name
1952of a directory has a trailing
1953.B /
1954appended to the name.
1955.P
1956.H 2 "Conditional Expressions"
1957The Bourne shell uses the \f5test\fP
1958command, or the equivalent \f5[\fP
1959command, to test files for attributes
1960and to compare strings or numbers.
1961The problem with \f5test\fP
1962is that the shell has expanded the words of the \f5test\fP
1963command and
1964split them into arguments before \f5test\fP begins execution.
1965\f5test\fP
1966cannot distinguish between operators and operands.
1967In most cases
1968\f5test\ "$1"\fP
1969will test whether argument 1 is non-null.
1970However,
1971if argument 1 is
1972.BR \-f ,
1973then \f5test\fP
1974will treat
1975.B \-f
1976as an operator and
1977yield a syntax error.
1978One of the most frequent errors with
1979\f5test\fP
1980occurs when its operands are not within double quotes.
1981In this case, the argument may expand to more than a single
1982argument or to no argument at all.  In either case this
1983will likely cause a syntax error.
1984What makes this most insidious is that these errors are frequently
1985data dependent.  A script that appears to run correctly may abort
1986if given unexpected data.
1987.P
1988To get around these problems,
1989\f5ksh\fP
1990has a compound command for conditional expression testing
1991as part of the language.
1992The reserved words
1993.B [[
1994and
1995.B ]]
1996delimit the range of the command.
1997Because they are reserved words, not operator characters,
1998they require spaces to separate them
1999from arguments.
2000The words between
2001.B [[
2002and
2003.B ]]
2004are not processed for field splitting or for pathname generation.
2005In addition, since \f5ksh\fP
2006determines the operators before parameter expansion,
2007expansions that yield no argument cause no problem.
2008The operators within
2009.BR [[ ... ]]
2010are almost the same as those for the \f5test\fP
2011command.
2012All unary operators are of the form
2013.BI \- letter
2014and are followed by a single operand.
2015Instead of
2016.B \-a
2017and
2018.BR \-o ,
2019.BR [[ ... ]]
2020uses
2021.B &&
2022and
2023.B \(bv\(bv
2024to indicate "and" and "or".
2025Parentheses are used without quoting for grouping.
2026.P
2027The right hand side of the string comparison operators
2028.B ==
2029and
2030.B !=
2031takes a pattern and tests whether the left hand operand
2032matches this pattern.  Quoting the pattern results
2033is a string comparison rather than the pattern match.
2034The operators
2035.B <
2036and
2037.B >
2038within
2039.BR [[ ... ]]
2040designate lexicographical comparison.
2041.P
2042In addition there are several other new comparison primitives.
2043The binary operators
2044.B \-ot
2045and
2046.B \-nt
2047compare the modification times
2048of two files to see which file is
2049.I "older than"
2050or
2051.I "newer than"
2052the other.
2053The binary operator
2054.B \-ef
2055tests whether two files
2056have the same device and i-node number,
2057i.\ e., a link to the same file.
2058.P
2059The unary operator
2060.B \-L
2061returns true if its operand is a symbolic link.
2062The unary operator
2063.B \-O
2064(\fB\-G\fP)
2065returns true if the owner (or group) of the file operand matches
2066that of the caller.
2067The unary operator
2068.B \-o
2069returns true when its operand is the name of an option that is
2070currently on.
2071.P
2072The following script illustrates some of the uses of
2073.BR [[ ... ]] .
2074The reference manual contains the complete list of operators.
2075.sp
2076.nf
2077.in .5i
2078.ta 4i
2079\f5for i
2080do      # execute foo for numeric directory
2081        if      [[ \-d $i && $i == +([0\-9]) ]]
2082        then    foo
2083        # otherwise if writable or executable file and not mine
2084        elif    [[ (\-w $i\(bv\(bv\-x $i) && ! \-O $i ]]
2085        then    bar
2086        fi
2087done\fP
2088.fi
2089.ta
2090.in
2091.sp
2092.H 2 "Input and Output"
2093\f5ksh\fP has
2094extended I/O capabilities to enhance the
2095use of the shell as a programming language.
2096As with the Bourne shell,
2097you use the I/O redirection operator,
2098.BR < ,
2099to control where input comes from,
2100and the I/O redirection operator,
2101.BR > ,
2102to control where output goes to.
2103Each of these operators can be preceded with a single digit that
2104specifies a file unit number to associate with the file stream.
2105Ordinarily you specify these I/O redirection operators with a specific
2106command to which it applies.
2107However, if you specify I/O redirections with the \f5exec\fP
2108command,
2109and don't specify arguments to \f5exec\fP,
2110then the I/O redirection applies to the current program.
2111For example, the command
2112\f5exec\ <\ foobar\fP
2113opens file \f5foobar\fP
2114for reading.
2115The \f5exec\fP
2116command is also used to close files.
2117A file descriptor unit can be opened as a copy of an existing
2118file descriptor unit by using either of the
2119.B <&
2120or
2121.B >&
2122operators and putting the file descriptor unit of the original file
2123after the
2124.BR & .
2125Thus, \f52>&1\fP means open standard error (file descriptor 2)
2126as a copy of standard output (file descriptor 1).
2127A file descriptor value of
2128.B \-
2129after the
2130.B &
2131indicates that the file should be closed.
2132To close file unit 5, specify
2133\f5exec\ 5<&-\fP.
2134There are two additional redirection operators with \f5ksh\fP
2135and the POSIX shell that are not part of the Bourne shell.
