Make later ToUnicode mappings supersede earlier ones CPDF_ToUnicodeMap kept the numerically lowest unicode when a charcode was mapped more than once. Adobe TN #5014 requires the last mapping to win: "Code mappings (unlike codespace ranges) may overlap, but succeeding maps superceded preceding maps." Subset fonts commonly emit a catch-all bfrange and then override individual codes, so this decoded those overrides incorrectly whenever the correct unicode was numerically above the charcode. In practice that meant accented letters and symbols: a font mapping 0xCE to U+00F3 extracted as U+00CE, and one mapping 0x9B to U+00BA extracted as a C1 control character. ASCII in the same font was unaffected, because there the target falls below the charcode, so affected pages looked almost correct. The behaviour was not deliberate. Deduplication was added to bound memory by [1], and [2] collapsed the multimap to a flat map using std::min while speeding up ReverseLookup(). Both were memory and speed changes rather than semantic ones. Apply the same rule to reverse_map_, so both maps agree about which mapping is the effective one. Where several charcodes map to one unicode, ReverseLookup() now answers with the last of them rather than the lowest, which is observable through CharCodeFromUnicode(). Deduplication is preserved in both maps, so that memory bound still holds. This change exposes a second bug, in the font subsetter rather than in the reader. The FPDFSaveWithFontSubsetEmbedderTest expectations that change here are a symptom of that one, not a consequence of correcting the precedence rule. The ToUnicode CMap that LoadCompositeFont() generates for a font loaded through FPDFText_LoadFont() already mapped some charcodes twice. It collects the font's cmap into a multimap keyed by glyph, so a glyph that several unicodes reach contributes several entries, and overlapping bfranges result. A reader following TN #5014 has always resolved those overlaps to the last mapping, so what these generated fonts extract to in other PDF viewers has never been what these tests recorded. The tests could not show that, because they check PDFium's writer using PDFium's reader, and the reader keeping the lowest mapping cancelled out the writer's ambiguity. Correcting the reader stops the two from cancelling, so the expectations now record what a conformant viewer has been seeing all along: U+4E00 rather than U+2F00, which is the better answer, but also U+F906 rather than U+53E5 and U+00A0 rather than U+0020, which are worse. The subsetter is fixed separately by [3], which makes each charcode map once so that no reader has an overlap to resolve. [1] https://pdfium-review.googlesource.com/c/pdfium/+/95390 [2] https://pdfium-review.googlesource.com/c/pdfium/+/139310 [3] https://pdfium-review.googlesource.com/c/pdfium/+/157050 Bug: 554790604 Change-Id: Ib76dae0f8188dadec7b92ef7b31d86fa11ed201c Reviewed-on: https://pdfium-review.googlesource.com/c/pdfium/+/156290 Reviewed-by: Tom Sepez <tsepez@chromium.org> Commit-Queue: Tom Sepez <tsepez@chromium.org> Reviewed-by: Lei Zhang <thestig@chromium.org>
PDFium uses the same build tooling as Chromium. See the platform-specific Chromium build instructions to get started, but replace Chromium's “Get the code” instructions with PDFium's.
The default architecture for Windows, Linux, and Mac is “x64”. On Windows, “x86” is also supported. GN parameter “target_cpu = "x86"” can be used to override the default value. If you specify Android build, the default CPU architecture will be “arm”.
It is expected that there are still some places lurking in the code which will not function properly on big-endian architectures. Bugs and/or patches are welcome, however providing this support is not a priority at this time.
PDFium aims to be compliant with the Chromium policy.
Currently this means Clang. Former MSVC users should consider using clang-cl if needed. Community-contributed patches for gcc will be allowed. No MSVC patches will be taken.
Run: download_from_google_storage --config and follow the authentication instructions. Note that you must authenticate with your @google.com credentials. Enter “0” if asked for a project-id.
Once you've done this, the toolchain will be installed automatically for you in the Generate the build files step below.
The toolchain will be in depot_tools\win_toolchain\vs_files\<hash>, and windbg can be found in depot_tools\win_toolchain\vs_files\<hash>\win_sdk\Debuggers.
If you want the IDE for debugging and editing, you will need to install it separately, but this is optional and not needed for building PDFium.
The name of the top-level directory does not matter. In the following example, the directory name is “repo”. This directory must not have been used before by gclient config as each directory can only house a single gclient configuration.
mkdir repo cd repo gclient config --unmanaged https://pdfium.googlesource.com/pdfium.git gclient sync cd pdfium
On Linux, additional build dependencies need to be installed by running the following from the pdfium directory.
