This HOW-TO is organized in three sections. You should read them sequentially because the later sections build upon the earlier ones.
The first thing you should understand is that properties are stored in separate documents inside the POI filesystem. (If you don't know what a POI filesystem is, read the POIFS documentation.) A document in a POI filesystem is also called a stream.
The following example shows how to read a POI filesystem's
"title" property. Reading other properties is similar. Consider the API
documentation of org.apache.poi.hpsf.SummaryInformation
to
learn which methods are available!
The standard properties this section focusses on can be found in a document called \005SummaryInformation located in the root of the POI filesystem. The notation \005 in the document's name means the character with the decimal value of 5. In order to read the title, an application has to perform the following steps:
SummaryInformation
from
that document.
SummaryInformation
instance's
getTitle()
method.
Sounds easy, doesn't it? Here are the steps in detail.
An application that wants to open a document in a POI filesystem (POIFS) proceeds as shown by the following code fragment. (The full source code of the sample application is available in the examples section of the POI source tree as ReadTitle.java.
The first interesting statement is
It creates a
org.apache.poi.poifs.eventfilesystem.POIFSReader
instance
which we shall need to read the POI filesystem. Before the application
actually opens the POI filesystem we have to tell the
POIFSReader
which documents we are interested in. In this
case the application should do something with the document
\005SummaryInformation.
This method call registers a
org.apache.poi.poifs.eventfilesystem.POIFSReaderListener
with the POIFSReader
. The POIFSReaderListener
interface specifies the method processPOIFSReaderEvent
which processes a document. The class
MyPOIFSReaderListener
implements the
POIFSReaderListener
and thus the
processPOIFSReaderEvent
method. The eventing POI filesystem
calls this method when it finds the \005SummaryInformation
document. In the sample application MyPOIFSReaderListener
is
a static class in the ReadTitle.java source file.
Now everything is prepared and reading the POI filesystem can start:
The following source code fragment shows the
MyPOIFSReaderListener
class and how it retrieves the
title.
The line
declares a SummaryInformation
variable and initializes it
with null
. We need an instance of this class to access the
title. The instance is created in a try
block:
The expression event.getStream()
returns the input stream
containing the bytes of the property set stream named
\005SummaryInformation. This stream is passed into the
create
method of the factory class
org.apache.poi.hpsf.PropertySetFactory
which returns
a org.apache.poi.hpsf.PropertySet
instance. It is more or
less safe to cast this result to SummaryInformation
, a
convenience class with methods like getTitle()
,
getAuthor()
etc.
The PropertySetFactory.create
method may throw all sorts
of exceptions. We'll deal with them in the next sections. For now we just
catch all exceptions and throw a RuntimeException
containing the message text of the origin exception.
If all goes well, the sample application retrieves the title and prints it to the standard output. As you can see you must be prepared for the case that the POI filesystem does not have a title.
Please note that a Microsoft Office document does not necessarily
contain the \005SummaryInformation stream. The documents created
by the Microsoft Office suite have one, as far as I know. However, an
Excel spreadsheet exported from StarOffice 5.2 won't have a
\005SummaryInformation stream. In this case the applications
won't throw an exception but simply does not call the
processPOIFSReaderEvent
method. You have been warned!
A couple of additional standard properties are not contained in the \005SummaryInformation stream explained above, for example a document's category or the number of multimedia clips in a PowerPoint presentation. Microsoft has invented an additional stream named \005DocumentSummaryInformation to hold these properties. With two minor exceptions you can proceed exactly as described above to read the properties stored in \005DocumentSummaryInformation:
SummaryInformation
by
DocumentSummaryInformation
.And of course you cannot call getTitle()
because
DocumentSummaryInformation
has different query methods. See
the Javadoc API documentation for the details!
