CHANGELOG

Changes in PDFxStream v4.0.0

Added support for extracting right-to-left (RTL) and bidirectional (bidi) text. This includes transparent and automatic handling of character reordering and bracket mirroring when using any Snowtide-provided OutputHandler for extracting text, as well as Java API updates to add the Span abstraction (see com.snowtide.pdf.layout.Line.getSpans()) for document model access to RTL and bidi text.
PDFxStream now requires Java 11 or higher.
To avoid unplanned service terminations due to accidental support enrollment expirations, PDFxStream now supports a subscription-oriented licensing model. It will automatically verify the status of licensing and technical support access, warning if one’s subscription is nearing expiration. For more information, and to learn how to disable this verification, please visit https://www.snowtide.com/help/telemetry
Hundreds of relatively miscellaneous PDF specification support improvements, performance optimizations, and bugfixes.

Changes in PDFxStream v3.9.8

Significantly improved PDFxStream’s document “repair” tactics, to (as needed) deeply examine dangling objects to infer missing document structure.
Improved handling of documents that refer to undefined fonts.
Optimized retrieval of document metadata to avoid unnecessary object resolution operations.
Fixed a bug where horizontal textual rule elision could sometimes be triggered on single dash or underscore characters.
Fixed a bug in the handling of certain cross-reference streams.

Changes in PDFxStream v3.9.6

Added support for /Alternate color space arrays
PDFxStream now ships with a version of ‘Century Expanded’ AFM font metrics
Improvements have been made to the process of repairing damaged or malformed PDF documents, specifically in the detection of incorrect xref tables
Fixed the calculation of ascent font metrics for a subset of common embedded AFM fonts

Changes in PDFxStream v3.9.5

Added com.snowtide.pdf.layout.TextUnit.Predicate and corresponding options in Configuration to allow developers to selectively filter characters (TextUnits) from extracted pages.
Significantly improved performance on Windows platforms using new nonblocking I/O in key areas.
Incrementally improved colorspace and gradient detection support.

Changes in PDFxStream v3.9.3

Significantly improved performance in decoding jbig2 and JPEG2000 image data
Adjust constraints associated with determining whether a character is visually underlined, based on updated machine learning model
Two security fixes associated with avoiding infinite loops in conjunction with improperly-structured PDFs.

Changes in PDFxStream v3.9.1

Fix NullPointerException thrown when loading a PDF that contains multiple versions of outline (bookmarks) objects
Fix handling of improperly-guarded inline images
Eliminate layout and performance problems associated with documents that embed images as Type3 font glyphs

Changes in PDFxStream v3.9.0

Added TableUtils.tableToStrings(), allowing easy access to the text-only content of Table instances without any document model programming
SECURITY FIX: Modified embedded image handling to avoid DOS possible via specially-constructed input PDF document (CVE pending). Issue affects allN extraction functions (text, images, or form data), so upgrade is strongly recommended for all customers.

Changes in PDFxStream v3.8.1

Significant improvement (minimization) in temporary allocations while parsing PDF dictionaries
Tuned preallocation behaviour to play nicer with JVM’s Garbage-First (G1) collector

Changes in PDFxStream v3.8.0

Significant improvement in image extraction performance on Java 10 and older
Added support for 256-bit AES decryption as defined in PDF specification ISO 32000-2:2020

Changes in PDFxStream v3.7.5

Comprehensive revision to handling of damaged/incomplete PDF documents, addressing dozens of classes thereof.

Changes in PDFxStream v3.7.2

Off-page rectangles are now ignored; previous handling caused VisualOutputTarget to incorrectly gauge the width of such pages, leading to wildly inappropriate spacing.
Characters that are only very partially occluded by filled rectangles are now excluded from text extracts.
Fixed a divergence in PDFxStream’s behaviour on Java versions 8+ compared to 7 and below, caused by a JDK change in how iteration order over certain collections was implemented.

Changes in PDFxStream v3.7.1

Improved text occlusion detection to properly account for filled rectangles affected by clipping paths.
SECURITY FIX: Modified page tree handling to avoid DOS possible via specially-constructed input PDF document (CVE pending).

