ksh 11/16/88i RELEASEa file
Larry Cipriani
month or so ago, the RELEASEa file showing changes made is below.
Licensing info is available from AT&T Toolchest 908-522-6900 @ 1200 baud,
login as guest.
----------------------------------------------------------------------
1. New features in 11/16/88[a-i]
a. All characters of alias names can contain non-special
characters, not just the first. A / is not a valid alias
character.
b. The vi % directive has been added to 11/16/88c
c. A new [[...]] -C option returns true for contiguous files.
This option returns false on systems that do not support
contiguous files.
d. Setting REPLY to the null string in a select command causes
the selection list to get refreshed in 11/16/88d.
e. Unary minus added for arithmetic expressions to increase
compatibility with system V test command.
f. Starting with 11/16/88e, The vi directive '#' has been changed
so that each line of the command is turned into a comment.
Also, if the first character of the command is a '#', the '#'
in front of each line is deleted.
g. The message for EOF with ignoreeof has changed in 11/16/88e.
h. A SOCKETS option has been added for machines with the BSD
socket calls so that scripts can connect to sockets via
redirection. See the README file for more details.
i. Starting with 11/16/88e, the behavior of 'ksh script' where
script is not a shell script has been changed. If ksh detects
that script is an a.out, a cannot exec error message is generated.
j. The s directive in vi was added to the documentation starting
with 11/16/88g.
k. The directory /usr/local/bin was added to the set of trusted
directories for /etc/suid_exec starting with 11/16/88g.
l. Starting in 11/16/88g, == can be used where = is used
for string comparison inside [[...]].
m. Starting in 11/16/88g, ulimit without any arguments will print
out a large number rather than the string "unlimited" when
the file size limit is unlimited.
n. With FS_3D option, assignemnts to VPATH cause 3d mounts
starting in 11/16/88i
2. Bugs fixed in 11/16/88a for default OPTIONS
a. Files containing a \ were not matched correctly with patterns.
b. read without arguments no longer references location 0.
c. The value of integer exported variables was not inherited
correctly in some cases.
d. The characters ()|&\ in IFS did cause word splitting when ksh
expanded commands.
e. $! expanded to 0 before any background jobs were run, rather
than to the null string. The value of $! was not reset to
null when running a script by name.
f. On some machines, the shell would core dump if the variable
LINENO was exported.
g. The [[ ]] command did not handle multiple && and || without
parenthesis grouping.
h. On some BSD implementations, terminal setting were being
incorrectly reset when using vi-mode.
i. The execution of traps that write to standard output got
lost when the trap occurred while executing command substitution
of a built-in command.
j. The noclobber option now only applies to ordinary files.
k. The value of RANDOM changes when ksh creates a subshell.
l. Some syntax error messages had extra backslashes in them.
m. The expansion ${#name[number]} was not returning the string
length of array element number.
n. ksh did not work correctly in some cases when there was
no search permission in the current directory.
o. A pipeline would hang if the last element of the pipeline
was a shell construct that created a lot of child processes.
For example, cat file | while read line;do /bin/true;done
p. getopts has been fixed to agree with the description in
the man page. Previous ksh initialized OPTIND to 0, not 1,
and did not handle the case with a leading : as described.
q. If the name of the directory you are in contained a :,
a cd command to a relative pathname did always work correctly.
r. read -s now saves commands in the history file even when
read has been redirected from a file.
s. Some here documents containing command substitution with
a command that read from standard input didn't work
correctly.
t. (exec >&-; echo $(command) >&2) no longer runs command
with file descriptor 1 closed.
u. The <ESC>* command of vi and emacs inserted an extra
space when exactly one file matched.
v. A bug in vi edit mode caused a 'p' command to fail
after a 'D' command on some machines.
w. Reseeding the random number generator now guarantees
the same random sequence.
x. The results command substitution with built-in commands
that produced over 1K of output was sometimes incorrect.
