The use of subshells in get-parent-package, get-child-packages and is-parent-package is inadvisable (i.e. the "echo | while" structure). This: get-parent-package () { local target=$1 parent child echo "$KDE_DERIVATION_MAP" | while read parent child; do if [ "$target" == "$child" ]; then echo $parent return 1 fi done [ "$?" == "0" ] && die "Package $target not found in KDE_DERIVATION_MAP, please report bug" } can generate warnings about pipe closure, when the while loop terminates before the echo has finished (it's a race, so doesn't always occur). Note the tricks necessary with the return code to deal with that. is-parent-package() { echo "$KDE_DERIVATION_MAP" | while read parent child; do [[ "${parent}" == "$1" ]] && return 1 done return 0 } fails because the "return 1" is in a subshell it does not set the return code of the function. Also, if the function is to be used in an if statement (as appears to be the case); e.g. if is-parent-package $pkg}; then it needs to return 0 for true and 1 for false. In general, it's best to avoid using a pipe to do this sort of thing; not only is it slower as it spawns a sub-process, but variable assignments etc in the subshell are not reflected in the caller - it's easy to forget that the while loop is a subshell. There are several ways to avoid them; the easiest is to use HERE docs. I'll attach an example re-implementation of these functions using HERE docs that improves performance by 35% or more. One other comment on the eclass in general; using '[' is a little slower than '[[', because it means forking a call to '['. '[...]' can be exchanged for '[[...]]' everywhere in eclasses (some conditions may need to be modified for bash syntax, and quoting is not necessary with '[[') - this should yield a performance benefit.
Created attachment 90758 [details, diff] Reimplementation of get-parent-package, get-child-packages and is-parent-package using HERE docs
Created attachment 90759 [details] test script illustrating speed of various methods For interest, here's a shell script I used to try a 'for' loop variant and the HERE doc variant, to get timing information and to test that the functions behave the way they should in comparison with the originals (aside from bugs in the originals).
Kev, I'm all for you committing these changes if you're happy with them.
ok; committed :) (for the record, the thing I said about '[' vs '[[' is not true, as '[' is a bash builtin)
