Roland Mainz wrote:
Darren J Moffat wrote:I don't see how the fix under review for this bug can be improving performance by using ksh93. Which is the rationale in the bug for switching nightly to ksh93.In my original review request email I wrote this: -- snip -- - This patch is _not_ intended to improve performace. This patch only switches "nightly.sh" from ksh88 to ksh93 to be in sync with "bldenv.sh" that users can actually use ksh93 statements in the "opensolaris.sh" config file and to prepare for later cleanup steps (my fear is that this patch will trigger a larger discussion again and a large patch may suffer from bitrot again (at least my last two nightly cleanup patches suffered that fate and I am not exactly willing anymore to write patches for /dev/null)). As said I now try to split the original monster-patch into smaller pieces to get them added to the tree faster. -- snip --
The bug needs to be updated with that - or a different bug number used since the bug doesn't say anything about bldenv.sh already using ksh93.
Given bldenv.sh already uses ksh93 then I think switching nightly to do so as well is reasonable.
As said this patch does not aim at performance - I'll do that later (as
a side-note: The script is I/O-dominated in many places and therefore
fixing the places which trigger the { |open()|, |stat()|, |lseek()|,
write()|, |close()| }-storms to |write()| ([1]) will greatly improve the
performace on NFS filesystems. Getting rid of the |fork()|-bomb
behaviour in other places will help a lot, too (e.g. the massive "egrep"
chain somewhere in the script can be reduced to one array and one (or
more (assuming this helps Niagara machines)) egrep statement ([2]) and
other places scream for improvements, too. Loudly.)).
But will we actually notice any difference given a build is much more likely dominated by cc and make ?
For example if I've just done a full clobber build and make no changes and do an incremental will these future proposed changes to nightly save any noticeable wall clock time ?
Do any of these changes mean that nightly no longer works with ksh88 ?
Given bldenv.sh is already ksh93 I see no issue with moving nightly that way.
-- Darren J Moffat _______________________________________________ opensolaris-code mailing list opensolaris-code@opensolaris.org http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
