Marius Volkhart [Sun, 14 Mar 2021 20:43:43 +0000 (20:43 +0000)]
Change Gradle to use java-library plugin
This plugin is specifically built for libraries. The major difference to the regular java plugin is that is allows defining dependencies as part of the api or implementation. Both are used by the project at compile/runtime, but only api dependencies are made available to dependent projects.
In our current setup, this doesn't matter much. We deploy to maven central using pre-built POMs. It's more of a future-proofing, and it makes it a little bit clearer which gradle projects actually require which dependencies.
Marius Volkhart [Sun, 14 Mar 2021 18:56:30 +0000 (18:56 +0000)]
Exclude batik-script dependency from OOXML artifact
We do not make use of the batik-script dependency. While this is likely to be true of a variety of the Batik dependencies, batik-script causes problems for our users who are using JPMS. See [bug-65103].
Marius Volkhart [Sun, 14 Mar 2021 18:51:12 +0000 (18:51 +0000)]
Specify more granular Batik dependencies
Batik-all is a strange artifact. It's POM declares dependencies on all the sub-JARs, but its JAR has all of the sub-jars repackaged. This results in multiple JARs with the same packages being added to consuming applications. This leads to problems for JPMS users. See [bug-65183].
The Ant build does not use batik-all, so the Maven and Gradle builds should not either.
Marius Volkhart [Sun, 14 Mar 2021 11:31:18 +0000 (11:31 +0000)]
Disable parallel tests on Gradle again
Something is causing parallel tests to fail on CI. I haven't been able to track down what it is. The symptoms look similar to others where the cause was a test modifying the test-data directory.
The integration tests also sometimes run into OutOfMemoryErrors when I run them in parallel.
Marius Volkhart [Sun, 14 Mar 2021 10:42:15 +0000 (10:42 +0000)]
Limit which tests can run in parallel
Some tests modify global resources. Those tests cannot be run in parallel with others, as they cause problems or become flaky. Where possible, indicate to JUnit the resources in contention. Otherwise, mark the tests as needing to run in isolation.
Reduce the number of map lookups necessary to compute the return values for methods that return collections of property details. Since we maintain parity between the `props` and `dictionary` contents, when retrieving property details, we can reference the `props` directly and avoid the `dictionary` indirection.
Marius Volkhart [Tue, 9 Mar 2021 19:26:31 +0000 (19:26 +0000)]
Additional debug logging for unknown records in HSLF
Recently, while debugging app behavior on HSLF documents, I had to dig into the OOXML that Microsoft PowerPoint places into files saved in PPT format. Having information in the logs about when records were not parsed by POI was very helpful. The hex identifier was critical in being able to quickly search the [MS-PPT] spec for what type of record it was, and the integer identifier was helpful in quickly finding the Record type in RecordTypes.java.
Marius Volkhart [Mon, 1 Mar 2021 00:25:23 +0000 (00:25 +0000)]
Deprecate functions that duplicate functionality
DrawingGroupRecord#processChildRecords and AbstractEscherHolderRecord#convertRawBytesToEscherRecords duplicate the functionality of AbstractEscherHolderRecord#decode. This makes the code harder to follow, as it is not clear when certain access patterns repeat. Accordingly, these functions are deprecated and flagged for removal.
Marius Volkhart [Mon, 1 Mar 2021 00:04:51 +0000 (00:04 +0000)]
Review EscherContainerRecord#getChildRecords() call sites for unnecessary work
This started off as wanting to add the EscherContainerRecord#getChildCount() function in order to do an efficient check for how many children the container has. This was desirable in new code for editing HSSF pictures. The existing option of calling getChildRecords().size() was undesirable as this requires a list copy first.
In the process of finding call sites that would benefit from replacing getChildRecords().size(), I realized that several other patterns would benefit from eliminating a copy, such as iterating over the children in a for-each loop, and indexed access to specific children.
Marius Volkhart [Sun, 28 Feb 2021 23:16:14 +0000 (23:16 +0000)]
Add the ability to edit HSLFPictureData contents
Pictures can now be edited by calling HSLFPictureData#setData(byte[]). The byte[] should contain the image data as an image viewer might read it.
To enable this functionality, a tighter coupling between the EscherBSERecords of the slideshow and the HSLFPictureData was required. This ensures that changes in image data size are accurately recorded in the records.
In the course of coupling the records and the HSLFPictureData, various scenarios arose where a mapping of records to pictures was non-trivial. Accordingly, the HSLFSlideShowImpl#matchPicturesAndRecords(...) function was added to perform a more sophisticated matching pass. This function is heavily exercised by org.apache.poi.hslf.usermodel.TestBugs.testFile[5] and PPTX2PNG.render[2], as well as the new TestPictures#testSlideshowWithIncorrectOffsets().
Marius Volkhart [Sun, 28 Feb 2021 19:18:13 +0000 (19:18 +0000)]
Rename EscherRecordHolder to OfficeArtContent
While the class does indeed hold EscherRecords, due to recent refactoring it is much more structured now than it was before. The contents of the class now closely resemble the OfficeArtContent structure referenced in the MS-DOC spec. Naming the class after the specification structure makes it easier to find and understand.
Marius Volkhart [Sun, 28 Feb 2021 18:49:42 +0000 (18:49 +0000)]
Rework EscherRecordHolder parsing
Modify the parsing done by EscherRecordHolder to be more deterministic. The format of the OfficeArtContent structure, which the EscherRecordHolder represents, is well defined in the MS-DOC spec. A clear class structure makes it easier to reason about the availability of data.
Marius Volkhart [Sat, 27 Feb 2021 00:43:56 +0000 (00:43 +0000)]
Fix bug with record indexes for HSSF Workbooks
We have encountered workbooks that do not have a TabIdRecord (see 55982.xls). However, the WorkbookRecordList#updateRecordPos() method would still increment the position of the TabIdRecord for such workbooks. Changing the default position of the record from 0 to -1 indicates that the record position has now been set.
This bug was discovered while adding support for editing pictures in HSSF documents.
Marius Volkhart [Fri, 26 Feb 2021 22:14:17 +0000 (22:14 +0000)]
Enable running more integration tests from Gradle.
This is particularly relevant for developers who import the Gradle project into their IDE and run tests using the Gradle test runner. Those developers were previously unable to debug integration tests.
Andreas Beeker [Mon, 22 Feb 2021 00:18:37 +0000 (00:18 +0000)]
Use upcoming XmlBeans 5.0.0 (working version 4.0.1)
Make batik optional as it doesn't work on the module-path
Use sub-components of batik as batik-all references them all again resulting in duplicated entries (maven poms haven't been migrated yet ...)
Remove SLF4j dependency
Marius Volkhart [Sun, 21 Feb 2021 20:57:25 +0000 (20:57 +0000)]
[bug-63046] Use Log4j 2 for logging
This removes the POILogger and POILogFactory mechanism for logging. This mechanism was created at a time when the Java landscape looked very different than it does today.
Log4j 2 is an Apache Foundation project that is well maintained and is widely regarded as having good performance and capabilities. We use only the Log4j API artifact. This lets application developers choose how they want to capture logging events if at all. Integrations with Log4j 2 Core and Logback are available from the Log4j project.
Dominik Stadler [Sat, 30 Jan 2021 18:41:57 +0000 (18:41 +0000)]
Enable storing an SXSSF workbook twice
Verify this via a unit-test for all types of workbooks
Add a test to verify the contents is byte-for-byte equal
Also remove an overwritten test which works the same now for all three workbook-types
projects / poi.git / log