Google Git Google Git
Sign in
apache/openoffice-org/refs/heads/services_https/./content/xml/optimization/optimization.html
blob: 283f6a083b20782f8ca73bd63e085831cfea0a68 [file]
<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 3.2//EN">
<HTML>
<head>
<META HTTP-EQUIV="CONTENT-TYPE" CONTENT="text/html; charset=iso-8859-1">
<TITLE></TITLE>
<META NAME="GENERATOR" CONTENT="StarOffice 6.0 (Win32)">
<META NAME="AUTHOR" CONTENT="Sascha Ballach">
<META NAME="CREATED" CONTENT="20010601;9311185">
<META NAME="CHANGEDBY" CONTENT="Daniel Vogelheim">
<META NAME="CHANGED" CONTENT="20010725;18030177">
</head>
<body>
<H1 ALIGN=CENTER>Loose list of possible optimizations in the xml
module</H1>
<P ALIGN=LEFT>This is a loose list of optimizations items, which the
guys which work on the xml project find out. This list expands if any
one found more items. The status tells what happens with this items.
If an item is implemented, then the status change and the build
version is add.</P>
<P ALIGN=LEFT>Here are also all items listed which have to do in
modules like sw, sc, graphics and so to optimize e.g. the save and
load performance of the new fileformat.</P>
<P ALIGN=LEFT>If anybody find also some items, where we may change
the code then he can write a mail to me
(<A HREF="mailto:sascha.ballach@germany.sun.com">Sascha.Ballach@germany.sun.com</A>)or
add a task into the <A HREF="http://xml.openoffice.org/project_issues.html">IssueZilla</A>.</P>
<P ALIGN=LEFT><A HREF="current_optimization_status.html">Here</A> can
anybody see the current state of our optimizations and the current
problems.</P>
<P ALIGN=LEFT><A HREF="current_optimization_status.sxc">Here</A> is
the complete overview as OpenOffice.org 6.0 file.</P>
<TABLE WIDTH=100% BORDER=1 BORDERCOLOR="#000000" CELLPADDING=4 CELLSPACING=0 FRAME=VOID RULES=GROUPS>
<COLGROUP>
<COL WIDTH=13*>
<COL WIDTH=20*>
</COLGROUP>
<COLGROUP>
<COL WIDTH=28*>
</COLGROUP>
<COLGROUP>
<COL WIDTH=36*>
</COLGROUP>
<COLGROUP>
<COL WIDTH=159*>
</COLGROUP>
<THEAD>
<TR VALIGN=TOP>
<TH COLSPAN=2 WIDTH=13%>
<P>Status</P>
</TH>
<TH WIDTH=11%>
<P>Module</P>
</TH>
<TH WIDTH=14%>
<P>Type</P>
</TH>
<TH WIDTH=62%>
<P>Description</P>
</TH>
</TR>
</THEAD>
<TBODY>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>ALL</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>At the moment is the <CODE>getPropertyStates</CODE> method
implemented as a loop which calls the <CODE>getPropertyState</CODE>
method for every Property. So the Property has to be searched
again and again. Therefore this method needs a lot of string
compares. We have to use a better implementation which use, that
the <CODE>getPropertyStates</CODE> method gets the PropertyNames
in alphabetical order.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>ALL</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>We have to implement a XMultiPropertySet interface and use it
instead of the XPropertySet interface to get and set the style
properties. (Done in some modules)</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>ALL</P>
</TD>
<TD WIDTH=14%>
<P>Code/Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Don’t export XML tokens as character pointer in an own object
file. This expand the size of each DLL which use any of these XML
tokens. This can change to a <CODE>GetToken(nId)</CODE> and
<CODE>IsToken(nId)</CODE>.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Remember the style names of the cells so they have not to get
again. (save spreadsheet document)</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Collect all cells and his styles and merge the cells to ranges
and set the styles on this ranges. (load spreadsheet document)</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Remember all API objects which are used again and again, so
they have not to create so often.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Find a better way to find out whether a cell contains a
annotation or not.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>The method <CODE>getFormula()</CODE> change in every document
which is not English the language to get the English formula.
