apologies in advance for this lengthy mail, but I am at a total loss
currently:
after upgrading my mac from 10.8 to 10.9 today I see regular/frequent
memory faults of ksh in my setup. this happens with the native ksh coming
with macos 10.9 (ksh 93u) as well as a binary of ksh 93u+ (but that one
was compiled under 10.8: my attempt to recompile from the ast-sources
failed today -- I might come back to the list in that respect, too, but
first have to look into it a bit).
the memory faults happen in some ksh-functions (actually they are adjusted
so as to work with bash and zsh as well) which I regularly have used for
some years now without any problems. together they act as a drop in
replacement for `cd' (by logging all `cd' actions to a file and
pattern-matching arguments to subsequent `cd' actions against this
"database", using the first hit after sorting all entries by frequency of
occurences. the whole script is ~ 1000 lines, so too long for this mail
(but I will gladly provide it if there is interest: it even is quite
useful in my view to rapidly navigate deep directory trees ...). instead I
include the part where the crash does happen below to give you an idea of
the context:
8<
function sdname { ## regex
#--
# ensure "default" behaviour of `cd' and return immediately
# if a path to an existing directory is specified
if [[ -z "$*" ]]; then
echo "$HOME"
return
elif [[ "$*" == "-" ]]; then
#zsh does not echo a single `-' (why not???), therefore, we here use
`printf'
#which works with all three targeted shells.
printf -- '-\n'
return
elif [[ -d "$*" ]]; then
echo "$*"
return
fi
#
typeset pat idx awkpat dname match matches
#treat $* as single argument and protect all `/' to allow them as part
of
#the search pattern:
#
pat="${*//\//\/}"
idx=${pat#$sdprefix}
if [[ $idx != $pat ]] && [[ -z ${idx/+([0-9])} ]]; then
#look up by numeric index
idx=${idx:-1} #`cd =' handled as `cd =1' (or whatever
prefix is used instead of `=')
awkpat="NR == $idx"
else
#look up by regex pattern matching.
#remember: the dir names are in field no. 3 in `sdlist'.
awkpat="\$3 ~ /$pat/"
fi
matches=$(echo "$sdlist" |\
$awk -F"\t" "$awkpat"' {
sub(/^~/, "'"$HOME"'", $3)
print $3
}')
# select first actually existing directory
typeset IFS=$'\n'
for match in $matches; do
[[ -d $match ]] && {
dname=$match
break
}
done
#if `dname' is empty at this point, we simply pass through $*
#(and let the final `cd' fail ...).
#
dname=${dname:-"$*"}
echo "$dname"
}
function sd { ## regex
#--
typeset dname
typeset IFS=$'\n'
dname=$(sdname "$*")
command cd "$dname" 2>/dev/null
(( $? == 0 )) || {
typeset -i lines
lines=$(sdlinecount "$sdlog")
(( lines > sdlines )) && {
typeset sdlinesOrig=$sdlines
((sdsilent == 1)) || echo "no match so far - extending stack"
sdirs -l $lines
dname=$(sdname "$*")
matches=$(sdirs "$*" | awk -F'\t' 'NR > 1 {print $3}')
sdirs -l $sdlinesOrig # reset the stack to original length
command cd "$dname" 2>/dev/null
(( $? == 0 )) && {
((sdsilent == 1)) || {
echo "using first valid match from:"
echo "$matches"
}
} || { echo "No matching directory"; return 1; }
}
}
# avoid logging `cd' to one of (home|current|root)-dir, since otherwise
#
# -- $HOME would always be at the top of the stack (really undesirable),
# -- an accidental/reaffirmative `cd .' would artificially increase
#frequency of visits to the respective directory (matter of taste),
# -- `/' would be logged although it's always reached as explicit path
#and the respective stack entry thus would be never used.
#
if [[ "$PWD" == !("$HOME"|"$OLDPWD"|/) ]]; then
typeset entry
# the following lines work with, _both_, ksh and bash which
# handle tilde expansion slightly different. when combining both
# lines into a single one (i.e. avoiding definition of `entry'),
# tilde expansion cannot be consistently suppressed in both
# shells. seemingly. (remember, that we need `"' quoting to
# account for possible blanks in file names).
#
entry=${PWD/#$HOME/\~}
sdnew+="$entry"$'\n'
sdlog=${sdlog0}${sdnew%$'\n'}
sdlist=$(sdstack)
fi
}
function cd {
sd "$*"
}
8<
where `sdlog' contains newline-separated directory names (those contained
in the database, i.e. visited previously) and
`sdlist' contains newline-separated records of the format
frequency TAB sorting-ra
