1N/AThis is a list of changes that have been made since the 12/28/93 version
1N/A1. New features in 12/28/93b
1N/A a. If IFS contains two consecutive identical characters belonging
1N/A to the [:space:] class, then this character is treated as
1N/A a non-space delimiter so that each instance will delimit
1N/A a field. For example, IFS=$'\t\t' will cause two consecutive
1N/A tabs to delimit a null field.
1N/A b. The getopts command has a -a name option that specifies a
1N/A name that will be used for usage messages.
1N/A2. New features in 12/28/93e
1N/A a. The math functions, atan2, hypot, fmod, and pow were added.
1N/A b. When a shared library is loaded, if the function lib_init()
1N/A is defined in the library, it is invoked the first time that
1N/A the library is loaded with builtin -f library.
1N/A3. New features in 12/28/93f
1N/A a. Hostnames in addition to host addresses can be given in
1N/A b. File name completion and expansion now quotes special
1N/A characters in file names from both emacs and vi edit modes.
1N/A4. New features in 12/28/93g
1N/A a. The pipefail option has been added. With pipefail
1N/A enabled, a pipeline will not complete until all
1N/A commands are complete, and the return value will
1N/A be that of the last command to fail, or zero if
1N/A all complete successfully.
1N/A b. When an executable is found on a given path,
1N/A the appropriate library path variable is prepended
1N/A with a corresponding library directory.
1N/A5. New features in 12/28/93h
1N/A a. The PATH search algorithm has been modified to look
1N/A for a file named .fpath in each bin directory and if
1N/A found, to search for functions in this directory if
1N/A it cannot find the command in that directory.
1N/A b. When performing pathname expansion, the shell checks
1N/A to see whether each directory it reads is case sensitive
1N/A or not, and performs the matching accordingly.
1N/A6. New features in 12/28/93i
1N/A a. Most of the built-in commands and ksh itself are now
1N/A self documenting. Running command --man will produce
1N/A screen output. Running command --html produces the
1N/A man page in html format.
1N/A b. The getopts builtin can process command description
1N/A strings to produce man pages.
1N/A7. Bugs fixed in 12/28/93a for default OPTIONS
1N/A a. An expansion bug which causes portions of a word after
1N/A a $((...)) expansion that contains a nested $var expansion
1N/A to be lost has been fixed.
1N/A b. A bug that caused a core dump when a script that did not
1N/A have PWD set and did a cd inside command substitution
1N/A c. A bug which caused a core dump on some machines when
1N/A the LANG variable was assigned to has been fixed.
1N/A d. A bug which incorrectly handled set disciplines that
1N/A performed arithmetic evaluation when the discipline
1N/A was called from the arithmetic evaluator has been fixed.
1N/A e. A bug caused by an EXIT trap inside a function that
1N/A was executed in a subshell was fixed.
1N/A f. If foo is a function, and not a program, then command foo
1N/A now reports that foo isn't found rather than invoking foo.
1N/A g. The previous version incorrectly listed -A as an
1N/A invocation option. The -A option is only for set.
1N/A h. A bug was fixed which caused ksh to loop when execution trace
1N/A was enabled and the PS4 prompt required command substitution.
1N/A i. A bug which could cause the job control switch character
1N/A to be disabled when a script that enabled monitor mode
1N/A terminated was fixed.
1N/A j. A bug in the macro expansion global replacement operator //,
1N/A when the pattern began with a [ or +( has been fixed.
1N/A k. A bug which prevented ~ expansion from occurring when
1N/A it was terminated with a colon inside an assignment
1N/A l. A bug in the dot command which prevented autoload functions
1N/A from working has been fixed.
1N/A m. A bug which caused a variable to be unset if the
1N/A its value were expanded inside a set discipline has
1N/A n. Whence -a now longer reports that a defined function
1N/A o. A bug on some systems in which $0 would be incorrect
1N/A in scripts invoked by name has been fixed.
1N/A p. Here documents with an empty body now work.
1N/A 1. A bug which disabled argument passing and resetting
1N/A of options for a script invoked by name inside a
1N/A function has been fixed.
