Bugzilla Bug 193979

When running revdep-rebuild from app-portage/gentoolkit-0.2.4-rc1, the progress
is halted when clean_vars() is called in a subshell. CPU usage goes to 100% and
stays there until the subshell is killed, it does not exit when the main
program is interuppted.

The line that triggers this is:
a="${a//+(\/\/)//}"


app-shells/bash-3.2_p17-r1 is used

Reproducible: Always

Steps to Reproduce:

This line also triggers the "hang"
a="${a//+([[:space:]])/$'\n'}"

After more investigation the part about it hanging was wrong.

The problem is that the line that handle spaces -> "\n" scale very poorly,
something like O(n²) where n is the number of characters in the string.

Created an attachment (id=132023) [edit]
revdep-rebuild-awkCleanVars-r448.patch

<sigh> I don't seem to be able to avoid using external commands for this
clean_vars(). This patch should fix the problem. Thanks to togge_ on
#gentoo-dev-help for helping me re-benchmark extglob.

gentoolkit-0.2.4_rc1 has been package masked due to this and other significant
problems.  Please downgrade back to gentoolkit-0.2.4_pre7

Created an attachment (id=132034) [edit]
revdep-rebuild-awkCleanVars-r448.patch

Same deal as my comment #3, only not broken (I typoed the regex the first time
through.)

(In reply to comment #4)
> gentoolkit-0.2.4_rc1 has been package masked due to this and other significant
> problems.  Please downgrade back to gentoolkit-0.2.4_pre7

I've noticed this in 0.2.4_pre7 too, when working on bug 177925. 8-10 control
files setting SEARCH_DIRS_MASK and the "configuring search environment" will
take ages with 100% CPU eaten (and even ctrl+c leaves that process crunching on
background). Also traced it down to clean_var(), didn't check which part of it
exactly... Thus I'm updating bug's name.

I've looked into this too. At a certain point, clear_vars is fed a variable
which is about 22000 chars (!) long. Most of the entries are dupes from
/usr/kde/3.5/lib and similar paths (from KDE). While the performance problem is
fixed (in a way) by using sed, I feel that generating said var might be
optimized. I don't really know how revdep-rebuild arrives at that monster of a
path list, so I might be wrong on that part.

*** Bug 209958 has been marked as a duplicate of this bug. ***

*** Bug 210101 has been marked as a duplicate of this bug. ***

Released in gentoolkit-0.2.4_rc2