Changes in PDFxStream v3.7.0

Now fully loading embedded TrueType font files to obtain character metrics
Change handling of
Include AFM files for Minion, Myriad, and Open Sans typefaces
Added support for out-of-spec documents produced by “Nuance PDF Create” that were previously reported as holding no content
Stop including zero-width spaces as characters in document model
Fix VisualOutputTarget to avoid an OutOfMemoryError when the source page contains oddly-skewed characters
Fix application of object decryption to avoid decrypting certain embedded cmaps twice

Changes in PDFxStream v3.6.0

Significantly improved layout and spacing calculations for Chinese, Japanese, and Korean
Significant performance improvements
Added option to control whether or not horizontal textual rules should be elided (see com.snowtide.pdf.Configuration.setElideHorizontalTextualRules(boolean))
Fixed default metrics for “Cambria” font
Adjust handling of overlapping characters so that reported spatial coordinates are faithful to their encoding in the source PDFs
Fixed handling of alternate descriptor for ICC colorspaces
Miscellaneous minor bugfixes

Changes in PDFxStream v3.5.0

PDFxStream now supports extracting JPX/JPEG2000 image data from PDF documents
Significantly improved page layout performance, up to 30% in many cases
Non-standard date/time strings lacking a timezone indicator are now parsed and interpreted within UTC
Fixed a bug where some embedded italic fonts were not recognized as such
Fixed a bug where characters that are only partially within the bounds of a page were previously excluded from extraction.
Fixed a bug where non-NUL control characters were previously disallowed from being mapped as source character codes in custom font encodings.

Changes in PDFxStream v3.4.0

PDFxStream.NET now requires .NET v4.0 or above
com.snowtide.pdf.RegionOutputTarget now provides a “minimum overlap” configuration option via .setMinimumOverlapPct(float), used to determine how much of each considered character must lie within defined regions in order to be included in the extracted content
A new configuration option has been added, pdfxs.layout.ignoreNonCardinalRotatedChars. When set via system property, environment variable, or com.snowtide.pdf.Configuration.setIgnoreNonCardinalRotatedChars(boolean), PDFxStream will ignore any characters that are rotated by non-cardinal angles, i.e. any angle other than 0, 90, 180, or 270.
Document metadata values that were (improperly) UTF-16 encoded are now repaired before being returned
The calculation of spacing between words has been improved.
Characters that overlap page bounds by any amount are now included in text extracts. (Previously, characters would need to be wholly within page bounds to be included.)
This release includes dozens of bug fixes and enhancements to PDFxStream’s support for PDF variants found in the field.

Changes in PDFxStream v3.3.7

Extracted images now each carry an identifier that (com.snowtide.pdf.layout.Image.id()) is unique to the underlying bitmap data. This makes it easy to never load/decode the same bitmap more than once, even if it appears in multiple places in a source PDF.
com.snowtide.pdf.forms.AcroRadioButtonGroupFields now properly report their pageNumber() and bounds() (the latter being calculated to be the MBR of the group’s component radio buttons).
Heuristics related to intra-word whitespace have been improved.
Fixed a bug where only the first column partition position added via com.snowtide.pdf.Page.addColumnPartition(int) would be recognized.
Fixed a bug where an inset table would potentially prevent the recognition of a column boundary.

Changes in PDFxStream v3.3.6

References are now recursively resolved when found as document metadata values.
The heuristics used to calculate the number of spaces between characters have been updated to be more accurate in properly recognizing and dealing with text that is justified, but tightly kerned.
Fixed a bug where AFM font metrics were being improperly applied to Type0 fonts, leading in certain cases to serious overruns of TextUnit bounds beyond where they actually should have been.
The single-threaded usage limitation has been removed when PDFxStream is used without a license file. Instead, PDFxStream will open a maximum of 500 PDF documents; this count is reset when you restart your program or application. This is intended to be a reasonable development and test limitation for most early evaluation purposes. Email us help@snowtide.com to obtain a license file that will remove this restriction for the duration of your development and testing.

Changes in PDFxStream v3.3.5

This release contains a significant fix to how text encodings and embedded character maps are unioned to produce efficient decoding of multibyte text encodings.

Further small fixes include:

The widths of named characters that are unmapped in AFM files are now properly applied when referenced in embedded text encoding specifications.
Fixed a regression where inline images were not being properly skipped.

Changes in PDFxStream v3.3.1

PDFxStream’s command-line support, provided by the com.snowtide.pdf.Console class, now includes an option to use VisualOutputTarget when extracting text content from a source PDF document.
VisualOutputTarget’s handling of pages that contain text rendered at different sizes has been significantly improved.
Fixed a bug where embedded character maps that don’t start with /CIDInit are now recognized.
Font descent metrics are now taken into account when calculating linebreak counts between lines.
A mismatch between the FontBBox and ascender/descender metrics was fixed in PDFxStream’s bundled Courier font descriptions.

Changes in PDFxStream v3.3.0

PDFxStream.NET now ships with two different PDFxStream assemblies: one for use with VB.NET, one for use with all other .NET languages. This addresses a problem where PDFxStream could not be used with recent VB.NET compilers.
Significantly improved the handling of overlapping rectangles when determining visibility of content.
Fixed a performance regression introduced in v3.2.0.

