[Scons-dev] max drift * issue 2001

[Tom Tanner (BLOOMBERG/ LONDON)]

I tried this and it actually makes a null build go slightly slower which is a little contrary to what I expected. I'm using plain md5 decider as Decider('md5-timestamp') checking goes horribly wrong for us.

It seems to me that this new check isn't entirely correct anyway.

Wouldn't the correct thing be to recalculate the content signature whenever you change the file - that is to say, whenever you copy it from cache or generate the file as the result of a build action?

Isn't that the necessary and sufficient condition?
----- Original Message -----
From: jason.l.kenny at intel.com
To: TOM TANNER (BLOOMBERG/ LONDON), scons-dev at scons.org
At: Oct  2 2012 17:09:06

I believe this is a misnomer.

What it is claim is that the build started 25% early, not that it was 25% faster. This only means that we falsely built more stuff, by making it out of data when using a time-stamp only  method for an update check. I found that this logic did not help anything built really go faster. It only meant I build a lot more, depending on what changed. I found that other methods gave  factor of ten or more build time speedup, and did not require build items that are up to date.

However the reason why this did not go in, probably was that it did not get added when we had SVN only. If you feel that this helps you out that much and you get the speed gains you like with it.. you should resubmit this as a push request.

Jason


-----Original Message-----
From: scons-dev-bounces at scons.org [mailto:scons-dev-bounces at scons.org] On Behalf Of Tom Tanner (BLOOMBERG/ LONDON)
Sent: Tuesday, October 02, 2012 11:02 AM
To: scons-dev at scons.org
Subject: [Scons-dev] max drift * issue 2001

There's a patch suggestion from a long time ago for getting rid of max drift which basically returns the cached signature if and only if the timestamps are identical with a claim of a 25% performance improvement (presumably due to not continually calling time())

Is there any reason this patch didn't go in?
_______________________________________________
Scons-dev mailing list
Scons-dev at scons.org
http://two.pairlist.net/mailman/listinfo/scons-dev