1N/A r. A bug in which an EXIT trap set the caller of a function
1N/A would be executed if a command called inside a function
1N/A was not found has been fixed.
1N/A s. A bug which allowed a script to trap signals that are
1N/A ignored at the time that the shell was invoked has
1N/A t. A bug which caused 2<&1- when applied to a shell built-in
1N/A to leave standard input closed has been fixed.
1N/A u. A bug which caused the shell to incorrectly parse
1N/A $() command substitutions with nested case statements
1N/A8. Bugs fixed in 12/28/93b for default OPTIONS
1N/A a. A bug which caused unset RANDOM to dump core has been
1N/A b. A bug which prevented return for terminating a profile
1N/A or ENV file has been fixed.
1N/A c. A bug which prevented standard input from being
1N/A monitor mode was turned off has been fixed.
1N/A d. Statements of the form typeset -options var[expr]=value
1N/A did not perform substitutions on expr as expected.
1N/A e. A bug which prevented the shell from sending a HUP
1N/A signal to some background jobs that were not disowned
1N/A f. A bug which allowed a script to trap signals that are
1N/A ignored at the time that the shell was invoked by exec
1N/A g. A bug which could cause a core dump when a discipline
1N/A function was unset within a discipline was fixed.
1N/A h. The typeset builtin now accepts a first argument of
1N/A + or - for compatibility with ksh88.
1N/A i. For compatibility with ksh88, the results of expansions
1N/A of command arguments will treat the extended character
1N/A match characters ()|& as ordinary characters.
1N/A j. A bug which caused read to fail on a file that was
1N/A was print or printf has been fixed.
1N/A k. When a job is suspended, it is put on the top of
1N/A the job list as required by the POSIX standard.
1N/A l. The value of OPTARG when an option that required
1N/A an argument but didn't have one was incorrect in the
1N/A case the the option string began with a :.
1N/A m. A bug which caused the terminal to get into a bad
1N/A state with some KEYBD traps in vi-mode has been fixed.
1N/A n. A bug which caused an invalid trap to cause a script
1N/A to terminate, rather than just return an error, has
1N/A o. Backreferencing sub-expressions in patterns and replacement
1N/A p. A bug in chmod which caused the -R option to fail has
1N/A9. Bugs fixed in 12/28/93c for default OPTIONS
1N/A a. The expansion of "$@" was incorrect when $1 was the null
1N/A b. A bug which could incorrectly report a syntax error in
1N/A a backquoted expression when a $ was preceded by \\
1N/A c. A bug which prevented the shell from exiting after
1N/A reporting an error when failing to open a script
1N/A d. A bug that could lead to memory corruption when a
1N/A large here document that required parameter or command
1N/A substitution was expanded has been fixed.
1N/A e. A bug that could cause a core dump on some systems
1N/A after ksh detected an error when reading a function
1N/A f. A bug which could cause a coprocess to hang when
1N/A reading from a process that has terminated has been fixed.
1N/A g. A bug which caused a script to terminate when set -e
1N/A was on and the first command of and && or || list
1N/A failed has been fixed.
1N/A h. A bug with here documents inside $(...) when the delimiter
1N/A word is an identifier has been fixed.
1N/A i. A bug which caused $0 to display the wrong value when
1N/A a script was invoked as an argument to the . command
1N/A and the eval command has been fixed.
1N/A j. A bug that could cause the built-in sleep to hang
1N/A k. A bug introduces in 12/28/93b which caused the backslash
1N/A to be removed when it was followed by digit inside double
1N/A quotes in some instances has been fixed.
1N/A l. A bug which could cause a core dump if ksh was invoked with
1N/A standard input closed has been fixed.
1N/A m. A bug which could cause a core dump if typeset -A was
1N/A specified for an existing variable has been fixed.
1N/A n. Variables that were unset but had attributes such as readonly
1N/A and export were not listed with readonly, export and typeset.
1N/A o. Several problems with signals have been fixed.
1N/A p. A bug which prevented ulimit -t from working has been fixed.
1N/A Also, a bug in which failed ulimits could cause a core dump
1N/A has also been fixed.
1N/A r. A bug which caused read -r on a line that contained only
1N/A blanks to get a non-null value has been fixed.