Changes in PDFxStream v3.2.1

The output of pdfts.examples.GoogleHTMLOutputHandler now includes appropriate META tags in order to ensure proper encoding and display of high-code-point Unicode characters.
Blocks in the PDFxStream document model are now split more aggressively in order to better correspond with obvious paragraph breaks.
Adobe Font Metrics (AFM) are now located properly even when font or font family names are hex-encoded (e.g. #20 where spaces should be)
Fixed a serious regression where vertically-aligned com.snowtide.pdf.layout.TextUnits carrying the same character (sequence) and rendered using monospace fonts would be omitted from text extracts entirely.
Fixed a bug where empty-string embedded encodings are now properly ignored.
Fixed a bug where (faulty) encoding information from an embedded font was being applied in favor of (accurate) ToUnicode character mappings.

Changes in PDFxStream v3.2.0

This release contains a number of new features and capabilities, as well as a large number of fixes of customer-reported bugs.

PDFxStream now supports accessing the “appearance” associated with extracted PDF annotations. This is manifested by a separate com.snowtide.pdf.Page for each annotation that defines its own appearance, accessible via com.snowtide.pdf.annot.Annotation.getAppearance().
com.snowtide.pdf.VisualOutputTarget now includes space at the beginning of each line as necessary corresponding to the left margin of the page. This makes it easier to concatenate the text extracted from multiple pages, and process the result cumulatively, which is useful for e.g. tabular data spanning multiple pages where columns are delineated by whitespace. The old behaviour can be recovered via com.snowtide.pdf.VisualOutputTarget.setMarginTrimmed(true).
Added com.snowtide.pdf.TextUnit.getFontSize()
com.snowtide.pdf.Console now offers an --attrs option, which will emit all of the document-level metadata attributes present in the provided input file.
Improved lazy loading of PDF image data (i.e. so that only data associated with the images that an application actually accesses will be loaded)
Added support for softmask images
Added support for form checkboxes as annotation widgets
Fixed bug where PDFxStream was attempting to decrypt certain encrypted PDF strings twice
Fixed bug where image data stored using interleaved encodings that include CCITT were not decoded properly
PDFxStream’s packaging process has now been changed so that intermediate package names will never collide with the name of a class.

Changes in PDFxStream v3.1.3

A wider range of Unicode space characters are now excluded from the PDFxStream document model, including “regular” non-breaking spaces, as well as zero-width non-breaking spaces.
Inference of sub- and super-scripted characters now yields more correct positioning of them within lines.
Minor changes have been made to the statistical models that determine the whitespace distance threshold for each line of extracted text.
Addressed intra-character whitespace calculation in cases where PDFs use embedded fonts that fail to specify the width of space characters or a default character width
Fixed a bug where certain single-byte character encodings were incorrectly treated as being multibyte.
Fixed a bug where form extraction would fail for “choice” fields that contained no selectable options.

Changes in PDFxStream v3.1.2

Adopted the Adobe Glyph List mappings for Unicode Private Use Area characters
Fixed an issue where PDFxStream would not load on certain v1.5.x JDKs

Changes in PDFxStream v3.1.1

Fixed a problem with the packaging of PDFxStream.NET bundles.

Changes in PDFxStream v3.1.0

This is the first public release of PDFxStream v3.x. It introduces a number of new capabilities and adds tons of smaller improvements over PDFTextStream v2.7.0, which preceeded it. Upgrading to PDFxStream should be relatively painless: steps have been taken to maximize the API compatibility between the two releases, though there are some minor breaking changes (mostly related to rebranding to “PDFxStream” as the main product name).

Since moving from v2.7.0 to v3.1.0 constitutes a major upgrade, existing PDFTextStream customers will need new PDFxStream v3.x license file(s). Please contact us to request issuance of your new PDFxStream license file(s).

New capabilities

PDF image extraction is now offered as a distinct feature (com.snowtide.PDF.Feature.Images). The key method is com.snowtide.pdf.Page.getImages(), which returns a collection of com.snowtide.pdf.layout.Image objects that can provide either encoded image data (as PNG, JPEG, etc) or “live” platform-suitable image objects (either java.awt.image.BufferedImage on the JVM, or System.Drawing.Bitmap on .NET).
Extensive set of enhancements to PDFxStream’s support for ideosyncratic PDFs, as produced en masse by large vendors (including Oracle, IBM, SAS, and Salesforce products).
Significant performance improvements, especially related to the identification of tabular and columnated regions of content.
Added com.snowtide.pdf.Page.getCharacters(), yielding a collection of com.snowtide.pdf.layout.TextUnits on a page without incurring the costs associated with page segmentation and read-ordering required by calling Page.getTextContent() or Page.pipe(OutputHandler).
Added support for extracting PDF attachments, both at the document level (com.snowtide.pdf.Document.getEmbeddedFiles()) and those associated with particular annotations (com.snowtide.pdf.annot.FileAttachmentAnnotation).
Added support for identifying the location of interactive form fields via addition of com.snowtide.pdf.DocumentLocation as superinterface of com.snowtide.pdf.forms.AcroFormField.
Wherever possible, the PDFxStream API has been generified to maximize the availability of static type information.