2136The
2137.B >|
2138operator overrides the effect of the
2139.B noclobber
2140option described earlier.
2141The
2142.B <\^>
2143operator causes a file to be opened for both reading and writing.
2144.P
2145\f5ksh\fP recognizes certain pathnames and treats them
2146specially.
2147Pathnames of the form
2148.BI /dev/fd/ n\^
2149are treated as equivalent to the file defined by file descriptor
2150.IR n .
2151These name can be used as the script argument to \f5ksh\fP
2152and in conditional testing as described above.
2153On underlying systems that support
2154.B /dev/fd
2155in the file system, these names can be passed to other commands.
2156Pathnames of the form
2157.BI /dev/tcp/ hostid / port
2158and
2159.BI /dev/udp/ hostid / port
2160can be used to create
2161.B tcp
2162and
2163.B udp
2164connections to services given by the
2165.I hostid\^
2166number and
2167.I port\^
2168number.
2169The
2170.I hostid\^
2171cannot use symbolic values. In practice these
2172numbers are typically generated by command substitution.
2173For example,
2174\f5exec\ 5>\ /dev/tcp/$(service\ name)\fP
2175would open file descriptor 5 for sending messages
2176to hostid and port number defined by the output of \f5service\ name\fP.
2177.P
2178The Bourne shell has a built-in command \f5read\fP
2179for reading lines from standard input (file descriptor 0)
2180and splitting it into fields based on the value of the
2181.B \s-1IFS\s+1
2182variable, and a command \f5echo\fP
2183to write strings to standard output.
2184(On some systems, \f5echo\fP
2185is not a built-in command and incurs considerable overhead to use.)
2186Unfortunately, neither of these commands
2187is able to perform some very basic tasks.
2188For example.
2189with the Bourne shell,
2190the \f5read\fP
2191built-in cannot read a single line that ends in
2192.BR \e .
2193With \f5ksh\fP
2194the \f5read\fP
2195built-in has a
2196.B \-r
2197option to remove the special meaning for
2198.B \e
2199which allows it to be
2200treated as a regular
2201character rather than the line continuation character.
2202With the Bourne shell,
2203there is no simple way to have more than one file open
2204at any time for reading.
2205\f5ksh\fP has options on the \f5read\fP
2206command to specify the file
2207descriptor for the input.
2208The fields that are read from a line can be stored into an indexed
2209array with the
2210.B \-A
2211option to read.
2212This allows a line to be split into an arbitrary number of fields.
2213.P
2214The way the Bourne shell uses the
2215\fB\s-1IFS\s+1\fP variable to
2216split lines into fields greatly limits its utility.
2217Often data files consist of lines that use a character such
2218as
2219.B :
2220to delimit fields with two adjacent delimiters that denote
2221a null field.
2222The Bourne shell treats adjacent delimiters as a single
2223field delimiter.
2224With \f5ksh\fP,
2225delimiters that are considered white space characters
2226have the behavior of the Bourne shell, but other
2227adjacent delimiters separate
2228null fields.
2229.P
2230The \f5read\fP command is often used in scripts that interact
2231with the user by prompting the user and then requesting some
2232input.
2233With the Bourne shell two commands are needed; one to
2234prompt the user, the other to read the reply.
2235\f5ksh\fP allows these two commands to be combined.
2236The first argument of the \f5read\fP
2237command can be followed by a
2238.B ?
2239and a prompt string which is used whenever the input
2240device is a terminal.
2241Because the prompt is associated with the \f5read\fP built-in,
2242the built-in command line editors will be able to re-output
2243the prompt whenever the line needs to be refreshed when
2244reading from a terminal device.
2245.P
2246With the Bourne shell,
2247there is no way to set a time limit for waiting for the user
2248response to read.
2249The
2250.B \-t
2251option to \f5read\fP takes a floating
2252point argument that gives the time in seconds,
2253or fractions of seconds that the shell should wait for a reply.
2254.P
2255The version of the \f5echo\fP command in System V
2256treats certain sequences beginning with
2257.B \e
2258as control sequences.
2259This makes it hard to output strings without interpretation.
2260Most BSD derived systems do not interpret
2261.B \e
2262control sequences.
2263Unfortunately, the BSD versions of \f5echo\fP accepts a
2264.B \-n
2265option to prevent a trailing new-line, but has no way to
2266cause the string
2267.B \-n
2268to be printed.
2269Neither of these versions is adequate. Also, because they
2270are incompatible, it is very hard to write portable shell scripts
2271using \f5echo\fP.
2272The \f5ksh\fP built-in, \f5print\fP,
2273outputs characters to the terminal or to a file and
2274subsumes the functions of all versions of \f5echo\fP.
2275Ordinarily, escape sequences in arguments beginning with
2276.B \e
2277are processed the same as for the System V \f5echo\fP command.
2278However \f5print\fP follows the standard conventions for
2279options and has options that make \f5print\fP very versatile.
2280The
2281.B \-r
2282option can be used to output the arguments without any special meaning.
2283The
2284.B \-n
2285option can be used here to suppress the trailing new-line
2286that is ordinarily appended.
2287As with \f5read\fP, it is possible to specify the file descriptor number
2288as an option to the command to avoid having to use
2289redirection operators with each occurrence of the command.
2290.P
2291The IEEE POSIX shell and utilities standard committee was unable
2292to reconcile the differences between the System V and BSD
2293versions of \f5echo\fP.