./build/install-build-deps.sh
PDFium uses GN to generate the build files and Ninja to execute the build files. Both of these are included with the depot_tools checkout.
PDFium may be built either with or without JavaScript support, and with or without XFA forms support. Both of these features are enabled by default. Also note that the XFA feature requires JavaScript.
Configuration is done by executing gn args <directory> to configure the build. This will launch an editor in which you can set the following arguments. By convention, <directory> should be named out/foo, and some tools / test support code only works if one follows this convention. A typical <directory> name is out/Debug.
use_remoteexec = false # Approved users only. Do necessary setup & authentication first. is_debug = true # Enable debugging features. # Set true to enable experimental Skia backend. pdf_use_skia = false # Set true to enable experimental Fontations backend. pdf_enable_fontations = false pdf_enable_xfa = true # Set false to remove XFA support (implies JS support). pdf_enable_v8 = true # Set false to remove Javascript support. is_component_build = false # Disable component build (Though it should work)
In order to minimize the build time for embedder applications, test executables like pdfium_test are not built by default. To build these, add the argument pdf_is_standalone = true.
By default, the entire project builds with C++20.
By default, PDFium expects to build with a clang compiler that provides additional chrome plugins. To build against a vanilla one lacking these, one must set clang_use_chrome_plugins = false.
When complete the arguments will be stored in <directory>/args.gn, and GN will automatically use the new arguments to generate build files. Should your files fail to generate, please double-check that you have set use_sysroot as indicated above.
You can build the standalone test program by running: ninja -C <directory> pdfium_test You can build the entire product (which includes a few unit tests) by running: ninja -C <directory> pdfium_all
The pdfium_test program supports reading, parsing, and rasterizing the pages of a .pdf file to .ppm or .png output image files (Windows supports two other formats). For example: <directory>/pdfium_test --ppm path/to/myfile.pdf. Note that this will write output images to path/to/myfile.pdf.<n>.ppm. Run pdfium_test --help to see all the options.
There are currently several test suites that can be run:
It is possible the tests in the testing directory can fail due to font differences on the various platforms. These tests are reliable on the bots. If you see failures, it can be a good idea to run the tests on the tip-of-tree checkout to see if the same failures appear.
If your change affects rendering, a pixel test should be added. Simply add a .in or .pdf file in testing/resources/pixel and the pixel runner will pick it up at the next run.
Make sure that your test case doesn't have any copyright issues. It should also be a minimal test case focusing on the bug that renders the same way in many PDF viewers. Try to avoid binary data in streams by using the ASCIIHexDecode simply because it makes the PDF more readable in a text editor.
To try out your new test, you can call the run_pixel_tests.py script:
$ ./testing/tools/run_pixel_tests.py your_new_file.in
To generate the expected image, you can use the make_expected.sh script:
$ ./testing/tools/make_expected.sh your_new_file.pdf
Please make sure to have optipng installed which optimized the file size of the resulting png.
.in files.in files are PDF template files. PDF files contain many byte offsets that have to be kept correct or the file won't be valid. The template makes this easier by replacing the byte offsets with certain keywords.
This saves space and also allows an easy way to reduce the test case to the essentials as you can simply remove everything that is not necessary.
A simple example can be found here.
To transform this into a PDF, you can use the fixup_pdf_template.py tool:
$ ./testing/tools/fixup_pdf_template.py your_file.in
This will create a your_file.pdf in the same directory as your_file.in.
There is no official style guide for the .in file, but a consistent style is preferred simply to help with readability. If possible, object numbers should be consecutive and /Type and /SubType should be on top of a dictionary to make object identification easier.
The public/ directory contains header files for the APIs available for use by embedders of PDFium. The PDFium project endeavors to keep these as stable as possible.
Outside of the public/ directory, code may change at any time, and embedders should not directly call these routines.
Code coverage reports for PDFium can be generated in Linux development environments. Details can be found here.
Chromium provides code coverage reports for PDFium here. PDFium is located in third_party/pdfium in Chromium‘s source code. This includes code coverage from PDFium’s fuzzers.
The current health of the source tree can be found here.
There are several mailing lists that are setup:
Note, the Reviews and Bugs lists are typically read-only.
PDFium uses this bug tracker.
Report security bugs via Google Bughunters (Chrome VRP).
Project members and embedders can directly report security bugs using Chromium's security bug template.
See the CONTRIBUTING document for more information on contributing to the PDFium project.