In the previous section the application simply caught all exceptions and was in no way interested in any details. However, a real application will likely want to know what went wrong and act appropriately. Besides any IO exceptions there are three HPSF resp. POI specific exceptions you should know about:
NoPropertySetStreamException
:PropertySet
instance from a stream that is not a
property set stream. (SummaryInformation
and
DocumentSummaryInformation
are subclasses of
PropertySet
.) A faulty property set stream counts as not
being a property set stream at all. An application should be prepared to
deal with this case even if it opens streams named
\005SummaryInformation or
\005DocumentSummaryInformation only. These are just names. A
stream's name by itself does not ensure that the stream contains the
expected contents and that this contents is correct.
UnexpectedPropertySetTypeException
SummaryInformation
or
DocumentSummaryInformation
) but the provided property
set is not of that type.MarkUnsupportedException
InputStream.mark(int)
operation. The POI filesystem uses
the DocumentInputStream
class which does support this
operation, so you are safe here. However, if you read a property set
stream from another kind of input stream things may be
different.Many Microsoft Office documents contain embedded
objects, for example an Excel sheet on a page in a Word
document. Embedded objects may have property sets of their own. An
application can open these property set streams as described above. The
only difference is that they are not located in the POI filesystem's root
but in a nested directory instead. Just register a
POIFSReaderListener
for the property set streams you are
interested in. For example, the POIBrowser application in the
contrib section tries to open each and every document in a POI filesystem
as a property set stream. If this operation was successful it displays the
properties.
Now comes the real hardcode stuff. As mentioned above,
SummaryInformation
and
DocumentSummaryInformation
are just special cases of the
general concept of a property set. This concept says that a
property set consists of properties and that each
property is an entity with an ID, a
type, and a value.
Okay, that was still rather easy. However, to make things more complicated, Microsoft in its infinite wisdom decided that a property set shalt be broken into one or more sections. Each section holds a bunch of properties. But since that's still not complicated enough, a section may have an optional dictionary that maps property IDs to property names - we'll explain later what that means.
The procedure to get to the properties is the following:
PropertySetFactory
class to
create a PropertySet
object from a property set stream. If
you don't know whether an input stream is a property set stream, just
try to call PropertySetFactory.create(java.io.InputStream)
:
You'll either get a PropertySet
instance returned or an
exception is thrown.PropertySet
's method getSections()
to get the sections contained in the property set. Each section is
an instance of the Section
class.F29F85E0-4FF9-1068-AB-91-08-00-2B-27-B3-D9
. You can
get the format ID with Section.getFormatID()
.Section
can be retrieved
with Section.getProperties()
. The result is an array of
Property
instances.Property
class has methods to retrieve them.Let's have a look at a sample Java application that dumps all property set streams contained in a POI file system. The full source code of this program can be found as ReadCustomPropertySets.java in the examples area of the POI source code tree. Here are the key sections:
The most important package the application needs is
org.apache.poi.hpsf.*
. This package contains the HPSF
classes. Most classes named below are from the HPSF package. Of course we
also need the POIFS event file system's classes and java.io.*
since we are dealing with POI I/O. From the java.util
package
we use the List
and Iterator
class. The class
org.apache.poi.util.HexDump
provides a methods to dump byte
arrays as nicely formatted strings.
The POIFSReader
is set up in a way that the listener
MyPOIFSReaderListener
is called on every file in the POI file
system.
The listener class tries to create a PropertySet
from each
stream using the PropertySetFactory.create()
method:
Creating the PropertySet
is done in a try
block, because not each stream in the POI file system contains a property
set. If it is some other file, the
PropertySetFactory.create()
throws a
NoPropertySetStreamException
, which is caught and
logged. Then the program continues with the next stream. However, all
other types of exceptions cause the program to terminate by throwing a
runtime exception. If all went well, we can print the name of the property
set stream.
The next step is to print the number of sections followed by the sections themselves:
The PropertySet
's method getSectionCount()
returns the number of sections.
To retrieve the sections, use the getSections()
method. This method returns a java.util.List
containing
instances of the Section
class in their proper order.