2294They introduced a new command named \f5printf\fP
2295which takes an ANSI-C format string and a list of options
2296and outputs the strings using the ANSI-C formatting rules.
2297Since \f5ksh\fP is POSIX conforming, it accepts \f5printf\fP.
2298However, there is a
2299.B \-f
2300options to \f5print\fP that can be used to specify
2301a format string which processes the arguments the same way that
2302\f5printf\fP does.
2303.P
2304The format processing for \f5print\fP and \f5printf\fP has
2305been extended slightly.
2306There are three additional formatting directives.
2307The
2308.B %b
2309format causes the
2310.B \e
2311escape sequences to be expanded as they are with the System V \f5echo\fP
2312command.
2313The
2314.B %q
2315format causes quotes to
2316be placed on the output as required
2317so that it can be used as shell input.
2318Special characters in the output of most \f5ksh\fP built-in commands
2319and in the output from an execution trace
2320are quoted in an equivalent fashion.
2321The
2322.B %P
2323format causes an extended regular expression string to
2324be converted into a shell pattern.
2325This is useful for writing shell applications that have
2326to accept regular expressions as input.
2327Finally, the escape sequence
2328.B \e\^E
2329which expands to the terminal escape character (octal 033)
2330has been added.
2331.P
2332The shell is frequently used as a programming language for
2333interactive dialogues.
2334The
2335\f5select\fP
2336statement has been added to the language
2337to make it easier to
2338present menu selection alternatives to the
2339user and evaluate the reply.
2340The list of alternatives is numbered and put in columns.
2341A user settable prompt,
2342\fB\s-1PS3\s+1\fP,
2343is issued and if the answer is
2344a number corresponding to one of the alternatives,
2345the select loop variable is set to this value.
2346In any case, the
2347.B \s-1REPLY\s+1
2348variable is used to store the user entered reply.
2349The shell variables
2350.B \s-1LINES\s+1
2351and
2352.B \s-1COLUMNS\s+1
2353are used to control the layout of select lists.
2354.H 2 "Option Parsing"
2355The \f5getopts\fP built-in command can be used
2356to process command arguments in a manner consistent
2357with the way \f5ksh\fP does for its own built-in commands.
2358.P
2359The \f5getopts\fP built-in allows users to specify options
2360as separate arguments or to group options that do not
2361take arguments together.  Options that require arguments
2362do not require space to separate them from the option argument.
2363The
2364.B \s-1OPTARG\s+1
2365variable stores the value of the option argument
2366after finding a variable that takes an argument.
2367The
2368.B \s-1OPTIND\s+1
2369variable holds the index of the current options argument.
2370After processing options, the arguments should be
2371shifted by
2372.B \s-1OPTIND\s+1\-1
2373to make the
2374remaining arguments be \f5"$@"\fP.
2375.P
2376The \f5getopts\fP argument description allows additional
2377information to be specified along with the options
2378that is used to generate \fIusage\fP messages for
2379incorrect arguments and for the option argument \fB\-?\fP.
2380The example in the APPENDIX uses \f5getopts\fP to process
2381its arguments.
2382.H 2 "Co-process"
2383\f5ksh\fP can spawn a
2384.I co-process
2385by adding a
2386.B "|&"
2387after a command.
2388This process will be run with its standard input and its
2389standard output connected to the shell.  The built-in command \f5print\fP
2390with the
2391.B \-p
2392option will write into the standard input of this
2393process and
2394the built-in command \f5read\fP
2395with the
2396.B \-p
2397option will read from the output of this process.
2398.P
2399In addition, the I/O redirection operators \fB<&\fP and \fB>&\fP can
2400be used to move the input or output pipe of the co-process
2401to a numbered file descriptor.
2402Use \f5exec\ 3>&\ p\fP to move the input of the co-process
2403to file descriptor \fB3\fP.
2404After you have connected to file descriptor \fB3\fP, you
2405can direct the output of any command to the co-process
2406by running \fIcommand\fP\f5\ >&3\fP.
2407Also, by moving the input of the co-process to a numbered descriptor,
2408it is possible to run a second co-process.
2409The output of both co-processes will be the file descriptor
2410associated with \f5read\ -p\fP.
2411You can use \f5exec\ 4<&\ p\fP to cause the output of these
2412co-processes to go to file descriptor \fB4\fP of the shell.
2413Once you have moved the pipe to descriptor \fB4\fP, it is possible
2414to connect a server to the co-process by running \fIcommand\fP\f5\ 4<&\ p\fP
2415or to close the co-process pipe with \f5exec\ 4<&\ -\fP.
2416.H 2 "Functions"
2417.P
2418Function definitions are of the form
2419.sp
2420.in +.5i
2421.nf
2422\f5function\fP \fIname\fP
2423.br
2424.B {
2425.br
2426        any shell script
2427.br
2428.B }
2429.fi
2430.sp
2431.in
2432A function whose name contains a \fB\s+2.\s-2\fP
2433is called a \fIdiscipline\fP function.
2434The portion of the name after the last \fB\s+2.\s-2\fP
2435is the name of the discipline.
2436Discipline functions named \f5get\fP, \f5set\fP, and \f5unset\fP
2437can be assigned to any variable to intercept lookups,
2438assignments and unsetting of the variable
2439defined by the portion of the name before the last \fB\s+2.\s-2\fP.
2440Applications can create additional disciplines for variables
2441that are created as part of user defined built-ins.
2442The portion of the name before the last \fB\s+2.\s-2\fP
2443must refer to the name of an existing variable.
