Bug 651349

+++ This bug was initially created as a clone of Bug #642575 +++
RHEL 5.6 clone

Description of problem:

Some filters filter devices wrongly.

The original cmd:
/sbin/pvs --config " devices { preferred_names = [\"^/dev/mapper/\"] write_cache_state=0 filter = [ \"a%/dev/mapper/36000402004ec06f56089aa1400000000|/dev/mapper/36000402004ec06f56089ab3a00000000|/dev/mapper/36000402004ec06f5608a865500000000%\", \"r%.*%\" ] } "

This is an irc log excerpt that shows agk trying to narrow down the issue.

[Tuesday 12 October 2010] [19:03:01] <agk> so it's do with pulling the suffix out
[Tuesday 12 October 2010] [19:03:12] <agk> /sbin/pvs --config " devices { write_cache_state=0 filter = [ \"a%8(9aa140000000|9ab3a0000000|a86550)%\", \"r%.*%\" ] } "
[Tuesday 12 October 2010] [19:03:15] <agk> vs
[Tuesday 12 October 2010] [19:03:26] <agk> /sbin/pvs --config " devices { write_cache_state=0 filter = [ \"a%8(9aa140000000|9ab3a0000000|a8655)%\", \"r%.*%\" ] } "
[Tuesday 12 October 2010] [19:03:44] <agk> removing that last '0' it starts to work again
[Tuesday 12 October 2010] [19:04:51] <agk> So the simplest case so far:
[Tuesday 12 October 2010] [19:04:56] <agk> "a%8(9aa140|9ab3a0|a86550)%\ fails
[Tuesday 12 October 2010] [19:05:11] <agk> a%8(9aa140|9ab3a|a86550)%\
[Tuesday 12 October 2010] [19:05:12] <agk> works
[Tuesday 12 October 2010] [19:05:36] <agk> so it must be going wrong when it sees a common suffix (0) and pulls it out of the expression
[Tuesday 12 October 2010] [19:06:30] <agk> but if there are only 2 things instead of 3 between the |s it works ok
[Tuesday 12 October 2010] [19:10:48] <agk> So: /tmp/bothcmd2 is my simplest test case so far
[Tuesday 12 October 2010] [19:10:55] <agk> bothcmds2
[Tuesday 12 October 2010] [19:11:07] <agk> it runs two supposedly identical cmds but gets different output


--- Additional comment from agk on 2010-10-22 17:03:21 EDT ---

Well the conclusion from irc was that removing a common suffix was broken.

But other parts of this code were changed since that was added, so we might find rather than fixing that, removing it no longer makes any noticeable performance difference.

So (1) try removing that optimisation and see how much it slows things down;
or (2) try to fix the optimisation.

--- Additional comment from agk on 2010-11-02 16:02:06 EDT ---

Patch added upstream for 2.02.76.

The optimiser was ignoring '| d e' and factoring out 'a' to convert
   a b | (a c | d e)
into
   a (b | c | d e).

--- Additional comment from agk on 2010-11-08 06:16:50 EST ---

http://sourceware.org/git/?p=lvm2.git;a=commitdiff;h=f56361e29fa9dc72a3966d0a45ab027a1d104923

http://sourceware.org/git/?p=lvm2.git;a=commitdiff;h=86a974c3c344de8e6265700a8dbbabb210c0026a