y. set -A, without an argument, now produces an error message.
set -A name ... no longer causes the positional parameters
to be reset as an unintentional side effect.
z. Unterminated here-documents do not generate syntax errors
unless the -n option is on.
aa. Job control now works for functions as it did with ksh-i.
bb. The exit status of a pipeline whose last element is a built-in
command is now correct.
cc. ksh no longer dumps core when the last element of a pipeline
is a built-in command with an invalid parameter expansion.
dd. A trap on CHLD is no longer triggered by processes in the
current pipeline.
ee. One systems without readdir(), filename generation left
open files by mistake.
ff. traps ignored in a shell script were sometimes reset to
caught when functions were invoked.
gg. The trace (set -x ) of [[ x < y ]] and [[ x > y]] was incorrect.
hh. foobar() ; no longer core dumps.
ii. A trap on continue from within a select loop no correctly
when the trap occurs while reading from the terminal.
jj. The r command correctly handles numerical replacements so
that r 3=4 now works.
kk. Long lines (>3K) no longer causes a core dump with read.
ll. The command substitution $(name=value) no longer causes a
core dump.
mm. With the vi edit mode on some systems, extra characters were
left on the screen when the kill or erase characters were
preceded by a backslash.
nn. After a dot script that couldn't be found, a trap command
in a script could cause a core dump.
oo. Redirection with read of the form, read x <&3, was causing
the seek position on unit 3 to be positioned incorrectly.
pp. When an invalid alias name is given, ksh now prints the
correct error message.
qq. On systems with job control, if you escaped to a shell that
terminated abnormally, the calling process would hang.
3. Bugs fixed in 11/16/88b for default OPTIONS
a. ${x-"{a}"b} now expands to {a}b when x is not set.
b. ksh no longer core dumps with print -s when HISTFILE changed.
c. Tab stripping of numbered here-documents now works.
d. The deprecated, sh -t, now works the same as System V shell.
e. Attempting to list autoloaded functions no longer core dumps.
4. Bugs fixed in 11/16/88c for default OPTIONS
a. The process id for background jobs killed from scripts was
incorrect.
b. The wait built-in did not always return the correct value when
there was more than one background job.
c. ksh dumped core while in read with stderr redirected and in
emacs mode when it received an signal that had a trap set.
5. Bugs fixed in 11/16/88d for default OPTIONS
a. With the monitor option on with systems without job control,
login shells no longer print an extraneous 'running jobs'
message with a|b, where b is not found.
b. On systems where the behavior of the echo builtin depends
on the location of echo in the PATH variable, ksh no longer
core dumps if the PATH does not have a directory containing
an echo command.
c. return with an argument greater than 256 no longer causes
a core dump.
d. Tests on files whose names are null return false on all systems.
e. In some instances $0 was being incorrectly set after the
dot command was invoked with arguments inside a script.
f. The line number enclosed in [] when ksh reports a syntax error
has been corrected to be the first line of the command that
it is reading.
g. A trap on EXIT after a syntax error sometimes inserted an
extra character before the trap command text.
h. for i in ...; do echo ${i[@]};done no longer causes a subscript
error.
i. If the read built-in was interrupted, the terminal was not
always restored to the state it was in before read.
j. Output from background status changes with the monitor mode are
now written to standard error.
k. Executing a function interactively no longer causes SIGTERM
to be caught.
l. Using _ as an array variable no longer causes a core dump.
m. Unsetting a function more than once no longer causes core
dumps or other undesired side effects.
n. The code to process the substring operator ## was changed
so that the time was no longer quadratic in the length
of the string.
o. A bug that caused a job to disappear from the jobs table
after sending it a CONT signal has been fixed.
p. On systems where a null pathname is not the same as .,
the command 'cd .' would fail if the environment was cleared
before executing ksh.
q. In vi-mode, the screen did not always get refreshed correctly
when an insert followed a delete in front of a control character.
r. A bug in getopts that could cause a core dump when getopts
was called with fewer arguments on subsequent calls was fixed.
s. In vi-mode, ^V^M now enters ^M, not ^J on most systems.
t. On systems with job control, the kill command now sends out
SIGCONT signal after the signal has been sent, not before.
u. If the shell was interrupted while processing the .profile
file, the $ENV file was sometimes echoed to the terminal.