2444Thus, if \f5p\fP is a reference to \f5PATH\fP, then
2445the function name \f5p.get\fP and \f5PATH.get\fP
2446refer to the same function.
2447.P
2448The function is invoked either
2449by specifying
2450.I name
2451as the command name
2452and optionally following it with arguments
2453or by using it as an option to the \fB\s+2.\s-2\fP
2454built-in command.
2455Positional parameters are saved before each
2456function call and restored when completed.
2457The arguments that follow the function name on the calling
2458line become positional parameters inside the function.
2459The \f5return\fP
2460built-in can be used to cause the function to return to
2461the statement following
2462the point of invocation.
2463.P
2464Functions can also be defined with the System V notation,
2465.sp
2466.in +.5i
2467.nf
2468\fIname\fP \f5()\fP
2469.br
2470.B {
2471.br
2472        any shell script
2473.br
2474.B }
2475.fi
2476.sp
2477.in
2478Functions defined with this syntax cannot be used as the first
2479argument to a \fB\s+2.\s-2\fP procedure.
2480\f5ksh\fP accepts this notation for compatibility only.
2481There is no need to use this notation when writing
2482\f5ksh\fP scripts.
2483.P
2484Functions defined with the \f5function\fP\ \fIname\fP syntax
2485and invoked by name
2486are executed in the current shell environment
2487and can share named variables with the calling program.
2488Options, other than execution trace
2489.BR \-x ,
2490set by the calling program are
2491passed down to a function.
2492The options are
2493not shared with
2494the function so that any options set within a function are
2495restored when the function exits.
2496Traps ignored by the caller are ignored within the function
2497and cannot be enabled.
2498Traps caught by the calling program are reset to their
2499default action within the function.
2500In most instances, the default action is
2501to cause the function to terminate.
2502A trap on
2503\fB\s-1EXIT\s+1\fP
2504defined within a function executes after the function
2505completes but
2506before the caller resumes.
2507Therefore,
2508any variable assignments and
2509any options set as part of a trap action will be effective
2510after the caller resumes.
2511.P
2512By default, variables are inherited by the function and shared
2513by the calling program.
2514However,
2515for functions defined with the \f5function\fP\ \fIname\fP syntax
2516that are invoked by name,
2517environment substitutions preceding the function call
2518apply only to the scope of the function call.
2519Also, variables whose names do not contain a \fB\s+2.\s-2\fP
2520that are defined with the \f5typeset\fP
2521built-in command are local to the function that they are declared in.
2522Thus, for the function defined
2523.sp
2524.nf
2525.in .5i
2526\f5function  name
2527{
2528     typeset -i x=10
2529     let z=x+y
2530     print $z
2531}\fP
2532.fi
2533.ta
2534.in
2535.sp
2536invoked as
2537\f5y=13\ name\fP,
2538\f5x\fP and \f5y\fP
2539are local variables with respect to the function
2540\f5name\fP
2541while
2542\f5z\fP
2543is global.
2544.P
2545Functions defined with the \fIname\fP\f5()\fP syntax,
2546and functions invoked as an argument to the \fB\s+2.\s-2\fP
2547command,
2548share everything other than positional parameters with the caller.
2549Assignments that precede the call remain in effect after the
2550function completes.
2551.P
2552Alias and function names are not passed down to shell scripts
2553or carried across separate
2554invocations of \f5ksh\fP.
2555The
2556.B $\s-1FPATH\s+1
2557variable gives a colon separated list of directories that
2558is searched for function definitions when trying to resolve
2559the command name.
2560Whenever a file name contained in
2561.B $\s-1FPATH\s+1
2562is found, the complete file is read and all functions
2563contained within become defined.
2564.P
2565Calls that reference functions can be recursive.
2566Except for special built-ins,
2567function names take precedence over built-in names and names
2568of programs when used as command names.
2569To write a replacement function that invokes the command that
2570you wish to replace,
2571you can use the \f5command\fP built-in command.
2572The arguments to \f5command\fP are the name and arguments
2573of the program you want to execute.
2574For example to write a
2575.B cd
2576function which changes the directory and prints out the directory name,
2577you can write
2578.sp
2579.nf
2580.in .5i
2581\f5function  cd
2582{
2583     if      command cd  "$@"
2584     then    print  -r -- $PWD
2585     fi
2586}\fP
2587.fi
2588.ta
2589.in
2590.sp
2591.P
2592The
2593\fB\s-1FPATH\s+1\fP
2594variable is a colon separated list that \f5ksh\fP
2595uses to search for function definitions.
2596When
2597\f5ksh\fP
2598encounters an autoload function,
2599it runs the
2600.B .
2601command on the script containing the function,
2602and then executes the function.
2603.P
2604For interactive shells,
2605function definitions may also be placed in the
2606\fB\s-1ENV\s+1\fP
2607file.
2608However, this
2609causes the shell to take longer to begin executing.
2610.H 2 "Process Substitution"
2611.P
2612This feature is only available
2613on versions of the UNIX operating system which support the
2614.B /dev/fd
2615directory for naming open files.
2616Each command argument of the form
2617\fB<(\fP\fIlist\^\fP\fB)\fP
2618or
2619\fB>(\fP\fIlist\^\fP\fB)\fP
2620will run process
2621.I list
2622asynchronously connected to some file in the
2623.B /dev/fd
2624directory.
2625The name of this file will become the argument to the command.