1N/A s. A bug introduced in the 'a' point release in which
1N/A ${x='\\'} expanded to \ when x was unset has been fixed.
1N/A t. A bug which prevented a trap on EXIT from being executed
1N/A when the last command in a script was a function invocation
1N/A u. A bug which caused an interactive shell ignore input when
1N/A standard error was redirected to a file with exec,
1N/A and then restored with exec 2>&1 has been fixed.
1N/A v. An interactive shell turns on monitor mode even when
1N/A standard error has been redirected to a file.
1N/A w. A bug which could cause standard input to be incorrectly
1N/A positioned for the last command of a script has been fixed.
1N/A y. A bug in the edit modes which allowed walking back in
1N/A the history file for more than HISTSIZE commands has
1N/A z. A bug which could cause a core dump if variable TMPDIR was
1N/A changed between two command substitutions has been fixed.
1N/A aa. A bug which prevented a trap on EXIT from being cleared
1N/A10. Bugs fixed in 12/28/93d for default OPTIONS
1N/A a. The \ character was not handled correctly in replacement
1N/A b. A bug with read in which the line did not end with
1N/A a new-line has been fixed.
1N/A c. A bug in file name generation which sometimes
1N/A appended a . for filenames that ended in / has
1N/A d. If a process is waited for after a status has
1N/A been returned by a previous wait, wait now
1N/A e. A bug with hist (fc) -e which prevented a command
1N/A to re-executed after it had been edited has been fixed.
1N/A f. A bug which prevented quoting from removing the meaning
1N/A of unary test operators has been fixed.
1N/A11. Bugs fixed in 12/28/93e for default OPTIONS
1N/A a. Empty command substitutions of the form $() now work.
1N/A b. whence -v foo now gives the correct result after calling
1N/A c. A bug in right to left arithmetic assignment for which
1N/A the arithmetic expression (( y = x = 1.5 )) did not
1N/A yield 1 for y when x was declared typeset -i was fixed.
1N/A d. printf has been fixed to handle format containing \0
1N/A and/or \0145 correctly. In addition, characters following
1N/A %b in the format string are no longer displayed when
1N/A the operand contains \c.
1N/A e. A bug in printf that could cause the %E format to
1N/A produce unnormalized results has been fixed.
1N/A f. A bug which causes some arithmetic expressions to be
1N/A incorrectly evaluated as integer expressions rather
1N/A that floating point has been fixed.
1N/A g. Functions defined inside a subshell no longer remain
1N/A defined when the subshell completes.
1N/A h. The error message from sh -c ';echo foo' has been
1N/A i. The format for umask -S has been changed to agree
1N/A with the specification in the POSIX standard.
1N/A j. A bug that caused side effects in subscript evaluation
1N/A when tracing was enabled for subscripts using ++ or --
1N/A k. To conform to the Posix standard getopts has been changed
1N/A so that the option char is set to ? when it returns with
1N/A a non-zero exit status.
1N/A l. The handling of \} inside ${name...} has been fixed so
1N/A that the \ quotes the }.
1N/A m. A bug that caused the read builtin to resume execution
1N/A after processing a trap has been fixed.
1N/A n. [[ -s file ]] has been fixed so that if file is open
1N/A by ksh, it is flushed first.
1N/A o. In some cases attributes and sizes for non exported
1N/A variables weren't being reset before running a script.
1N/A p. The value of TMOUT was affected by changes make to
1N/A q. The jobs command did not reflect changes make by
1N/A sending the CONT signal to a command.
1N/A r. The error message for ksh -o unknown was incorrect.
1N/A s. Functions invoked as name=value name, did not use
1N/A values from the calling scope when evaluating value.
1N/A t. A bug in which the shell would reexecute previously
1N/A executed code when a shell script or coprocess was
1N/A run in the background has been fixed.
1N/A u. A bug in which an empty here-document would leave
1N/A a file descriptor open has been fixed.
1N/A v. A bug in which $(set -A array ...) would leave a
1N/A side effect has been fixed.
1N/A w. A discipline function for a global variable defined
1N/A within a function defined with the function keyword,
1N/A incorrectly created a local variable of the same name
1N/A and applied the discipline to it.