6. Bugs fixed in 11/16/88e for default OPTIONS
a. The code for item n from the d point release was omitted
but included in this release.
b. A script that uses edit modes now restores the terminal
modes with it receive a termination signal. Also, the
initial process group is restored for job control shells.
c. The standard input for background jobs was not set to
/dev/null in some cases where it should have been.
d. The value of $? now gets set correctly to indicate an
error after eval fails because of a syntax error.
e. The line discipline ^O bit is now cleared by ksh on some
systems that use sys/termios.h. Earlier versions only cleared
the ^O bit for BSD systems.
f. The suid_exec program has been modified to fix security
problems that could arise on systems that have symbolic links
but do not have the seteuid() system call.
g. The precedence for compound arithmetic assignment operators
such as *= was incorrectly set at the precedence of the
operator rather than the precedence for assignment.
h. The suspend character is now disabled when the last process of
a pipeline is not stoppable to prevent deadlock.
i. An extra secondary prompt was printed whenever a here-document
line contained a partial delimiter.
j. The EXIT trap will be executed for each function in the calling
chain when exit is called and when the maximum recursion level
(currently 128) is reached. Also return now behaves the same
as exit outside a function or a dot script and inside a subshell.
k. The value of $? no longer changes with a compound [[...]] command.
l. The jobs command now correctly reports 'no such job' when you
specify a pid that is not part of any running job.
m. The legal names for restricted shells have been reduced to
rsh, krsh, and rksh.
n. The vi, r, f, and @ directives now process ^V correctly. Also,
the r directive updates the screen correctly after you back
up to a control character and replace it with a non control char.
o. The command, $x > file, where x is not set, no longer core dumps.
p. When the read is redirected from a file it no longer issues the
secondary prompt when the line is continued with a \.
q. Variables that have attributes, but whose values are unset,
are now displayed with typeset, export, and readonly.
s. Set, without arguments, now displays the name-value pairs even
when preceded with a variable assignment list.
t. set -a now causes assignments of the form ${x=value} to get
the export attribute.
u. Arithmetic expansion, $((...)), inside here-documents no longer
causes subsequent parameter expansions to have extra \ inserted.
v. A bug that could cause the shell to go into an infinite
loop while creating the history has been fixed.
w. true and false are now built-in commands rather than aliases so
that correct results are produced when invoked with extraneous
arguments.
x. A tab character near the end of a long prompt no longer
causes the shell to hang.
y. An expansion of the form $((...)) is processed as an arithmetic
expansion only when the inner parenthesis match. Otherwise,
it will be processed as a nested command substitution.
z. Unsetting a function no longer causes the whence builtin to
give incorrect results for that name.
aa. Errors detected while expanding the PS1 prompt no longer
leave the shell in an undefined state.
bb. A bug that caused the shell to hang in read when a
coprocess (cmd |&) terminated has been fixed.
cc. A bug in getopts that caused a core dump when standard error
was redirected has been fixed.
dd. Testing for file access with null file names now returns false
now on all systems.
ee. A bug with [[...]] that caused quoting of patterns following the
= and != operators to give incorrect results has been fixed.
ff. A bug with nested $() of builtins, where inner $() was enclosed
in double quotes has been fixed.
gg. A bug that caused incorrect results when the -x trace option
was enabled and command substitution of a builtin command
produced no output was fixed.
hh. A bug that prevented set -e from exiting with pipelines has
been fixed.
ii. Interactive shells that trap TERM, and then reset the trap,
no longer exit when they receive a TERM signal.