2626If the form with
2627.B >
2628is selected then writing on this file will provide input for
2629.IR list .
2630If
2631.B <
2632is used,
2633then the file passed as an argument will contain the output of the
2634.I list
2635process.
2636For example,
2637.sp
2638.nf
2639.in .5i
2640\f5paste  <(cut \-f1 \fP\fIfile1\fP\f5)  <(cut \-f2 \fP\fIfile2\fP\f5) | tee >(\fP\fIprocess1\fP\f5)  >(\fP\fIprocess2\fP\f5)\fP
2641.fi
2642.ta
2643.in
2644.sp
2645extracts
2646fields 1 and 3 from
2647the files
2648.I file1
2649and
2650.I file2
2651respectively,
2652places the
2653results side by side, and
2654sends it
2655to the processes
2656.I process1
2657and
2658.IR process2 ,
2659as well as putting it onto the standard output.
2660Note that the file which is passed as an argument to the command is
2661a UNIX system
2662.IR pipe (2)
2663so that the programs that expect to
2664.IR lseek (2)
2665on the file will not work.
2666.H 2 "Finding Commands"
2667.P
2668The addition of aliases, functions,
2669and more built-ins
2670has made it substantially more difficult to know what
2671a given command name really means.
2672.P
2673Commands that begin with reserved words
2674are an integral part of the shell language itself
2675and typically define the control flow of the language.
2676Some control flow commands are not reserved words in
2677the language but are \fIspecial\fP built-ins.
2678Special built-ins are built-ins that are considered a
2679part of the language rather than user definable commands.
2680The best examples of commands that fit this description
2681are \f5break\fP and \f5continue\fP.
2682Because they are not reserved words, they can be the
2683result of shell expansions and are not effected by quoting.
2684These commands have the following special properties:
2685.BL
2686.LI
2687Assignments that precede them apply to the current shell process,
2688not just to the given command.
2689.LI
2690An error in the format of these commands cause a shell script
2691or function that contains them to abort.
2692.LI
2693They cannot be overridden by shell functions.
2694.LE
2695.P
2696Other commands are built-in because they perform side effects
2697on the current environment that would be nearly impossible
2698to implement otherwise.
2699Built-ins such as \f5cd\fP and \f5read\fP
2700are examples of such built-ins.
2701These built-ins are semantically equivalent to commands that
2702are not built-in except that they don't take a path search
2703to locate.
2704.P
2705A third reason to have a command built-in is so that
2706it will be unaffected by the setting of the
2707.B \s-1PATH\s+1
2708variable.
2709The \f5print\fP  command fits this category.
2710Scripts that use \f5print\fP will be portable
2711to all sites that run \f5ksh\fP.
2712.P
2713The final reason for having a command be a built-in is
2714for performance.
2715On most systems it is more than an order of magnitude
2716faster to initiate a command that is built-in than
2717to create a separate process to run the command.
2718Examples that fit this category are \f5test\fP
2719and \f5pwd\fP.
2720.P
2721Given a command name \f5ksh\fP decides what it means using
2722the following order:
2723.BL
2724.LI
2725Reserved words define commands that form part of the shell
2726grammar.
2727They cannot be quoted.
2728.LI
2729Alias substitutions occur first as part of the reading of commands.
2730Using quotes in the command name will prevent alias substitutions.
2731.LI
2732Special built-ins.
2733.LI
2734Functions.
2735.LI
2736Commands that are built-in that are not associated with a pathname
2737such as \f5cd\fP and \f5print\fP.
2738.LI
2739If the command name contains a
2740.BR / ,
2741the program or script corresponding to the given name is executed.
2742.LI
2743A path search locates the pathname corresponding to the command.
2744If the pathname where it is found matches the pathname associated
2745with a built-in command, the built-in command is executed.
2746If the directory where the command is found is listed in the
2747.B \s-1FPATH\s+1
2748variable, the file is read into the shell
2749like a dot script, and a function by that name is invoked.
2750Once a pathname is found, \f5ksh\fP remembers its location
2751and only checks relative directories in \fB\s-1PATH\s+1\fP
2752the next time the command name is used.
2753Assigning a value to \fB\s-1PATH\s+1\fP
2754causes \f5ksh\fP to forget the location of all command names.
2755.LI
2756The
2757.B \s-1FPATH\s+1
2758variable is searched and files found are treated as described above.
2759.LE
2760.P
2761The first argument of the \f5command\fP built-in, described earlier,
2762skips the checks for reserved words and for function definitions.
2763In all other ways, \f5command\fP behaves like a built-in
2764that is not associated with a pathname.
2765As a result, if the first argument of \f5command\fP is
2766a special built-in, the special properties of this built-in
2767do not apply.
2768For example, whereas, \f5exec\ 3<\ foo\fP will cause a script containing
2769it to abort if the open fails, \f5command\ exec\ 3<\ foo\fP
2770results in a non-zero exit status but does not abort the script.
2771.P
2772You can get a complete list of the special built-in commands
2773with \f5builtin\ -s\fP.
2774In addition \f5builtin\fP without arguments gives a list of
2775the current built-ins and the pathname that they are associated with.
2776A built-in can be bound to another pathname by giving
2777the pathname for the built-in.  The basename of this path must
2778be the name of an existing built-in for this to succeed.
2779Specifying the name of the built-in without a pathname causes
2780this built-in to be found before a path search.
2781A built-in can be deleted  with the \fB\-d\fP option.