1N/A12. Bugs fixed in 12/28/93f for default OPTIONS
1N/A a. A bug which would cause the secondary prompt to be
1N/A displayed when a user entered a literal carriage
1N/A return has been fixed.
1N/A b. I bug which caused ksh read -s name to core dump was
1N/A c. I bug with the expansion of \} and \] inside double
1N/A quoted strings that also contained variable expansions
1N/A d. Changes in the 'e' point release caused autoload
1N/A functions invoked from within command substitution
1N/A to fail. This has been fixed.
1N/A e. A bug in the processing of here-documents that could
1N/A prevent variable substitution to occur after $(...) command
1N/A substitution for long here documents has been fixed.
1N/A f. A bug caused by a race condition that could cause SIGTERM
1N/A to be ignored by a child process has been fixed.
1N/A g. A bug which prevented the startup of a coprocess immediately
1N/A after killing a running coprocess has been fixed.
1N/A h. ulimit foobar, where foobar is not an arithmetic
1N/A expression, now gives an error message as it did with ksh88
1N/A instead of setting the file size limit to 0.
1N/A i. A bug which could cause an interactive shell to terminate when
1N/A the last process of a pipeline was a POSIX function was fixed.
1N/A j. A bug which could cause command substitution of a shell script
1N/A to core dump has been fixed.
1N/A k. A security hole was fixed in suid_exec.
1N/A l. Arithmetic functions such as pow() that take more than
1N/A one argument, did not work if arguments other than the
1N/A first contained parenthesized sub-expression.
1N/A m. The error message from a script containing an incomplete
1N/A arithmetic expression has been corrected.
1N/A n. A bug which caused a core dump on some machines when
1N/A the value of a name reference contained a positional
1N/A parameter and the name reference was not defined inside
1N/A a function has been fixed.
1N/A o. Arithmetic expressions now correctly handle hexidecimal
1N/A p. A bug in which integer variables could be expanded
1N/A with a leading 10# when declared with typeset -i
1N/A multiple times has been corrected.
1N/A q. A bug in which IFS wasn't correctly restored when
1N/A set within command substitution has been fixed.
1N/A r. The _ character is now considered as part of a word
1N/A with the M-f and M-b emacs directives as it was in ksh88.
1N/A13. Bugs fixed in 12/28/93g for default OPTIONS
1N/A a. A bug in which a name reference could be created to
1N/A itself and later cause the shell to get into an infinite
1N/A loop has been fixed.
1N/A b. A bug in shcomp relating to compound variables was fixed.
1N/A c. A bug introduced in 'e' in which leading 0's in -Z
1N/A fields caused the value to be treated as octal for arithmetic
1N/A evaluation has been fixed.
1N/A d. A bug when a name reference with a shorter name than
1N/A the variable it references was the subject of a compound
1N/A assignment has been fixed.
1N/A e. A bug which in which assignment to array variables in
1N/A a subshell could effect the parent shell has been
1N/A f. read name?prompt was putting a 0 byte at the end of the
1N/A prompt on standard error.
1N/A g. A bug in [[ string1 > string2 ]] when ksh was run with -x
1N/A k. A bug in which the escape character was not processed
1N/A correctly inside {...} when brace expansion is enabled
1N/A has been fixed, for example {\$foo}.
1N/A l. A bug in line continuation in here-documents has been
1N/A m. The default base when not specified with typeset -i is
1N/A 10 in accordance with the documentation. Previously,
1N/A the value was determined by the first assignment.
1N/A n. A parsing bug in which a # preceded alphanumeric
1N/A characters inside a command substitution caused
1N/A a syntax error to be reported has been fixed.
1N/A o. A bug in which a decimal constant represented as 10#ddd
1N/A where ddd was more than five digits generated a syntax
1N/A error has been fixed.
1N/A p. A bug in here document expansion in which ${...} expansions
1N/A were split across buffer boundaries has been fixed.
1N/A14. Bugs fixed in 12/28/93h for default OPTIONS
1N/A a. I bug in shcomp for compilation of unary operators with [[...]]
