How To Use the POIFS APIs
How To Use the POIFS APIs

This document describes how to use the POIFS APIs to read, write, and modify files that employ a POIFS-compatible data structure to organize their content.

Target Audience

This document is intended for Java developers who need to use the POIFS APIs to read, write, or modify files that employ a POIFS-compatible data structure to organize their content. It is not necessary for developers to understand the POIFS data structures, and an explanation of those data structures is beyond the scope of this document. It is expected that the members of the target audience will understand the rudiments of a hierarchical file system, and familiarity with the event pattern employed by Java APIs such as AWT would be helpful.

Glossary

This document attempts to be consistent in its terminology, which is defined here:

Directory
A special file that may contain other directories and documents.
DirectoryEntry
Representation of a directory within another directory.
Document
A file containing data, such as word processing data or a spreadsheet workbook.
DocumentEntry
Representation of a document within a directory.
Entry
Representation of a file in a directory.
File
A named entity, managed and contained by the file system.
File System
The POIFS data structures, plus the contained directories and documents, which are maintained in a hierarchical directory structure.
Root Directory
The directory at the base of a file system. All file systems have a root directory. The POIFS APIs will not allow the root directory to be removed or renamed, but it can be accessed for the purpose of reading its contents or adding files (directories and documents) to it.
The different ways of working with POIFS

The POIFS API provides ways to read, modify and write files and streams that employ a POIFS-compatible data structure to organize their content. The following use cases are covered:

Reading a File System

This section covers reading a file system. There are two ways to read a file system; these techniques are sketched out in the following table, and then explained in greater depth in the sections following the table.

Conventional Reading with POIFSFileSystem
  • Simpler API similar to reading a conventional file system.
  • Can read documents in any order.
  • Well tested read and write support.
  • If created from an InputStream, all files are resident in memory. (If created from a File, only certain key structures are)
Event-Driven Reading
  • Reduced footprint -- only the documents you care about are processed.
  • Improved performance -- no time is wasted reading the documents you're not interested in.
  • More complicated API.
  • Need to know in advance which documents you want to read.
  • No control over the order in which the documents are read.
  • No way to go back and get additional documents except to re-read the file system, which may not be possible, e.g., if the file system is being read from an input stream that lacks random access support.
Conventional Reading with POIFSFileSystem

In this technique for reading, certain key structures are loaded into memory, and the entire directory tree can be walked by the application, reading specific documents at leisure.

If you create a POIFSFileSystem instance from a File, the memory footprint is very small. However, if you createa a POIFSFileSystem instance from an input stream, then the whole contents must be buffered into memory to allow random access. As such, you should budget on memory use of up to 20% of the file size when using a File, or up to 120% of the file size when using an InputStream.

Preparation

Before an application can read a file from the file system, the file system needs to be opened and core parts processed. This is done using the org.apache.poi.poifs.filesystem.POIFSFileSystem class. Once the file system has been loaded into memory, the application may need the root directory. The following code fragment will accomplish this preparation stage:

Assuming no exception was thrown, the file system can then be read.

Reading the Directory Tree

Once the file system has been loaded into memory and the root directory has been obtained, the root directory can be read. The following code fragment shows how to read the entries in an org.apache.poi.poifs.filesystem.DirectoryEntry instance:

Reading a Specific Document

There are a couple of ways to read a document, depending on whether the document resides in the root directory or in another directory. Either way, you will obtain an org.apache.poi.poifs.filesystem.DocumentInputStream instance.

DocumentInputStream

The DocumentInputStream class is a simple implementation of InputStream that makes a few guarantees worth noting:

  • available() always returns the number of bytes in the document from your current position in the document.
  • markSupported() returns true.
  • mark(int limit) ignores the limit parameter; basically the method marks the current position in the document.
  • reset() takes you back to the position when mark() was last called, or to the beginning of the document if mark() has not been called.
  • skip(long n) will take you to your current position + n (but not past the end of the document).

The behavior of available means you can read in a document in a single read call like this:

The combination of mark, reset, and skip provide the basic mechanisms needed for random access of the document contents.