2782.P
2783On systems with run time loading of libraries, built-in commands
2784can be added with the \f5builtin\fP command.
2785Each command that is to be built-in must be written as a
2786C function whose name is of the form \f5b_\fP\fIname\fP, where
2787\fIname\fP is the name of the built-in that is to be added.
2788The function has the same argument calling convention as
2789\f5main\fP.  The lower eight bits of the return value become
2790the exit status for this built-in.
2791Builtins are added by specifying the pathname of the library
2792as an argument to the \fB\-f\fP option of \f5builtin\fP.
2793.P
2794The built-in command,
2795\f5whence\fP,
2796when used with the
2797.B \-v
2798option, tells how a given command is bound.
2799A line is printed for each argument to \f5whence\fP
2800telling what would happen if this argument were used as a command name.
2801It reports on reserved words, aliases, built-ins, and
2802functions.
2803If the command is none of the above,
2804it follows the path search rules and prints the full path-name,
2805if any, otherwise it prints an error message.
2806.H 2 "Symbolic Names"
2807To avoid implementation dependencies, \f5ksh\fP
2808accepts and generates symbolic names
2809for built-ins that use numerical values in the Bourne shell.
2810The
2811.B \-S
2812option of the
2813\f5umask\fP built-in command
2814accepts and displays
2815default file creation permissions
2816symbolically.
2817It uses the same symbolic notation as the \f5chmod\fP command.
2818.P
2819The \f5trap\fP and \f5kill\fP built-in commands
2820allows the signal names to be given symbolically.
2821The names of signals and traps
2822corresponding to signals are the same as the signal name with
2823the
2824.B \s-1SIG\s+1
2825prefix removed.
2826The trap
2827.B 0
2828is named
2829\fB\s-1EXIT\s+1\fP.
2830.H 2 "Additional Variables"
2831In addition to the variables discussed earlier, \f5ksh\fP
2832has other variables that it handles specially.
2833The variable \fB\s-1RANDOM\s+1\fP
2834produces a random number in the range 0 to 32767 each time it is referenced.
2835Assignment to this variable sets the seed for the
2836random number generator.
2837.P
2838The parameter \fB\s-1PPID\s+1\fP
2839is used to generate the process id of the process which invoked this shell.
2840.H 2 "Added Traps"
2841A new trap named
2842\fB\s-1ERR\s+1\fP
2843has been added.
2844This trap is invoked whenever the shell would exit if the
2845.B \-e
2846option were set.
2847This trap is used by
2848Fourth Generation Make\*(Rf
2849.RS
2850G. S. Fowler,
2851.I "The Fourth Generation Make,"
2852Proceedings of the Portland USENIX meeting, pp. 159-174, 1985.
2853.RF
2854which runs \f5ksh\fP
2855as a co-process.
2856.P
2857A trap named
2858\fB\s-1DEBUG\s+1\fP
2859gets executed after each command.
2860This trap can be used for debugging and other purposes.
2861.P
2862The
2863\fB\s-1KEYBD\s+1\fP
2864trap was described earlier.
2865.H 2 Debugging
2866The primary method for debugging Bourne shell scripts is to
2867use the
2868.B \-x
2869option to enable the execution trace.
2870After all
2871the expansions have been performed,
2872but before each command is executed,
2873the trace writes to standard error the name and arguments
2874of each command preceded by a
2875.BR + .
2876While the trace is very useful, there is no way
2877to find out what line of source a given trace line
2878corresponds to.
2879With
2880\f5ksh\fP
2881the
2882\fB\s-1PS4\s+1\fP
2883variable
2884is evaluated for parameter expansion and
2885is displayed before each command,
2886instead of the
2887.BR + .
2888.P
2889The
2890\fB\s-1LINENO\s+1\fP
2891variable is set to the current line number relative to the
2892beginning of the current script or function.
2893It is most useful as part of the
2894\fB\s-1PS4\s+1\fP
2895prompt.
2896.P
2897The
2898\fB\s-1DEBUG\s+1\fP
2899trap can be used to write a break point shell
2900debugger in \f5ksh\fP.
2901An example of such a debugger is \f5kshdb\fP.\*(Rf
2902.RS
2903Bill Rosenblatt,
2904.IR "Debugging Shell Scripts with \f5kshdb\fP" ,
2905Unix World, Volume X, No. 5, 1993.
2906.RF
2907.H 2 "Timing Commands"
2908.P
2909Finding the time it takes to execute commands
2910has been a serious problem with the Bourne shell.
2911Since the \f5time\fP command is not part of the
2912language, it is necessary to write a script
2913in order to time a \f5for\fP or \f5while\fP loop.
2914The extra time in invoking the shell and processing
2915the script is accumulated along with the time
2916to execute the script.
2917.P
2918More seriously, the Bourne shell does not give correct
2919times for pipelines.
2920The reason for this is that the times for some members
2921of a pipeline are not counted when computing the time.
2922As an extreme example,
2923running \f5time\fP on the script
2924.ce
2925\f5cat < /dev/null | sort -u bigfile | wc\fP
2926with the Bourne shell will show very little
2927user and system time no matter how
2928large \f5bigfile\fP is.
2929.P
2930To correct these problems,
2931a reserved word \f5time\fP
2932has been added to replace
2933the \f5time\fP
2934command.
2935Any function, command or pipeline can be preceded by this reserved word
2936to obtain information about the elapsed, user, and system times.
2937Since I/O redirections bind to the command, not to
2938\f5time\fP,
2939parentheses should be used to redirect the timing information which
2940is normally printed on file descriptor 2.