jj. When inserting text in vi mode, the kill character deletes
only the insertion, not the whole line.
kk. A bug with eval that occurred when a word ended in a \ has
been fixed.
ll. typeset -f fname now returns false after function fname
has been unset.
mm. A bug in trap that caused a command to be interpreted as a
trap number if the first character was a digit has been fixed.
nn. A bug that caused a trap action of return to not work correctly
when an interrupt occurred during read has been fixed.
oo. A bug that could cause the exported value of an integer variable
that has been imported and modified to be incorrect has been fixed.
pp. A bug that caused set -n to incorrectly report subscript errors
has been fixed.
qq. The select list variable is now set to null when none of the
given choices is not selected. Previously it was unset.
rr. The pattern */foo, now also matches symbolic links to non-existent
files.
ss. The whence command no longer terminates when it doesn't find
one of the names.
7. Bugs fixed in 11/16/88f for default OPTIONS
a. Running command&, did not cause the standard input to be
redirected from /dev/null when following a built-in command
that had standard input redirected.
b. The path search no longer terminates with exec returns the
ENAMETOOLONG error message because the name need not be
too long for other directories in the PATH.
c. A bug that could cause a file descriptor to be closed-on-exec
has been fixed.
d. A bug with logical or (||) and logical and (&&) in an arithmetic
expression that occurred when the left hand side was an identifier
has been fixed.
e. A bug in the 'e' point release that caused typeset -f name to
return true when name was the name of a built-in has been fixed.
f. Changing the attribute of a variable from integer to left-justified
no longer causes a core dump when the variable doesn't have a value.
g. Changing the attribute of an array variable that contains a
single element whose index is non-zero, now works correctly.
h. A bug in the 'e' point release that prevented an exit command
within a .profile or ENV file to exit the shell has been fixed.
i. The edit modes no longer uses <cr> when positioning the cursor
with read when you do not specify a prompt. This prevents
characters from earlier echo and commands from being erased.
j. An error from a built-in executed in a subshell no longer
causes the sub-shell to exit, for example, (cd /dev/null;echo $?).
k. On invocation, ksh now causes the arguments after -c string to
become $0 $1 ... as it does with System V shell. Previously,
they became $1 $2 ....
l. A bug in the 'e' point release that caused some comment lines
in the ENV file to be entered into the history file has been
fixed.
m. A right parenthesis no longer causes an argument expected error
message when used a an operand within a unary test such as
test -d ")".
n. The whence builtin no longer prepends two // when an executable
is in / and / is a component in PATH. Also, explicit references
to . in the PATH variable will not show up in whence output.
o. The terminal state is correctly restored when read is interrupted
when in the edit modes. On some systems the state wasn't being
restored when a trap was set and an interrupt occurred.
p. When exit is invoked without an argument within a trap action,
the exit status is the last command executed before the trap.
The documentation was ambiguous, and the code was incorrect.
q. When xxx is not an executable, the error message produced in the
'e' point release by "ksh -u xxx" was misleading when.
s. A script that ends with an incomplete here-document caused
an extra new-line to be appended to the here-document text.
t. When an error occurs while processing the .profile file,
the error message now correctly begins with .profile.
u. When the effective userid was 0, test -x returned true
for all files. Now, at least one execute bit must be
on for regular files.
v. If the ERR trap was set inside a subshell (), it failed to
be executed when the last command was not a built-in and failed.
w. A trailing / at the end of a file pattern will match only
directories on all systems. Previously, the result depended
on the underlying operating system.
x. The command cd foo/.., where foo was found in CDPATH, works
correctly now.
y. A bug which caused a core dump when the vi 'v' directive was
entered by a user while entering the input to a read command
that is contained in a script has been fixed. The 'v' directive
is now an error in this case.
z. A bug that could cause /tmp files to remain when it terminates
because of an interrupt has been fixed.
aa. The readonly attribute can no longer be disabled with typeset +r.
bb. A bug which would cause the interrupt key to be ineffective
when the last part of a pipeline was a compound command on
job control systems was fixed.
cc. From an interactive shell, kill 0, now sends the given signal
to all process groups created by the shell, not just the
process group of the shell.
dd. A bug on some systems which caused pipelines started in a dot
script or ENV file to be listed on the jobs command was fixed.
ee. When the shell dies because of SIGHUP, it no longer resends
the signal to the current process group.
ff. kill -s without an options argument produces an error message
rather than a core dump.