Breaking changes

The Lucene integration API provided by classes in the com.snowtide.pdf.lucene package are no longer included in PDFxStream. They will be open-sourced separately in the near future according to customer demand.
com.snowtide.pdf.PDFTextStreamConfig has been renamed Configuration.
The memory-mapping option previously offered by PDFTextStreamConfig has been removed: memory mapping is now never used by PDFxStream. This addresses various problems with memory-mapping PDF files on Windows, and eliminates an option that no longer had any benefit due to improvements made in how PDFxStream utilizes I/O. Concretely, this change eliminates the following methods and other facilities, with no replacement in the PDFxStream API:
- PDFTextStreamConfig.setMemoryMappingEnabled(boolean)
- PDFTextStreamConfig.isMemoryMappingEnabled()
- The pdfts.mmap.disable system property / environment variable
com.snowtide.pdf.layout.Rectangle is no longer an empty marker interface; it is now the concrete, default implementation of com.snowtide.pdf.layout.Region. The Rectangle interface was likely never used by any code consuming PDFTextStream, so breakage associated with this change should be minimal to nonexistent.
com.snowtide.pdf.PDFTextStream no longer extends java.io.Reader, or implements java.lang.Readable. Based on customer feedback, these affordances were never used.
Many classes in com.snowtide.pdf.layout that used to implement the Region interface now implement Bounded instead. At worst, this change will require that customer code that used to access the spatial properties of a Region now must obtain a Region from a Bounded object first. e.g., this code:
```
  Block block = ...;
  float xposition = block.xpos();
```
must be changed to:
```
  Block block = ...;
  float xposition = block.bounds().xpos();
```
Various methods on com.snowtide.pdf.EncryptionInfo were originally made public in error, they are of no use outside of PDFxStream’s internals. These methods are now private.
com.snowtide.pdf.Page.getStream() has been renamed to getDocument().
The public static main (String[]) method previously provided by com.snowtide.pdf.PDFTextStream has been consolidated into the catch-all main method provided by com.snowtide.pdf.Console.
com.snowtide.pdf.forms.Form, its sub-interfaces, and its implementations are now generified based on the type of form fields they contain.
com.snowtide.pdf.EncryptionInfo.getErrorType() now returns an instance of EncryptionInfo.ErrorType, a new enumeration.
com.snowtide.pdf.FaultyPDFException’s constructors are now private.
com.snowtide.pdf.PDFVersion is now defined as an enumeration, instances of which are returned by com.snowtide.pdf.Document.getPDFVersion()
com.snowtide.pdf.Page.getPdfName() has been removed; use Page.getDocument().getName() instead.
com.snowtide.pdf.annot.Annotation now implements com.snowtide.pdf.DocumentLocation; its old methods getRect() and getPageNumber() have been renamed to match the analogous methods defined by DocumentLocation.

Miscellaneous additions / fixes / changes

com.snowtide.pdf.PDFTextStream is now deprecated, though its API remains to prevent immediate breakage of code referencing it. Please update your projects to open PDF files via com.snowtide.PDF.open(), and use the com.snowtide.pdf.Document interface as the type representing those files.

Changes in PDFTextStream v2.7.0

PDFTextStream.NET now uses and ships with IKVM 0.46.0.4
ASCII “control characters” (0-8) are no longer added to page document models
Add support for /CXXX character names (observed in the wild, naming unicode code points in hex), which yields proper decoding of certain documents
Added support for embedded CMAP files that use CR characters for linebreaks
Added support for embedded font files that assume Windows-1250 text encoding
Added workaround for certain PDF documents with malformed embedded CMAP files that would cause an infinite loop / hang in PDFTextStream’s processing of said files.
Fixed bug in PDF file merge facility that resulted in certain object stream-encoded objects from being included in the result of the merge.
Fixed bug where certain signature form fields were not decoded properly, which was causing AcroSignatureField.getValue() to return null (instead of a Map of the field’s properties).
Fixed bug where compressed PDF object references were unnecessarily loaded repeatedly under .NET 4.5 on Windows
Fixed bug where some PDF document merge operations would yield an incorrect reference to /Metadata objects, and thus prevent Acrobat Reader from printing them