This is not necessary.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>ALL</P>
</TD>
<TD WIDTH=14%>
<P>Code</P>
</TD>
<TD WIDTH=62%>
<P>Some Attributes contains the old virtual <CODE>importXML</CODE>
and <CODE>exportXML</CODE> methods. These can remove, because the
XML attribute export is changed</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>ALL</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Use MultipropertySet in the API for PageDescriptor’s</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>The package component has two major performance problems:</P>
<OL>
<LI><P>When saving, all zip file data is written to a buffer in
memory which is then passed to the UCB to be written to disk. If
there are multiple OLE objects, this could be crippling. Data
should be written to disk, buffered.</P>
<LI><P>When loading, whole streams are read into memory at once
and are then read from memory. They should be uncompressed and
written to temporary files and be read from there instead</P>
</OL>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SW</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>The XML load/save creates every time a DrawModel
(SwDoc::MakeDrawModel). This must change, so its only created if
its needed.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>On save and load of edit cells the method <CODE>GetTextForwarder()</CODE>
makes to much to often. This is not necessary. On every creation
of a new API Object like XTextRange, XTextCursor and so on on the
same cell creates a new object and set all the properties and so
on. The method should remember the created object and give this
again and again.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>C</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>On save of a document the method <CODE>GetOutputStringImpl</CODE>
ask for the numberformat on every cell. On documents with more
languages this needs very long, because the language is every
time changed. The calling method knows the numberformat, but
there is no way to give this to the method with the API.</P>
<P>This optimization is changed to a optimization of the
<CODE>ChangeIntl</CODE> method and the use of an English
NumberFormatter to ask whether a string cell use a NumberFormat.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Extend the API to get all formatted ranges with one call. This
means to make a possibility to get all formatted ranges with the
same format as XCellRanges object. So we have to ask for the
autostyle only one time.</P>
<P>The same should be possible while loading a document. Here we
need a possibility to create a XCellRanges object and give them
all ranges with the same format and then set the format on all
this ranges in one time.</P>
<P>Finally we have to use this extensions.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Use a English NumberFormatter to reduce the calls of the
<CODE>ChangeIntl<CODE> method.</CODE></CODE></P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>SC</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>We have also to optimize the <CODE>GetTextForwarder</CODE>
method for the saving of the PageStyles.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P>&nbsp;</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>The <CODE>GetKeyByAttrName</CODE> method should be optimized
so the string will be created only once and then the string
should be cached. This increase the performance, because the same
strings are used very often.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>Creating the <CODE>PropertySetMapper</CODE>s takes a long
time, mainly due to string creation. This can be sped up by
storing the string length in the maps.</P>
<P><I>dvo: As it turns out, the string length doesn't help a
whole lot. Copying the string from sal_Char* to sal_Unicode* and
the corresponding malloc seems to take most of the time. So I'm
afraid there's nothing we can do about this.</I></P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P>When loading/saving small documents, reading/setting the
document settings takes a long time. In the writer, reading the
document settings takes about as long as importing 200
paragraphs. This should be investigated.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The NumberFormatter takes quite long to find the
first data style for a particular language. Thus, exporting the
first textfield that uses a number format takes about as long as
exporting 25 paragraphs.</P>
<P ALIGN=LEFT><I>dvo: I can't think of any way to avoid this,
though.</I></P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The NumberFormatsSupplier is always obtained from
the document, even though it is not used. That takes quite some
time in small documents that don't use it.</P>
<P ALIGN=LEFT><I>dvo: creating the NumberFormatsSupplier
on-demand fixes the problem. However, this requires an
incompatible build, so it is not checked in yet.</I></P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>Resetting existing styles to default takes a long
time during import. (use <CODE>XMultiPropertsStates</CODE>
interface). Currently this uses getPropertySet and then sets them
to default one-by-one.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>P</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML/Writer</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>All styles for a style family are obtained from the
document through an <CODE>XIndexAccess</CODE>, which is then used
to iterate over all styles. The current Writer implementation for
that is really, really slow (probably true for the other apps as
well, but I haven't checked).</P>
<P ALIGN=LEFT><I>dvo: I have tried replacing the XIndexAccess
with an XNameAccess. This speeds obtaining all styles up from
280ms to 10ms, which is almost a factor of 30. However, I'm not
sure if other apps have the same performance characteristics.