Reading a Document From the Root Directory

If the document resides in the root directory, you can obtain a DocumentInputStream like this:

Reading a Document From an Arbitrary Directory

A more generic technique for reading a document is to obtain an org.apache.poi.poifs.filesystem.DirectoryEntry instance for the directory containing the desired document (recall that you can use getRoot() to obtain the root directory from its file system). From that DirectoryEntry, you can then obtain a DocumentInputStream like this:

Event-Driven Reading

The event-driven API for reading documents is a little more complicated and requires that your application know, in advance, which files it wants to read. The benefit of using this API is that each document is in memory just long enough for your application to read it, and documents that you never read at all are not in memory at all. When you're finished reading the documents you wanted, the file system has no data structures associated with it at all and can be discarded.

Preparation

The preparation phase involves creating an instance of org.apache.poi.poifs.eventfilesystem.POIFSReader and to then register one or more org.apache.poi.poifs.eventfilesystem.POIFSReaderListener instances with the POIFSReader.

POIFSReaderListener

org.apache.poi.poifs.eventfilesystem.POIFSReaderListener is an interface used to register for documents. When a matching document is read by the org.apache.poi.poifs.eventfilesystem.POIFSReader, the POIFSReaderListener instance receives an org.apache.poi.poifs.eventfilesystem.POIFSReaderEvent instance, which contains an open DocumentInputStream and information about the document.

A POIFSReaderListener instance can register for individual documents, or it can register for all documents; once it has registered for all documents, subsequent (and previous!) registration requests for individual documents are ignored. There is no way to unregister a POIFSReaderListener.

Thus, it is possible to register a single POIFSReaderListener for multiple documents - one, some, or all documents. It is guaranteed that a single POIFSReaderListener will receive exactly one notification per registered document. There is no guarantee as to the order in which it will receive notification of its documents, as future implementations of POIFSReader are free to change the algorithm for walking the file system's directory structure.

It is also permitted to register more than one POIFSReaderListener for the same document. There is no guarantee of ordering for notification of POIFSReaderListener instances that have registered for the same document when POIFSReader processes that document.

It is guaranteed that all notifications occur in the same thread. A future enhancement may be made to provide multi-threaded notifications, but such an enhancement would very probably be made in a new reader class, a ThreadedPOIFSReader perhaps.

The following describes the three ways to register a POIFSReaderListener for a document or set of documents:

registers listener for all documents.
registerListener(POIFSReaderListener listener)
registers listener for a document with the specified name in the root directory.
registerListener(POIFSReaderListener listener, String name)
registers listener for a document with the specified name in the directory described by path
registerListener(POIFSReaderListener listener, POIFSDocumentPath path, String name)
POIFSDocumentPath

The org.apache.poi.poifs.filesystem.POIFSDocumentPath class is used to describe a directory in a POIFS file system. Since there are no reserved characters in the name of a file in a POIFS file system, a more traditional string-based solution for describing a directory, with special characters delimiting the components of the directory name, is not feasible. The constructors for the class are used as follows:

Constructor example Directory described
new POIFSDocumentPath() The root directory.
new POIFSDocumentPath(null) The root directory.
new POIFSDocumentPath(new String[ 0 ]) The root directory.
new POIFSDocumentPath(new String[ ] { "foo", "bar"} ) in Unix terminology, "/foo/bar".
new POIFSDocumentPath(new POIFSDocumentPath(new String[] { "foo" }), new String[ ] { "fu", "bar"} ) in Unix terminology, "/foo/fu/bar".
Processing POIFSReaderEvent Events

Processing org.apache.poi.poifs.eventfilesystem.POIFSReaderEvent events is relatively easy. After all of the POIFSReaderListener instances have been registered with POIFSReader, the POIFSReader.read(InputStream stream) method is called.

Assuming that there are no problems with the data, as the POIFSReader processes the documents in the specified InputStream's data, it calls registered POIFSReaderListener instances' processPOIFSReaderEvent method with a POIFSReaderEvent instance.

The POIFSReaderEvent instance contains information to identify the document (a POIFSDocumentPath object to identify the directory that the document is in, and the document name), and an open DocumentInputStream instance from which to read the document.

Writing a File System

Writing a file system is very much like reading a file system in that there are multiple ways to do so. You can load an existing file system into memory and modify it (removing files, renaming files) and/or add new files to it, and write it, or you can start with a new, empty file system:

POIFSFileSystem fs = new POIFSFileSystem();
The Naming of Names

There are two restrictions on the names of files in a file system that must be considered when creating files:

  1. The name of the file must not exceed 31 characters. If it does, the POIFS API will silently truncate the name to fit.
  2. The name of the file must be unique within its containing directory. This seems pretty obvious, but if it isn't spelled out, there'll be hell to pay, to be sure. Uniqueness, of course, is determined after the name has been truncated, if the original name was too long to begin with.
Creating a Document

A document can be created by acquiring a DirectoryEntry and calling one of the two createDocument methods:

createDocument(String name, InputStream stream)
  • Simple API
  • Increased memory footprint (document is in memory until file system is written).
createDocument(String name, int size, POIFSWriterListener writer)
  • Decreased memory footprint (only very small documents are held in memory, and then only for a short time).
  • More complex API.
  • Determining document size in advance may be difficult.
  • Lose control over when document is to be written.