Changes in PDFTextStream v2.6.4

Added table detection enablement option to com.snowtide.pdf.PDFTextStreamConfig (PDFTextStreamConfig.isTableDetectionEnabled() and PDFTextStreamConfig.setTableDetectionEnabled(boolean))
Eliminated potential NullPointerException when cropping a page prior to its layout being initially calculated (com.snowtide.pdf.Page.crop(Rect))
Fixed bug where incorrect character spacing data was applied to Adobe-standard fonts embedded in PDFs (resulting in poor/nonexistent word spacing)

Changes in PDFTextStream v2.6.3

Significantly improved ‘repair’ procedure for damaged or malformed PDF documents
Fixed bug in PDF merge functionality that would occasionally manifest as a blank page

Changes in PDFTextStream v2.6.2

Fixed rendering issue associated with usage of TrueType fonts with multibyte-encoded text streams.
Added compatibility fix for PDF documents that contain spurious (out of range) non-printing bytes.

Changes in PDFTextStream v2.6.1

Enhanced “repair” procedure for PDF documents with one-off stream encoding errors.
Fixed handling of text encoding found in PDF documents generated within Mac OS X Lion (whitespace in cmap codepoints)

Changes in PDFTextStream v2.6.0

New OutputHandler: com.snowtide.pdf.SelectionOutputTarget, implementing text extraction based on a “selection coordinates”, as commonly found in user-facing PDF viewer UIs.
PDFTextStream is now free for use in single-threaded applications; all previous “evaluation” limitations no longer apply when PDFTextStream is operated without a license file.

Changes in PDFTextStream v2.5.0

Added support for decryption of AES-encrypted PDF documents (includes support for 256-bit and variable bit length ciphers)
PDFTextStream for Java now requires v1.5.0 or higher of the JVM/JRE
PDFTextStream.NET is now tested and supported under Mono
PDFTextStream.NET now uses and ships with IKVM 0.46.0.1, and requires .NET 2.0 or higher.
com.snowtide.pdf.PDFTextStream now implements java.io.Closeable
com.snowtide.pdf.OutputTarget and its subclasses now accept java.lang.Appendables instead of strictly java.lang.StringBuffers
com.snowtide.pdf.PDFTextStream now offers String-based (file path) constructors
Dozens of performance and PDF document compatibility enhancements
added LinkAnnotation.getTargetPageNumber(); LinkAnnotation no longer improperly shadows Annotation.getPageNumber()
Fixes a fatal character decoding bug on IBM J9 JVMs
Fixes support for Windows ANSI-encoded PDF text
Fixes support for tracking the position and rotation of PDF media boxes (no longer just height/width)
The “NOCJK” build of PDFTextStream (all the same functionality, but without the font encoding files needed to extract CJK character sets) is no longer offered
PDF merge capability (com.snowtide.pdf.util.MergeUtil) has been deprecated
Memory-mapping of opened PDF files is now disabled by default, and has been deprecated

Changes in PDFTextStream v2.3.2

Fixed issue where PDFTextStream would fail to initialize when the default system locale was set to Shift_JIS (i.e. SJIS, MS932, Windows-31J)
Fixed an issue where certain Chinese, Japanese, and Korean fonts were not being loaded properly when specific encoding config data was missing.
Fixed an octal string parsing bug that could lead to a PDF parsing failure.
Added crop box attribute to com.snowtide.pdf.Page interface
An expanded set of control characters are now treated as whitespace.
Added support for non-compliant PDF documents produced by TXT2PDF for OS/390.

Changes in PDFTextStream v2.3.1

Added methods to VisualOutputTarget to enable the optional exclusion of rotated content from its output (523)
Fixed a bug where rotated characters were reporting a rotation angle (theta) of 0 when presented to VisualOutputTarget. (519)
Fixed a bug where use of PDFTextStream.NET in a multithreaded environment could produce garbled or missing text extracts in very limited circumstances.
Added support for PDFs that contain malformed arrays in their graphics output streams (509)
Fixed a bug where text rendered using a Type3 font that has a proper unicode mapping was being omitted from extracts (507)
Significantly improved the emission of whitespace between words on lines with large amounts of tracking (506)
Fixed character mapping for ‘ã’ and ‘-’ (“middle dot”) (502)
Fixed a bug affecting VisualOutputTarget and RegionOutputTarget where smaller characters would not be included in resulting text extracts. (499)
Fixed an issue where string values held in compressed object streams were being re-encrypted (primarily affecting key/value PDF attributes) (495)
Fixed an issue where PDF documents generated by PDFSharp were improperly handled, leading to significant degradation of extraction accuracy. (490)
Fixed an issue where CFF font encodings were being applied inappropriately, potentially leading to garbled extracts. (479)
fixed a bug related to zero-length cross-reference entry codes that was resulting in a improper FaultyPDFException being thrown (450)

