qpdf-10.0.0

[Douglas R. Reno]

New major version

[Bruce Dubbs]

10.0.0: April 6, 2020

Performance Enhancements

The qpdf library and executable should run much faster in this version than in the last several releases. Several internal library optimizations have been made, and there has been improved behavior on page splitting as well. This version of qpdf should outperform any of the 8.x or 9.x versions.

Incompatible API (source-level) Changes (minor)

The QUtil::srandom method was removed. It didn't do anything unless insecure random numbers were compiled in, and they have been off by default for a long time. If you were calling it, just remove the call since it wasn't doing anything anyway.

Build/Packaging Changes

Add a openssl crypto provider, which is implemented with OpenSSL and also works with BoringSSL. Thanks to Dean Scarff for this contribution. If you maintain qpdf for a distribution, pay special attention to make sure that you are including support for the crypto providers you want. Package maintainers will have to weigh the advantages of allowing users to pick a crypto provider at runtime against the disadvantages of adding more dependencies to qpdf.

Allow qpdf to built on stripped down systems whose C/C++ libraries lack the wchar_t type. Search for wchar_t in qpdf's README.md for details. This should be very rare, but it is known to be helpful in some embedded environments.

CLI Enhancements

Add objectinfo key to the JSON output. This will be a place to put computed metadata or other information about PDF objects that are not immediately evident in other ways or that seem useful for some other reason. In this version, information is provided about each object indicating whether it is a stream and, if so, what its length and filters are. Without this, it was not possible to tell conclusively from the JSON output alone whether or not an object was a stream. Run qpdf --json-help for details.

Add new option --remove-unreferenced-resources which takes auto, yes, or no as arguments. The new auto mode, which is the default, performs a fast heuristic over a PDF file when splitting pages to determine whether the expensive process of finding and removing unreferenced resources is likely to be of benefit. For most files, this new default will result in a significant performance improvement for splitting pages. See Section 3.8, “Advanced Transformation Options” for a more detailed discussion.

The --preserve-unreferenced-resources is now just a synonym for --remove-unreferenced-resources=no.

If the QPDF_EXECUTABLE environment variable is set when invoking qpdf --bash-completion or qpdf --zsh-completion, the completion command that it outputs will refer to qpdf using the value of that variable rather than what qpdf determines its executable path to be. This can be useful when wrapping qpdf with a script, working with a version in the source tree, using an AppImage, or other situations where there is some indirection.

Library Enhancements

Random number generation is now delegated to the crypto provider. The old behavior is still used by the native crypto provider. It is still possible to provide your own random number generator.

Add a new version of QPDFObjectHandle::StreamDataProvider::provideStreamData that accepts the suppress_warnings and will_retry options and allows a success code to be returned. This makes it possible to implement a StreamDataProvider that calls pipeStreamData on another stream and to pass the response back to the caller, which enables better error handling on those proxied streams.

Update QPDFObjectHandle::pipeStreamData to return an overall success code that goes beyond whether or not filtered data was written successfully. This allows better error handling of cases that were not filtering errors. You have to call this explicitly. Methods in previously existing APIs have the same semantics as before.

The QPDFPageObjectHelper::placeFormXObject method now allows separate control over whether it should be willing to shrink or expand objects to fit them better into the destination rectangle. The previous behavior was that shrinking was allowed but expansion was not. The previous behavior is still the default.

When calling the C API, any non-zero value passed to a boolean parameter is treated as TRUE. Previously only the value 1 was accepted. This makes the C API behave more like most C interfaces and is known to improve compatibility with some Windows environments that dynamically load the DLL and call functions from it.

Add QPDFObjectHandle::unsafeShallowCopy for copying only top-level dictionary keys or array items. This is unsafe because it creates a situation in which changing a lower-level item in one object may also change it in another object, but for cases in which you know you are only inserting or replacing top-level items, it is much faster than QPDFObjectHandle::shallowCopy.

Add QPDFObjectHandle::filterAsContents, which filter's a stream's data as a content stream. This is useful for parsing the contents for form XObjects in the same way as parsing page content streams.

Bug Fixes

When detecting and removing unreferenced resources during page splitting, traverse into form XObjects and handle their resources dictionaries as well.

The same error recovery is applied to streams in other than the primary input file when merging or splitting pages.

[Bruce Dubbs]

Fixed at revision 22968.

[Bruce Dubbs]

Milestone renamed

[Bruce Dubbs]

Milestone renamed