2941.H 1 SECURITY
2942There are several documented problems associated with the security of
2943shell procedures\*(Rf.
2944.RS
2945F. T. Grampp and R. H. Morris,
2946.I "UNIX Operating System Security,"
2947AT&T Bell Labs Tech. Journal, Vol. 63, No. 8, Part 2, pp. 1649-1671, 1984.
2948.RF
2949These security holes occur primarily because a user can manipulate the
2950.I environment
2951to subvert the intent of a
2952.I setuid
2953shell procedure.
2954Sometimes, shell procedures are initiated from
2955binary programs, without the author's
2956awareness, by library routines which invoke shells to carry out
2957their tasks.
2958When the binary program is run
2959.I setuid
2960then the shell procedure runs with the permissions afforded to the
2961owner of the binary file.
2962.P
2963In the Bourne shell,
2964the
2965.B \s-1IFS\s+1
2966parameter is used to split each word into separate command arguments.
2967If a user knows that some
2968.I setuid
2969program will run
2970\f5sh\ -c\ /bin/pwd\fP
2971(or any other command in
2972.BR /bin )
2973then the user sets and exports
2974.BR \s-1IFS\s+1=\^/ .
2975Instead of running
2976.B /bin/pwd
2977the shell will run
2978.B bin
2979with
2980.B pwd
2981as an argument.
2982The user puts his or her own \f5bin\fP
2983program into the current directory.
2984This program can
2985create a copy of the shell,
2986make this shell
2987.IR setuid ,
2988and then run the \f5/bin/pwd\fP
2989program so that the original program continues to run successfully.
2990This kind of penetration is not possible with
2991\f5ksh\fP
2992since the
2993.B \s-1IFS\s+1
2994parameter only splits arguments that result from command or parameter
2995substitution.
2996.P
2997Some
2998.I setuid
2999programs run programs using
3000.I system()
3001without giving the full pathname.
3002If the
3003user sets the
3004.B \s-1PATH\s+1
3005variable so that the desired command will be found
3006in his or her local bin, then the same technique described above can
3007be employed to compromise the security of the system.
3008To close up this and other security holes,
3009\f5ksh\fP
3010resets the effective user id to the real user id and the effective
3011group id to the real group id unless the
3012.I privileged
3013option
3014.RB ( \-p\^ )
3015is specified at invocation.
3016In
3017this mode, the
3018.B privileged
3019mode, the
3020.B .profile
3021and
3022.B \s-1ENV\s+1
3023files are not processed.
3024Instead, the file
3025.B /etc/suid_profile
3026is read and executed.
3027This gives an administrator control over the
3028environment to set the
3029.B \s-1PATH\s+1
3030variable or to log setuid shell invocations.
3031Clearly security of the system is compromised if
3032.B /etc
3033or this file is publicly writable.
3034.P
3035Some versions of the UNIX operating system look for the characters
3036\f5#!\fP
3037as the first two characters of an executable file.
3038If these characters are found, then the next word on this line is taken
3039as the interpreter to
3040invoke
3041for this command and the interpreter is
3042.IR exec ed
3043with the name of the script as argument zero and argument one.
3044If the
3045.I setuid
3046or
3047.I setgid
3048bits are on for this file, then the interpreter
3049is run with the effective uid and/or gid set accordingly.
3050This scheme has three major drawbacks.
3051First of all,
3052putting the pathname of the interpreter into the script
3053makes the script less portable since the interpreter
3054may be installed in a different directory on another system.
3055Secondly, using the
3056\f5#!\fP
3057notation forces an
3058.B exec
3059of the interpreter even when the call is invoked from the interpreter
3060which it must exec.  This is inefficient since
3061\f5ksh\fP can handle a failed exec much faster than starting up
3062again.
3063More importantly,
3064.I setuid
3065and
3066.I setgid
3067procedures provide an easy target for intrusion.
3068By linking a
3069.I setuid
3070or
3071.I setgid
3072procedure to a name beginning with a
3073.B \-
3074the interpreter is fooled into thinking that it is being invoked with
3075a command line option rather than the name of a file.
3076When the interpreter is the shell, the user gets a privileged
3077interactive shell.
3078There is code in
3079\f5ksh\fP
3080to guard against this simple form of intrusion.
3081.P
3082A more reliable way to handle
3083.I setuid
3084and
3085.I setgid
3086procedures is provided with
3087\f5ksh\fP.
3088The technique does not require any changes to the operating system
3089and provides better security.
3090Another advantage to this method is that it also allows scripts which
3091have execute permission but no read permission to run.  Taking away read
3092permission makes scripts more secure.
3093.P
3094The method relies on a setuid
3095.B root
3096program to authenticate the
3097request and exec the shell with the correct mode bits to carry out
3098the task.  This shell is invoked with the requested file already open
3099for reading.  A script which cannot be opened for reading or which
3100has its setuid and/or setgid bits turned on causes this setuid
3101.B root
3102program to get \fBexec\fPed.
3103For security reasons, this program is given the full
3104pathname
3105\f5/etc/suid_exec\fP.
3106A description of the implementation of the
3107\f5/etc/suid_exec\fP
3108program can be found in
3109a separate paper\*(Rf.
3110.RS
3111D. G Korn
3112.I "Parlez-vous Kanji?"
3113TM-59554-860602-03, 1986.
3114.RF
3115.H 1 "CODE CHANGES"
3116\f5ksh\fP is written in ANSI-C as a reusable library.