Changes in PDFTextStream v2.3.0

Added an .isStruckThrough() method to com.snowtide.pdf.TextUnit, indicating whether a character has a strikethrough drawn through it.
Improved PDFTextStream’s support for embedded character mappings.
The calculation of whitespace between words has been fixed to properly account for whitespace that is explicitly encoded in the source PDF documents.
Improved PDFTextStream’s handling of composite content encodings, which previously could fail resulting in some ranges of PDF content being ‘ignored’ during extraction.
Fixed a bug in VisualOutputTarget where text from a single line would be split over multiple lines
Improved vertical alignment of text extracted using VisualOutputTarget
Improved VisualOutputTarget-produced extracts to eliminate spurious additional whitespace between closely-adjacent words

Changes in PDFTextStream v2.2.5

Added support for extracting XFA forms data as XML
Significantly improved the performance of text extraction using VisualOutputTarget
Added support for PDF documents larger than 2GB
Fixed a bug where the encodings from embedded Type1 fonts were previously not being applied properly in some circumstances.
Fixed a problem where newer content in updated PDF documents were sometimes being ignored.
Fixed a problem where PDFDocEncoding-encoded bookmarks and metadata were not being decoded properly
added .getDestinationName() method to com.snowtide.pdf.Bookmark

Changes in PDFTextStream v2.2.1

PDFTextStream.NET now ships with ikvm v0.3.4, which fixes a number of problems that prevented PDFTextStream from functioning properly across multiple AppDomains (598)
Added PDFTextStream.loadLicense(URL) function (475)
Added a ‘spacing scale’ property to VisualOutputTarget which allows applications to control the amount of horizontal whitespace that should be emitted per physical amount of whitespace found in the source document (528)
PDFTextStream will now attempt to load a license file from the host application’s current directory before checking the current classpath / AppDomain (661)
Fixed a problem where pathological embedded Unicode character encodings were causing PDFTextStream to strings of control characters rather than reasonable extracted content. (428)
Fixed a bug in PDFTextStream’s handling of cross reference entries that caused fatal errors in some documents (620)
Fixed a problem where UTF-16 encoded bookmark titles were not being decoded properly (618)

Changes in PDFTextStream v2.2

Added support for Apache Lucene v2.1 and v2.2 to PDFTextStream’s Lucene integration module (com.snowtide.pdf.lucene.PDFDocumentFactory)
Added com.snowtide.pdf.PDFTextStreamConfig, which enables simple static and runtime configuration of PDFTextStream
Added new PDFTextStream constructors that accept customized PDFTextStreamConfig instances, and a setConfiguration(PDFTextStreamConfig) function to set a PDFTextStream instance’s configuration at runtime
PDFTextStream now joins adjacent rectangles that have similar stroke and fill colors, which improves various page segmentation results
Improved table detection processes to adaptively recognize very small “variant” table cells
Improved pdfts.examples.XMLOutputTarget to build an XML DOM Document instead of constructing XML using a StringBuffer; block elements now include a type attribute of “table” if the block is a table
Significantly improved the quality of PDF documents generated when merging PDF files (com.snowtide.pdf.util.MergeUtil) and when saving updated PDF forms (com.snowtide.pdf.forms.AcroForm#writeUpdatedDocument(OutputStream))
Rotated text blocks are now properly grouped within bounded regions
Changed pdfts.cjk.disable and pdfts.mmap.disable system properties to pdfts.cjk.enable and pdfts.mmap.enable, respectively
Fixed an overflow bug in PDFTextStream’s PDF data parser
Fixed a bug where the ascent and descent characteristics of some fonts were defaulting to improper values
Fixed a bug where lines and rectangles drawn with a Separation color space were not being recognized properly
Fixed a bug where an error would result when reading a PDF file with a non-conforming linebreak sequence after the `stream’ tag
Fixed a bug where tables containing underlined text would not be recognized properly
Fixed a bug where edges of rectangles were improperly recognized as text underlines
Fixed a bug where PDFTextStream wouldn’t recognize PDF data stream filter name abbreviations

Changes in PDFTextStream v2.1.6

Added com.snowtide.pdf.util.TableUtils, which provides a set of CSV conversion functions for exporting the contents of tables
Added options to specify path to load PDFTextStream license file via pdfts_license_path environment variable or system property
Added com.snowtide.pdf.PDFTextStream.loadLicense(String) - programmatic way to specify path from which to load PDFTextStream license file
Changed PDFTextStream’s default page segmentation algorithms to not eliminate empty table cells, making it simpler to export tabular content to Excel, etc.
Fixed bug in VisualOutputTarget where vertically-adjacent lines of text were being inappropriately combined
Fixed text encoding bug where text extracted from PDF documents generated by Adobe InDesign v4.0 - v5.0 would be “scrambled”, or appear to be series of Chinese glyphs
Fixed bug where AFM font mappings were sometimes applied in an incorrect order, leading to spot errors in text extracts
Fixed bug where certain embedded Type1 font encodings were not being loaded correctly, resulting in single-character extraction errors