1N/A b. A bug in which the value of $? was changed when executing
1N/A a keyboard trap has been fixed.
1N/A c. The handling of SIGCHLD has been changed so that the
1N/A trap is not triggered while executing trap commands
1N/A to avoid recursive trap calls.
1N/A d. I bug in which a local variable in a function declared readonly
1N/A would generated an error when the function went out of
1N/A scope has been fixed.
1N/A e. I bug in which \<new_line> entered from the keyboard
1N/A with the KEYBD trap enabled has been fixed.
1N/A f. The error message for a misplaced ((, for example print ((3),
1N/A was often garbled and has been fixed.
1N/A g. I bug in the KEYBD trap in which escape sequences of the form
1N/A <ESC>[#~ were not being handled as a unit has been fixed.
1N/A h. A bug in which ksh would consider expressions like [[ (a) ]]
1N/A as syntax errors has been fixed.
1N/A i. A function defined as foo() without a function body
1N/A was not reported as a syntax error.
1N/A j. A bug in which ksh could run out of file descriptors when
1N/A a stream was repeatedly opened with exec and read from
1N/A k. A bug introduced when fixing item n from the 'g' point
1N/A release has been fixed.
1N/A15. Bugs fixed in 12/28/93i for default OPTIONS
1N/A a. A bug in which a script could terminate when getopts
1N/A encountered an error when invoked inside a function
1N/A b. When a symbolic link was specified as the name of
1N/A the script to invoke by name, the value of $0 was
1N/A set to the real file name rather than the link name
1N/A in some cases and this has been fixed.
1N/A16. Bug fixes for specific non-default option combinations.
1N/A a. More signal names have been added for Solaris
1N/A b. A bug fixed for the v directive in vi MULTIBYTE has been
1N/A c. Code to for IFS handling of multibyte characters has
1N/A d. The displaying of multibyte strings in export, readonly,
1N/A typeset, and execution traces has been fixed.
1N/A e. A bug with type ahead and KEYBOARD traps with the
1N/A MULTIBYTE option set has been fixed.
1N/A f. The k-shell information abstraction database option, KIA,
1N/A has been revamped for the 'e' point release.
1N/A g. A bug in brace pattern expansions that caused expressions
1N/A such as {foo\,bar,bam} to expand incorrectly have been fixed.
1N/A h. On the
U/WIN version for Window 95 and Windows NT,
1N/A when a directory beginning with a letter followed by
1N/A a colon is given to cd, it is assumed to be an absolute
1N/A i. There was a bug in the compile option that does not
1N/A use fork() in which the current option settings where
1N/A not propagated to sub-shells.
1N/A for the MULTIBYTE option was fixed in release 'h'.
1N/A k. A bug in which the precision given as an argument
1N/A to printf was not working has been fixed.
1N/A17. Other changes to 12/28/93[abcdefghi]
1N/A a. A couple of minor changes to make adding built-ins easier.
1N/A b. Variables inside functions are now statically scoped.
1N/A The previous behavior was never documented.
1N/A c. A few changes have been made to the name-value library
1N/A that affect built-ins that use disciplines. The
1N/A changes allow disciplines to be shared by variables
1N/A and should make it possible to add new disciplines
1N/A without recompilation.
1N/A d. The name-value library interface has undergone significant
1N/A change for this revision. See the new nval.3 man page.
1N/A e. Builtin functions can take a third argument which is
1N/A f. The nv_scan() function can restrict the scope of a walk
1N/A to the top scope. Starting in 'f', nv_scan() has an
1N/A additional pointer argument that is passed to each invoked
1N/A g. Starting with release 'f', an empty for list behave like
1N/A a for list with null expansions. It produces a warning
1N/A h. Starting with release 'f' the code has been modified to
1N/A work with EBCDIC as well as ASCII.
1N/A i. Starting with the release 'g', the name-value pair library
1N/A uses the cdt library rather than the hash library.
1N/A j. The sh_fun() function now takes third argument which
1N/A is an argument list for the invoked discipline function
1N/A k. A callback function can be installed which will give
1N/A notification of file duplications and file closes.
1N/A18. Incompatibilities with 12/28/93 version.