The sample code shows a loop that retrieves the Section
objects one by one and prints some information about each one. Here is
the complete body of the loop:
The first method called on the Section
instance is
getFormatID()
. As explained above, the format ID of the
first section in a property set determines the type of the property
set. Its type is ClassID
which is essentially a sequence of
16 bytes. A real application using its own type of a custom property set
should have defined a unique format ID and, when reading a property set
stream, should check the format ID is equal to that unique format ID. The
sample program just prints the format ID it finds in a section:
As you can see, the getFormatID()
method returns a
ClassID
object. An array containing the bytes can be
retrieved with ClassID.getBytes()
. In order to get a nicely
formatted printout, the sample program uses the hex()
helper
method which in turn uses the POI utility class HexDump
in
the org.apache.poi.util
package. Another helper method is
out()
which just saves typing
System.out.println()
.
Before getting the properties, it is possible to find out how many
properties are available in the section via the
Section.getPropertyCount()
. The sample application uses this
method to print the number of properties to the standard output:
Now its time to get to the properties themselves. You can retrieve a
section's properties with the method
Section.getProperties()
:
As you can see the result is an array of Property
objects. This class has three methods to retrieve a property's ID, its
type, and its value. The following code snippet shows how to call
them:
The output of the sample program might look like the following. It
shows the summary information and the document summary information
property sets of a Microsoft Word document. However, unlike the first and
second section of this HOW-TO the application does not have any code
which is specific to the SummaryInformation
and
DocumentSummaryInformation
classes.
There are some interesting items to note:
Properties in the same section are distinguished by their IDs. This is similar to variables in a programming language like Java, which are distinguished by their names. But unlike variable names, property IDs are simple integral numbers. There is another similarity, however. Just like a Java variable has a certain scope (e.g. a member variables in a class), a property ID also has its scope of validity: the section.
Two property IDs in sections with different section format IDs don't have the same meaning even though their IDs might be equal. For example, ID 4 in the first (and only) section of a summary information property set denotes the document's author, while ID 4 in the first section of the document summary information property set means the document's byte count. The sample output above does not show a property with an ID of 4 in the first section of the document summary information property set. That means that the document does not have a byte count. However, there is a property with an ID of 4 in the second section: This is a user-defined property ID - we'll get to that topic in a minute.
So, how can you find out what the meaning of a certain property ID in
the summary information and the document summary information property set
is? The standard property sets as such don't have any hints about the
meanings of their property IDs. For example, the summary
information property set does not tell you that the property ID 4 stands
for the document's author. This is external knowledge. Microsoft defined
standard meanings for some of the property IDs in the summary information
and the document summary information property sets. As a help to the Java
and POI programmer, the class PropertyIDMap
in the
org.apache.poi.hpsf.wellknown
package defines constants
for the "well-known" property IDs. For example, there is the
definition
These definitions allow you to use symbolic names instead of numbers.
In order to provide support for the other way, too, - i.e. to map
property IDs to property names - the class PropertyIDMap
defines two static methods:
getSummaryInformationProperties()
and
getDocumentSummaryInformationProperties()
. Both return
java.util.Map
objects which map property IDs to
strings. Such a string gives a hint about the property's meaning. For
example,
PropertyIDMap.getSummaryInformationProperties().get(4)
returns the string "PID_AUTHOR". An application could use this string as
a key to a localized string which is displayed to the user, e.g. "Author"
in English or "Verfasser" in German. HPSF might provide such
language-dependend ("localized") mappings in a later release.
Usually you won't have to deal with those two maps. Instead you should
call the Section.getPIDString(int)
method. It returns the
string associated with the specified property ID in the context of the
Section
object.