Changes in PDFTextStream v2.1.5

Significant improvements in the handling and standard output of rotated content
Added com.snowtide.pdf.layout.TextUnit.getTheta()

Changes in PDFTextStream v2.1.3

Added com.snowtide.pdf.Font.isItalic() – indicates whether a font is italicized
Added com.snowtide.pdf.layout.TextUnit.isUnderlined() – indicates whether a character is underlined
Added tagging of italic text regions in pdfts.examples.XMLOutputTarget

Changes in PDFTextStream v2.1.2

Fixed page rotation detection bug when processing PDF documents generated by Crystal Reports

Changes in PDFTextStream v2.1.1

Significant improvements in output of VisualOutputTarget, especially for pages with many different font sizes
Fixed calculation of character widths for Type0 font that have a recognized AFM base font name

Changes in PDFTextStream v2.1

Added support for updating text, checkbox, radio button, and choice interactive form fields
Added support for Kodak print job data extraction (%KDK commands) via com.snowtide.pdf.util.KodakPrintData
Exposed the AcroFormField.isReadOnly() function
Added ByteBuffer-based buildPDFDocument() functions to com.snowtide.pdf.lucene.PDFDocumentFactory
Added the pdfts.logfactory' andpdfts.loggingtype’ system variables to simplify the customization of logging via com.snowtide.util.logging.LoggingRegistry
java.util.logging is now the default logging toolkit; `pdfts.loggingtype’ may be used to change that. Refer to the LoggingRegistry javadocs for more info.
Improved documentation significantly
Fixed a problem where merged PDF documents that contained empty dictionaries would be improperly generated
Fixed a problem where the “rich text” value of text interactive form fields would not be loaded

Changes in PDFTextStream v2.0.5

Fixed handling of text spacing that was causing some columnated text to overrun column boundaries improperly
Fixed a problem where text from adjacent lines would be inappropriately intermingled
Changed unlicensed functionality so that evaluation use would not require a special evaluation license file; specifically, PDFTextStream will randomize some digits in text extracts when it is operating unlicensed, and the 8-page extract limitation has been removed

Changes in PDFTextStream v2.0.2

Added com.snowtide.pdf.RegionOutputTarget to support region-specific content extraction
Added ability to derive encoding and spatial metrics of Type3 fonts; added pdfts.type3.derive system property to disable derivation if necessary (359)
Fixed problem with com.snowtide.pdf.VisualOutputTarget, where lines would sometimes be inappropriately combined (356)

Changes in PDFTextStream v2.0.1

Better indication of corrupted or otherwise unreadable PDF files (com.snowtide.pdf.FaultyPDFException)
Added pipe(OutputHandler) function to com.snowtide.pdf.layout.Line
Added pdfts.mmap.disable system property option to disable memory-mapping of PDF files - avoids JDK bug #4724038 (355)

Changes in PDFTextStream v2.0

PDFTextStream now available for .NET and Python
Added support for extraction of Chinese, Japanese, and Korean text (CJK)
Added support for accessing derived table structure (com.snowtide.pdf.layout.Table)
Significantly improved performance
Significantly improved accuracy of extraction of rotated text
Added support for Lucene v1.9 and v2.0
Added “visual” text layout output target (com.snowtide.pdf.VisualOutputTarget)
Added PDF merge capability (com.snowtide.pdf.util.PDFMergeUtil)
Added support for Type1C embedded font files (274)
Fixed issue where some bookmarks would have invalid page number attributes (i.e. -1) (324)
Fixed issues where blocks, lines, and textunits that represented rotated text reported inaccurate positions on the page
Fixed issue where xref table was not being rebuilt when object locator was simply missing (338)
Eliminated com.snowtide.pdf.PDFTextStreamOptions (deprecated in v1.3)

Changes in PDFTextStream v1.4

Added support for interactive PDF forms (AcroForms) (com.snowtide.pdf.forms.* and com.snowtide.pdf.PDFTextStream.getFormData()) (118)
Added support for derivation of ‘graphical’ font encoding (Type3) (297)
Added com.snowtide.pdf.OutputHandler base class for OutputTarget
Added PDFTextStream constructor that takes a java.nio.ByteBuffer, enabling completely in-memory operation
Added an example class that extracts form data as XML (pdfts.examples.XMLFormExport)
Added sample implementation of com.snowtide.pdf.OutputHandler that outputs PDF text as XML, indicating document structure and where bolded text ranges exist (pdfts.examples.XMLOutputTarget)
Added sample OutputHandler implementation that exports PDF text content as an XHTML document (pdfts.examples.GoogleHTMLOutputHandler)
Fixed bug where inline images were not being properly skipped (308)
Fixed bug where destination bounds of some bookmarks and annotations were not being properly set (307)
Fixed bug where text properties (font size, character encoding, etc) would persist beyond where they should (298)

