|
From: | lilypond |
Subject: | Re: [Lilypond-auto] Issue 2839 in lilypond: Current master non-deterministic |
Date: | Sun, 16 Sep 2012 20:22:57 +0000 |
Comment #1 on issue 2839 by address@hidden: Current master non-deterministic http://code.google.com/p/lilypond/issues/detail?id=2839Well, we still have the oscillating graphviz tests and oscillating property accessing statistics. We also have some code that depends intermediately on memory layout order. The point is that the processing results should, after observing all object relations, be independent of those intermediate orders, and if they aren't, we have a different problem. If we make the intermediate layers work in the same order always, we won't have any indications that the subsequent processing stages are not doing their job of evaluating, prioritizing and arranging the intermediate results.
So we should be wary of enforcing a deterministic arrangement of data that is supposed to result in order-independent output after processing. If there is no good reason for order-dependent results, taking the easy way out is giving a false sense of security.
[Prev in Thread] | Current Thread | [Next in Thread] |