Community
Participate
Working Groups
Build Identifier: I20100608-0911 Opening a big VHDL-file (2> MB) is very slow. The file is already part of my project so it's just opening that's costly. I'm using an Intel(R) Core(TM) i5 CPU 750 @ 2.67GHz and it takes 22 seconds. The profiler indicates that ParsetreeUtil.getOffset is using a lot of CPU cycles. The causes are threefold: 1) Iterating over a list by using an index (O(n^2) time) 2) Too much casting 3) Recursively visiting the parse tree within a recursive loop (I guess that's also (O(n^2)). Reproducible: Always
I have a patch that reduces opening time from 22 seconds to 14 seconds: /** * @param abstractNode * @return */ public static int getOffset(AbstractNode abstractNode) { if (abstractNode instanceof LeafNode) { LeafNode leafNode = (LeafNode) abstractNode; return leafNode.getTotalOffset(); } else { BooleanHolder hasLeafNodes = new BooleanHolder(false); CompositeNode compNode = (CompositeNode) abstractNode; return getOffset(compNode, hasLeafNodes); } } private static int getOffset(CompositeNode node, BooleanHolder hasLeafNodes) { for (AbstractNode child : node.getChildren()) { if (child instanceof CompositeNode) { CompositeNode childNode = (CompositeNode) child; BooleanHolder childrenHaveLeafNodes = new BooleanHolder(false); int offset = getOffset(childNode, childrenHaveLeafNodes); if (childrenHaveLeafNodes.isVal()) { hasLeafNodes.setVal(true); return offset; } } else { LeafNode leafNode = (LeafNode) child; hasLeafNodes.setVal(true); if (!leafNode.isHidden()) return leafNode.getTotalOffset(); } } return node.getTotalOffset(); }
A quick question. I can't use injection (I think) for this patch. Should I build my own Xtext version then or is there a better way?
(In reply to comment #2) > A quick question. I can't use injection (I think) for this patch. Should I > build my own Xtext version then or is there a better way? I'm not aware of a better way. Some remarks: 1) Iterating over a list by using an index (O(n^2) time) As the lists are RandomAccessLists, using an index is effectivly O(1). In fact, it should be slightly faster for really deep recursion since there is no need to instantiate an iterator. 2) Too much casting 3) Recursively visiting the parse tree within a recursive loop (I guess that's also (O(n^2)). This should be the main problem. Please note that the class ParsetreeUtil does no longer exist in Xtext 2.0 since we rewrote the node model.
Forgot to mention a little helper class: private final static class BooleanHolder { private boolean val; public BooleanHolder (boolean val) { this.val = val; } public boolean isVal() { return val; } public void setVal(boolean val) { this.val = val; } }
This should be obsolete with the new node model. Please reopen if the problem persists.
Closing all bugs that were set to RESOLVED before Neon.0