Changes in PDFTextStream v1.3.6

Fixed potential OutOfMemoryError caused by complex graphical regions (295)
Fixed bug where out-of-date content might be extracted from updated PDF documents (296)

Changes in PDFTextStream v1.3.5

Added PDF annotation API (com.snowtide.pdf.annot.*) (76)
Added PDF bookmark API (com.snowtide.pdf.Bookmark and com.snowtide.pdf.PDFTextStream.getBookmarks()) (284)
Significantly improved performance parsing PDF data containing very complex illustrations (282)
Improved triage procedures for handling damaged or malformed PDF files (292)
Fixed bug where com.snowtide.pdf.Page.getPageNumber() was reporting 1-indexed page numbers; it now properly reports 0-indexed page numbers (283)
Fixed parsing bug related to zero-length PDF names (290)

Changes in PDFTextStream v1.3.4

Improved rectangle and line detection to avoid skipping graphics that impact text layout (272)
Improved the algorithm used to calculate the number of line breaks to be outputted between lines of text (271)
Improved detection and handling of malformed PDF documents to prevent potential infinite loops (278)
Fixed compatibility problem with PDFs generated by IBM Manyimage tool
Fixed compatibility problem with PDFs generated by SAP R/3 (276)
Fixed error thrown when some blank pages are encountered (270)

Changes in PDFTextStream v1.3.3

Expanded support for referenced form XObjects; results in more complete text extracts (263)
Improved font lookup routines; now caching frequently-referenced fonts for improved performance
Fixed logging classloading issue on JDK 1.3.1_01

Changes in PDFTextStream v1.3.2

Significant performance enhancement through improved usage of java.nio.* classes; available only on JDK 1.4+

Changes in PDFTextStream v1.3.1

Fixed integration with JDK v1.4 java.util.logging toolkit

Changes in PDFTextStream v1.3

Added ability to retrieve PDF document page attributes (height, width, rotation, etc) (94)
Added ability to retrieve PDF document pages one at a time (94)
Added ability to retrieve PDF document encryption parameters (99)
Added ability to retrieve PDF file specification version number (91)
Added pipe() method to PDFTextStream and retrieved PDF pages, allowing easy redirection of content to a buffer to file (92)
Significantly improved page segmentation and document read-ordering, resulting in more semantically-consistent text extracts
Significantly improved extraction of rotated text
Significantly improved extraction of line-bounded tables (107)
Deprecated PDFTextStreamOptions class: strictEncoding and page header options no longer used (87, 98)
PDFTextStream now always produces Unicode text; the ASCII-only option is no longer provided, as it proved to be unreliable (87)
Fixed some minor Unicode text extraction issues related to selecting the proper character encoding for Type 1 fonts (86)
Fixed PDFTextStream’s implementation of the PDF graphics state stack to more closely conform to the PDF spec (90)
Fixed problem where certain monospaced character might be omitted from output
Fixed problem where text might be scrambled on a line that contains certain monospaced text (182)

Changes in PDFTextStream v1.2

Added support for retrieving document-level Adobe XMP data (document metadata in an XML format) (66)
Added support for PDF v1.5 files encrypted using crypt filters that specify an invalid decryption key length (63)
Improved overview documentation of metadata access in Javadoc and Developer’s Guide (70)
Fixed support for decrypting updated PDF v1.4 files encrypted with 128-bit passwords (62)
Fixed internal error that might have occurred in connection with processing updated PDF documents (72)

Changes in PDFTextStream v1.1.2

Enhanced the core parsing routines to accept PDF files that use improper (or nonexistant) string escape sequences
Fixed a bug that caused hard errors when processing some PDF v1.5 documents.
Fixed a bug where a particular text mapping (hex / CIDFont mappings) used in some PDF’s would be misinterpreted, resulting in space characters being outputted instead of ‘regular’ characters

Changes in PDFTextStream v1.1.1

Fixed a problem where some PDF’s that use a particular type of TrueType font were converted into useless text content

Changes in PDFTextStream v1.1

JDK v1.3 is now fully supported.
Significant improvements have been made in the layout and formatting of rotated text.
All logging is now channeled through Jakarta’s commons-logging library to enable usage of logging toolkits other than log4j.