Unlike reading, you don't have to choose between the in-memory and event-driven writing models; both can co-exist in the same file system.

Writing is initiated when the POIFSFileSystem instance's writeFilesystem() method is called with an OutputStream to write to.

The event-driven model is quite similar to the event-driven model for reading, in that the file system calls your org.apache.poi.poifs.filesystem.POIFSWriterListener when it's time to write your document, just as the POIFSReader calls your POIFSReaderListener when it's time to read your document. Internally, when writeFilesystem() is called, the final POIFS data structures are created and are written to the specified OutputStream. When the file system needs to write a document out that was created with the event-driven model, it calls the POIFSWriterListener back, calling its processPOIFSWriterEvent() method, passing an org.apache.poi.poifs.filesystem.POIFSWriterEvent instance. This object contains the POIFSDocumentPath and name of the document, its size, and an open org.apache.poi.poifs.filesystem.DocumentOutputStream to which to write. A DocumentOutputStream is a wrapper over the OutputStream that was provided to the POIFSFileSystem to write to, and has the responsibility of making sure that the document your application writes fits within the size you specified for it.

If you are using a POIFSFileSystem loaded from a File with readOnly set to false, it is also possible to do an in-place write. Simply call writeFilesystem() to have the (limited) in-memory structures synced with the disk, then close() to finish.

Creating a Directory

Creating a directory is similar to creating a document, except that there's only one way to do so:

DirectoryEntry createdDir = existingDir.createDirectory(name);
Using POIFSFileSystem Directly To Create a Document Or Directory

As with reading documents, it is possible to create a new document or directory in the root directory by using convenience methods of POIFSFileSystem.

DirectoryEntry Method Signature POIFSFileSystem Method Signature
createDocument(String name, InputStream stream) createDocument(InputStream stream, String name)
createDocument(String name, int size, POIFSWriterListener writer) createDocument(String name, int size, POIFSWriterListener writer)
createDirectory(String name) createDirectory(String name)
Modifying a File System

It is possible to modify an existing POIFS file system, whether it's one your application has loaded into memory, or one which you are creating on the fly.

Removing a Document

Removing a document is simple: you get the Entry corresponding to the document and call its delete() method. This is a boolean method, but should always return true, indicating that the operation succeeded.

Removing a Directory

Removing a directory is also simple: you get the Entry corresponding to the directory and call its delete() method. This is a boolean method, but, unlike deleting a document, may not always return true, indicating that the operation succeeded. Here are the reasons why the operation may fail:

  • The directory still has files in it (to check, call isEmpty() on its DirectoryEntry; is the return value false?)
  • The directory is the root directory. You cannot remove the root directory.
Changing a File's contents

There are two ways available to change the contents of an existing file within a POIFS file system. One is using a DocumentOutputStream, the other is with POIFSDocument.replaceContents

If you have available to you an InputStream to read the new File contents from, then the easiest way is via POIFSDocument.replaceContents. You would do something like:

Alternately, if you either have a byte array, or you need to write as you go along, then the OutputStream interface provided by DocumentOutputStream will likely be a better bet. Your code would want to look somewhat like:

For an example of an in-place change to one stream within a file, you can see the example org/apache/poi/hpsf/examples/ModifyDocumentSummaryInformation.java

Renaming a File

Regardless of whether the file is a directory or a document, it can be renamed, with one exception - the root directory has a special name that is expected by the components of a major software vendor's office suite, and the POIFS API will not let that name be changed. Renaming is done by acquiring the file's corresponding Entry instance and calling its renameTo method, passing in the new name.

Like delete, renameTo returns true if the operation succeeded, otherwise false. Reasons for failure include these:

  • The new name is the same as another file in the same directory. And don't forget - if the new name is longer than 31 characters, it will be silently truncated. In its original length, the new name may have been unique, but truncated to 31 characters, it may not be unique any longer.
  • You tried to rename the root directory.