8. Bugs fixed in 11/16/88g for default OPTIONS
a. A bug in test that caused "test -" to report that an argument
was expected has been fixed.
b. A bug in the 'f' point release of cd which caused ../.. not to
work when .. by itself was root was fixed.
c. CDPATH=value cd, now uses value for CDPATH.
d. A bug on some systems that caused pwd -P to change $PWD
has been fixed.
e. A bug is case which caused patterns of the form "$x"foo*, with
x was null, not to match correctly has been fixed.
f. A bug which cause the shell to hang when a long prompt ended
with a tab has been fixed.
g. A bug which could cause a core dump when a here document was
used inside command substitution was fixed.
h. With fc, when a user terminates the editor without changing
the file, the command is not run.
i. A script containing $(<file) or `<file` no longer aborts when
file cannot open.
j. A bug in emacs mode that could cause core dumps for lines > 256
characters was fixed.
k. The vi directive # only inserts # at the beginning of each
line not before each | and ;. Also, the kill character no
longer advances to the next line; even if echok is enabled.
l. Syntax errors are now correctly reported when ; is the first
token on a line with a brace group.
m. In emacs mode, the prompt is now drawn after two successive
line kills.
n. A bug that caused ksh to go into an infinite loop when the
last word of an eval was a command whose name was an alias
of an alias that was being evaluated has been fixed.
o. A bug on systems that support #! that prevented a script
whose name is the same as an executable in the PATH from
being invoked by full pathname from the directory
that its resides has been fixed.
p. Another bug on systems that support #! that caused a script
beginning with #! that did not have read permission to
recursively create processes has been fixed.
q. A bug that caused command completion not to work within
compound commands in some cases has been fixed.
r. A bug which could cause portions of a pipeline to join
a pre-existing job when early elements of the pipeline
completed before later ones were started has been fixed.
s. A bug on some systems which cause the shell to go into an
infinite loop in vi mode on devices for which read returns 0
(rather than -1) when disconnected has been fixed.
t. The DEBUG trap is now triggered after returning from a function.
Previously, the DEBUG trap was only triggered after a command
that was not a built-in was executed.
u. A bug that could cause a core dump on some systems for command
substitution of compound commands has been fixed.
v. A bug which could cause a core dump when more than CHILD_MAX
jobs had not completed and reported their completion status
was fixed.
w. A bug with caused typeset -ft to be ignored in the case
that the function was an autoload function that was not
yet loaded has been fixed.
x. A bug which caused /tmp files from here-documents defined in
a function not to be removed when the last command of a script
is exec has been fixed.
y. A bug in pipelines that contained one or more here documents in
a compound command that was not the last element has been
fixed. Earlier versions could produce the message
/tmp/sh123.1: cannot open.
z. A bug which casused 0> file to be treated as 1>file has been
fixed.
aa. A bug which prevented the terminal state from being restored
after resuming a stopped job has been fixed.
bb. A bug which caused some dot scripts not to work when file
descriptors less than 3 were closed has been fixed.
cc. A bug which caused a script to treat a stopped child process
as if it had completed has been fixed.
