Do you have any classes from Don's change in your local playpen?
On 3/17/2011 5:52 PM, Max Carlson wrote:
Not approved.
Can you update your patch? It applies cleanly but fails to compile in
trunk r18902. I see the patch was made for r18843.
compile:
[javac] Compiling 672 source files to
/Users/maxcarlson/openlaszlo/trunk2/WEB-INF/lps/server/build
[javac]
/Users/maxcarlson/openlaszlo/trunk2/WEB-INF/lps/server/src/org/openlaszlo/sc/WithThisAnalyzer.java:41:
cannot find symbol
[javac] symbol : method indexOf(org.openlaszlo.sc.parser.SimpleNode)
[javac] location: class org.openlaszlo.sc.parser.SimpleNode
[javac] int index = parent.indexOf(node);
[javac] ^
[javac] Note: Some input files use or override a deprecated API.
[javac] Note: Recompile with -Xlint:deprecation for details.
[javac] Note: Some input files use unchecked or unsafe operations.
[javac] Note: Recompile with -Xlint:unchecked for details.
[javac] 1 error
On 3/17/11 9:11 AM, André Bargull wrote:
Change bargull-20110317-tSC by bargull@Bargull02 on 2011-03-17 15:36:51
in /home/anba/src/svn/openlaszlo/trunk
for http://svn.openlaszlo.org/openlaszlo/trunk
Summary: improve compilation time for dhtml
New Features: LPP-9833 (speed up compilation time for dhtml)
Technical Reviewer: hqm
QA Reviewer: max
Doc Reviewer: (pending)
Overview:
By using Java VisualVM to profile the compilation time of the
amazon-demo in dhtml-mode, I've found a couple of hot spots which needed
too much time. With this change those hot spots should require less time
or even be removed completely.
#1 issue:
To read out the bounds of an image, DHTMLWriter used the superclass
method from ObjectWriter, but ObjectWriter simply transcoded the image
to swf and then used the jgenerator API to get the bounds. So this was
really a time consuming operation only to find out the image's size and
the transcoded swf wasn't even used anymore later since we're compiling
to dhtml.
#2 issue:
ImageMontageMaker used Thread.sleep() to wait for an image to load,
overall this led to about 1600ms of pure waiting time on my computer.
#3 issue:
Calling File#getCanonicalFile() is expensive (at least on Windows), but
it's used multiple times in DefaultFileResolver#resolveInternal(), even
for non-existing files. (Canonical files are necessary because the file
is later stored in a HashSet)
The other changes are extra improvements, but not linked to the named
hot spots.
Details:
ImageMontageMaker:
- use MediaTracker to track image loading instead of periodically
checking the image's width
ImageHelper:
- helper class to return the size of an image
ObjectWriter:
- split getResource(String,String,boolean) in several methods to improve
modularity
-- getAndCheckMimeType() returns the mime-type of the input, may throw
an error if the mime-type is unsupported
-- createResourceInfoElement() returns a jdom.Element for the resource
to collect resource information
-- importImage() transcodes the input image to swf and returns the
Resource object
ObjectWriter.Resource:
- added getBounds() method to return the resource's bounds, either from
the FlashDef if available, otherwise from the width/height properties
- getBounds() returns the bounds in TWIP units for backward compatibility
DHTMLWriter:
#importResource(File,String,Offset2D)
-- use Resource#getBounds() instead of accessing the resource's FlashDef
which may be null
-- use constants instead of magic numbers
#importImage(String,String,String,boolean)
-- override superclass method to avoid transcoding the image to swf if
possible
FileResolver:
- only call File#getCanonicalFile() when necessary
ViewSchema:
- removed isMouseEventAttribute() in favour of
containsMouseEventAttribute()
- this may only result in an almost negligible improvement, but well
it's there so why not
NodeModel:
- use containsMouseEventAttribute()
CompilerUtils:
- remove recursion and replace with plain iteration
Parser:
- removed xmltostring() and expandChildrenIncludes(), both methods
aren't used anywhere
#readExpanded(File,Set,CompilationEnvironment)
- don't copy the hashset for every round, just add the current file and
remove it later again
#evaluateSwitchStatement(Element,CompilationEnvironment)
- use emptyList() instead of returning a new, empty ArrayList
#expandIncludes(Element,Set,CompilationEnvironment)
- the method copied the element's children two times only to avoid
ConcurrentModificationError
- this was necessary because Element#getChildren() only returns a view
on Element#getContent()
- directly accessing and modifying Element#getContent() doesn't need any
copies, so that's clearly a better choice
- all ancestorIsDataSet() calls can simply be removed if you don't
descend into dataset elements in the first place
Tests:
profile application with visualvm, see snapshots at bug report
run amazon demo in dhtml
smokecheck x {swf10,dhtml}
Files:
M WEB-INF/lps/server/src/org/openlaszlo/media/ImageMontageMaker.java
A WEB-INF/lps/server/src/org/openlaszlo/media/ImageHelper.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/ObjectWriter.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/DHTMLWriter.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/ViewSchema.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/FileResolver.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/NodeModel.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/CompilerUtils.java
M WEB-INF/lps/server/src/org/openlaszlo/compiler/Parser.java
Changeset:
http://svn.openlaszlo.org/openlaszlo/patches/bargull-20110317-tSC.tar