3117The code can be compiled with C++ and older K&R C as well.
3118The code uses the IEEE POSIX 1003.1 and ISO 9945-1 standard\*(Rf
3119.RS
3120.I "POSIX \- Part 1: System Application Program Interface,"
3121IEEE Std 1003.1-1990, ISO/IEC 9945-1:1990.
3122.RF
3123wherever possible so that \f5ksh\fP should be able to run
3124on any POSIX compliant system.  In addition, it is possible
3125to compile \f5ksh\fP for older systems.
3126.P
3127Unlike earlier version of the Bourne shell,
3128\f5ksh\fP treats eight bit characters transparently
3129without stripping off the
3130leading bit.
3131There is also a compile time switch to enable handling multi-byte
3132and multi-width characters sets.
3133.P
3134On systems with dynamic libraries, it is possible to add built-in
3135commands at run time  with the built-in command \f5builtin\fP
3136described earlier.
3137It is also possible to embed \f5ksh\fP in applications in
3138a manner analogous to \f5tcl\fP.
3139.H 1 "EXAMPLE"
3140.P
3141An example of a \f5ksh\fP script is included
3142in the Appendix.
3143This one page program is a variant of the UNIX system
3144\f5grep\fP(1) program.
3145Pattern matching for this version of \f5grep\fP
3146means shell patterns.
3147.P
3148The first half uses the \f5getopts\fP command to
3149find the option flags.
3150Nearly all options have been implemented.
3151The second half goes through each line of each file
3152to look for a pattern match.
3153.P
3154This program is not intended to serve as a
3155replacement for \f5grep\fP
3156which has been highly tuned for performance.
3157It does
3158illustrate the programming power of \f5ksh\fP.
3159Note that no auxiliary processes are spawned by this script.
3160It was written and debugged in under two hours.
3161While performance is acceptable for small files,
3162this program runs at only one tenth
3163the speed of \f5grep\fP
3164for large files.
3165.H 1 "PERFORMANCE"
3166.P
3167\f5ksh\fP executes many scripts faster than the System V Bourne shell;
3168in some cases more than 10 times as fast.
3169The primary reason for this is that \f5ksh\fP creates fewer
3170processes.
3171The time to execute a built-in command or a function is one or two
3172orders of magnitude faster than performing a \f5fork\fP() and
3173\f5exec\fP() to create a separate process.
3174Command substitution and commands inside parentheses
3175are performed without creating another process, unless necessary
3176to preserve correct behavior.
3177.P
3178Another reason for improved performance is the use of the \fBsfio\fP\*(Rf,
3179.RS
3180David Korn and Kiem-Phong Vo,
3181.IR "SFIO - A Safe/Fast String/File I/O,"
3182Proceedings of the Summer Usenix,
3183pp. 235-255, 1991.
3184.RF
3185library for I/O.  The \fBsfio\fP library buffers all I/O
3186and buffers are flushed only when required.
3187The algorithms used in \fBsfio\fP perform better than
3188traditional versions of standard I/O so that programs that
3189spend most of their time
3190formatting output may actually perform better
3191than versions written in C.
3192.P
3193Several of the internal algorithms have been changed
3194so that the number of subroutine calls has been
3195substantially reduced.
3196\f5ksh\fP uses variable sized hash tables for variables.
3197Scripts that rely heavily on referencing variables execute faster.
3198More processing is performed while reading the script
3199so that execution time is saved while running loops.
3200These changes are not noticeable for scripts that \f5fork()\fP
3201and run processes,
3202but they reduce the time that it takes to interpret commands by
3203more than a factor of two.
3204.P
3205Most importantly, \f5ksh\fP provide mechanisms to write applications
3206that do not require as many processes.
3207The arithmetic provided by the shell eliminates the need for the
3208\f5expr\fP command.
3209The pattern matching and substring capabilities eliminate the
3210need to use \f5sed\fP or \f5awk\fP to process strings.
3211.P
3212The architecture of \f5ksh\fP makes it easy to make commands
3213built-ins without changing the semantics at all.
3214Systems that have run-time binding of libraries allow
3215applications to be sped up by supplying the critical
3216programs as shell built-in commands.
3217Implementations on other systems can add built-in commands
3218at compile time.
3219The procedure for writing built-in commands that can be loaded
3220at run time is in a separate document.\*(Rf,
3221.RS
3222David Korn,
3223.IR "Guidelines for writing \f5ksh-93\fP built-in commands,"
3224to be published, 1994.
3225.RF
3226.H 1  "CONCLUSION"
3227.P
3228The 1988 version of \f5ksh\fP has tens of thousands of regular users
3229and is a suitable replacement for the Bourne shell.
3230The 1993 version of \f5ksh\fP is essentially upward compatible with
3231both the 1988 version of \f5ksh\fP and with the recent IEEE POSIX
3232and ISO shell standard.
3233The 1993 version offers many advantages for programming applications,
3234and it has been rewritten so that it can be used in embedded applications.
3235It also offers improved performance.
3236.SG dgk \"  signature typist initials
3237\" .CS 14 24 38 0 0 16  \" cover sheet for TM
3238.bp
3239.ce
3240\fIAPPENDIX\fP
3241.nf
3242\f5
3243.ta .66i 1.33i 2i 2.66i 3.33i 4i 4.66i 5.33i 6i 6.66i 7.33i 8i
3244.so grep.mm
3245.fi
3246\fP
3247
3248
3249