Also, a properly implemented getByIndex should resolve this
issues, too. Therefore, I will not commit this 'fix'.</I></P>
<P ALIGN=LEFT><I>dvo: I submitted a performance bug about this
(#90251#)</I></P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>Writer</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The conversion between UI names and
internationalized names is slow.</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>Writer</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The IsPhysical property with styles is much slower
than most other properties.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML / Writer</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>In exportTextContentEnumeration, determining which
(relevant) service is implemented by the current XTextContent is
quite slow. For a simple document with 14 text content objects
(which run twice through the method) the total time needed to
determine the service takes as much time as exporting 80 simple
paragraphs.</P>
<P ALIGN=LEFT><I>dvo: A more radical approach to this problem
would be to give an XTextContent a required property which
returns this type as a string or an integer key. However, this
would require a modified TextContent service.</I></P>
<P ALIGN=LEFT><I>dvo: Most of the time seems to be spent in
asking a shape whether it happens to be of some type of service.
The way the interfaces are, there aren't too many way to speed up
the shape implementation. Simply asking the shape early solves
this problem nicely. For the 'einfach' Test-Dokument, checking
for shapes early reduces time for the entire
exportTextContentEnumeration by 10%.</I></P>
<P ALIGN=LEFT><I>dvo: I submitted a performance Bug #90240# about
this.</I></P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>I</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>Writer</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The 'FileLink' property of text sections is
implemented fairly slow.</P>
</TD>
</TR>
<TR BGCOLOR="#e6e6e6" VALIGN=TOP>
<TD WIDTH=5%>
<P>D</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>The FilterPropertyInfo_mpl::AddProperty method is
quite slow because it used an (albeit hacked-up) version of
insertion sort.</P>
<P ALIGN=LEFT><I>dvo: Using STL sort during GetApiNames() speeds
up the operation by about 30%.</I></P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=5%>
<P>C</P>
</TD>
<TD WIDTH=8%>
<P><BR>
</P>
</TD>
<TD WIDTH=11%>
<P>XML</P>
</TD>
<TD WIDTH=14%>
<P>Runtime</P>
</TD>
<TD WIDTH=62%>
<P ALIGN=LEFT>Importing a single control is about as slow as
importing 150 simple paragraphs.</P>
<P ALIGN=LEFT><I>dvo: The time is spent in loading several DLLs
required for the controls. This is probably unavoidable. The
second control is quite speedy.</I></P>
</TD>
</TR>
</TBODY>
</TABLE>
<P>Last change: 07/25/2001</P>
<P><B>Agenda</B>:</P>
<P STYLE="margin-bottom: 0.01cm">Status Symbols:</P>
<TABLE WIDTH=100% BORDER=0 CELLPADDING=0 CELLSPACING=0>
<COL WIDTH=16*>
<COL WIDTH=240*>
<TR VALIGN=TOP>
<TD WIDTH=6%>
<P>I</P>
</TD>
<TD WIDTH=94%>
<P>Idea
</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=6%>
<P>D</P>
</TD>
<TD WIDTH=94%>
<P>Done</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=6%>
<P>P</P>
</TD>
<TD WIDTH=94%>
<P>in Progress</P>
</TD>
</TR>
<TR VALIGN=TOP>
<TD WIDTH=6%>
<P>C</P>
</TD>
<TD WIDTH=94%>
<P>Cancelled</P>
</TD>
</TR>
</TABLE>
<P STYLE="margin-bottom: 0.01cm"><BR><BR>
</P>
</body>
</HTML>