Above you learned that property IDs have a meaning in the scope of a section only. However, there are two exceptions to the rule: The property IDs 0 and 1 have a fixed meaning in all sections:
Property ID | Meaning |
---|---|
0 | The property's value is a dictionary, i.e. a mapping from property IDs to strings. |
1 | The property's value is the number of a codepage, i.e. a mapping from character codes to characters. All strings in the section containing this property must be interpreted using this codepage. Typical property values are 1252 (8-bit "western" characters) or 1200 (16-bit Unicode characters). |
A property is nothing without its value. It is stored in a property set
stream as a sequence of bytes. You must know the property's
type in order to properly interpret those bytes and
reasonably handle the value. A property's type is one of the so-called
Microsoft-defined "variant types". When you call
Property.getType()
you'll get a long
value
which denoting the property's variant type. The class
Variant
in the org.apache.poi.hpsf
package
holds most of those long
values as named constants. For
example, the constant VT_I4 = 3
means a signed integer value
of four bytes. Examples of other types are VT_LPSTR = 30
meaning a null-terminated string of 8-bit characters, VT_LPWSTR =
31
which means a null-terminated Unicode string, or VT_BOOL
= 11
denoting a boolean value.
In most cases you won't need a property's type because HPSF does all the work for you.
When an application wants to retrieve a property's value and calls
Property.getValue()
, HPSF has to interpret the bytes making
out the value according to the property's type. The type determines how
many bytes the value consists of and what
to do with them. For example, if the type is VT_I4
, HPSF
knows that the value is four bytes long and that these bytes
comprise a signed integer value in the little-endian format. This is
quite different from e.g. a type of VT_LPWSTR
. In this case
HPSF has to scan the value bytes for a Unicode null character and collect
everything from the beginning to that null character as a Unicode
string.
The good new is that HPSF does another job for you, too: It maps the variant type to an adequate Java type.
Variant type: | Java type: |
---|---|
VT_I2 | java.lang.Integer |
VT_I4 | java.lang.Long |
VT_FILETIME | java.util.Date |
VT_LPSTR | java.lang.String |
VT_LPWSTR | java.lang.String |
VT_CF | byte[] |
VT_BOOL | java.lang.Boolean |
The bad news is that there are still a couple of variant types HPSF does not yet support. If it encounters one of these types it returns the property's value as a byte array and leaves it to be interpreted by the application.
An application retrieves a property's value by calling the
Property.getValue()
method. This method's return type is the
abstract Object
class. The getValue()
method
looks up the property's variant type, reads the property's value bytes,
creates an instance of an adequate Java type, assigns it the property's
value and returns it. Primitive types like int
or
long
will be returned as the corresponding class,
e.g. Integer
or Long
.
The property with ID 0 has a very special meaning: It is a dictionary mapping property IDs to property names. We have seen already that the meanings of standard properties in the summary information and the document summary information property sets have been defined by Microsoft. The advantage is that the labels of properties like "Author" or "Title" don't have to be stored in the property set. However, a user can define custom fields in, say, Microsoft Word. For each field the user has to specify a name, a type, and a value.
The names of the custom-defined fields (i.e. the property names) are stored in the document summary information second section's dictionary. The dictionary is a map which associates property IDs with property names.
The method Section.getPIDString(int)
not only returns with
the well-known property names of the summary information and document
summary information property sets, but with self-defined properties,
too. It should also work with self-defined properties in self-defined
sections.
The property with ID 1 holds the number of the codepage which was used to encode the strings in this section. The present HPSF codepage support is still very limited: When reading property value strings, HPSF distinguishes between 16-bit characters and 8-bit characters. 16-bit characters should be Unicode characters and thus be okay. 8-bit characters are interpreted according to the platform's default character set. This is fine as long as the document being read has been written on a platform with the same default character set. However, if you receive a document from another region of the world and want to process it with HPSF you are in trouble - unless the creator used Unicode, of course.
There are still some aspects of HSPF left which are not covered by this HOW-TO. You should dig into the Javadoc API documentation to learn further details. Since you've struggled through this document up to this point, you are well prepared.