9. Bugs fixed in 11/16/88h for default OPTIONS
a. The bug fix for item cc above caused stopped jobs to hang
in some instances.
b. A bug that caused variables with attributes but without
values to cause the environment of grandchild shells to
be truncated has been fixed.
c. A bug introduced in the 'g' point release caused noclobber
to fail on systems that use type 0 for regular files has
been fixed.
d. A bug introduced in the 'g' point release with [[ x < y ]]
has been fixed.
e. A bug which could prevent the shell from timing out when
it fork() fails because of process limits has been fixed.
f. A bug which cause assignments to MAILCHECK to be ignored
in some cases has been fixed.
g. A bug in cd which caused cd -P .. to fail when you were
one directory down from the root to fail has been fixed.
h. The standard input is no longer redirected to /dev/null
for background jobs when it has been closed already.
10. Bugs fixed in 11/16/88i for default OPTIONS
a. A bug which caused searches to fail when the history
size was larger than 32K commands has been fixed.
b. A bug that could cause a coshell to hang on read
when a coprocess exited has been fixed.
c. A bug which could cause the ksh file to omit a file
during file expansion when a pattern occured on the
non-terminal component has been fixed.
d. When in interrupt occured while executing a dot script
that was called by a function, the script terminated
rather than executing a trap that may have been set
by the caller of the function.
e. A bug that cause the inline edit modes to be disabled
for one command after resuming a stopped /bin/su (or any
other command that uses raw mode but does not catch TSTP)
has been fixed.
f. A bug that caused return without arguments to work incorrectly
when a script is invoked via exec script inside a trap action
has been fixed.
g. A bug that caused read and select to use the global REPLY
variable when a local one has been defined has been fixed.
h. A bug that caused a core dump on some systems, when
an export variable that had additional attributes was removed
from the environment by a intermediate process has been fixed.
i. A bug that gave incorrect results when the addtibute of
an array variable was changed to an integer and then back was
fixed.
j. A bug that on some systems could cause a character to lost
when reading from a pipe has been fixed.
k. Another getopts bug that sometimes caused OPTIND to have the
wrong value before finishing option paring been fixed.
11. Bug fixes for specific non-default option combinations.
a. VERSIONS: MULTIBYTE no longer automatically enabled.
b. ACCT: The accdefs.h include removed.
c. BRACEPAT: No longer references location 0 and dumps core.
d. EMACSPLUS: Now enables additional EMACS features.
e. FLOAT: arith.c now compiles.
f. VI off: e_runvi is no longer unsatisfied external.
g. SUID_EXEC: execute only works when file descriptor 10 is open.
h. MULTIBYTE: Changing LANG and LC_TYPE now work.
i. FLOAT: $((...)) now works starting with 11/16/88g.
j. FS_3D: Noclobber allows top level version to be created.
k. DEVFD: bug fix in 11/16/88g.
l. FS_3D: test -V fixed for version files.
12. Other changes to 11/16/88[a-i]
a. sh.1 corrections.
b. sh.memo corrections.
c. ksh.mk modified to build in a cross development environment.
d. jobs.c modified to compile on machines that define the new
line discipline, NTTYDISC, but not the old, OTTYDISC.
e. RELEASE file corrections.
f. Some fixes to the stand-alone edit libraries.
g. VERSIONS option renamed FS_3D
h. POSIX option added to try out some IEEE POSIX 1003.2 proposals.
i. VFORK bugs fixed. I still don't recommend this option.
j. OLDTERMIO option added to 11/16/88c. This option allows older
termio ioctls, like TCGETA to be called after newer termios
ioctls' fail. Only effective for machines that have both termio.h
and termios.h.
k. universe builtin added to 11/16/88d for systems with /bin/universe.
l. code changes to 11/16/88e to remove sbrk() dependence and to
use the standard library version of malloc().
m. A tests directory that contains some ksh regression test has
been added. See the README file for more details.
n. The way builtin commands are coded has been changed to make
it easier to add builtins.
m. With 11/16/88g, the makelibedit script has been changed, and
a man page for the stand-alone libedit.a has been added.
13. Incompatibilities with 11/16/88 version.
None intentional.
--
Larry Cipriani -- l.v.ci...@att.com or attmail!lcipriani