CHANGES.md

Change Log

Version 0.15.3 - 2026-04-01

• (fix) The 0.15.2 release did not include the legacy FUSE v2 driver binaries (dwarfs2) in the static release tarballs.

• (fix) The op_readlink implementation is now guaranteed to terminate the returned string, as required by the FUSE specification.

• (build) Warning suppressions for (newer) GCC warnings triggered warnings on older GCC versions. These have been fixed.

• (other) Some automatic code refactoring from clang-tidy was applied and some minor issues were fixed in the process.

Version 0.15.2 - 2026-03-31

• (fix) The image size was not correctly passed to an instance of the file system parser, which resulted in errors when trying to load an image embedded in a larger (e.g., multi-layer) file. Thanks to Ruan Formigoni for the pull request that fixed this issue.

• (fix) The performance monitor timer for op_lseek in the FUSE driver was not set up correctly, which could lead to segmentation faults or bus errors due to an uninitialized index into a std::deque. This has been fixed, and an additional check has been added to make sure these kinds of errors are caught in the future.

• (fix) A few test failures have popped up in openSUSE Tumbleweed native 32-bit builds. One was caused by glibc requiring _FILE_OFFSET_BITS=64 for 64-bit file operations. This caused some sparse file tests that created files larger than 4 GiB to fail. This was never needed in musl-libc builds, so the problem was not caught before. The other failures were mostly cosmetic and only showed up in GCC builds. The affected tests were checking the formatting code for times and ratios, which relied heavily on floating-point arithmetic and was not deterministic across platforms and compilers. The code has been completely rewritten to use integer arithmetic and avoid any platform-specific behavior. The final issue was a bug in the test code that, interestingly, also surfaced only with GCC. Addresses GitHub issue #354.

• (fix) The FUSE drivers (dwarfs, dwarfs2) have gone through several stages of refactoring to fix a large number of very subtle issues. For example, it is now guaranteed that you will see an error for a misspelled option rather than an error caused by the absence of the option you were trying to use. For example, if you used -o image_size=1234 instead of the correct -o imagesize=1234, you would likely get an error saying that the file system could not be loaded rather than an error about the unknown option. Before the refactoring, the code was a horrible mess of preprocessor conditionals, and basically every flavor of the FUSE driver (high-level, low-level, Windows, FUSE v2, or FUSE v3) had a different code path during startup, exposing a different set of quirks. The code is now much more readable, a lot more consistent, and even better tested than before.

• (fix) The manual pages shown with --man for all tools unintentionally contained the license header that had been added in XML comments at the top of the source files. The renderer has been fixed to ignore those comments.

• (build) After benchmarking the latest mimalloc memory allocator, it turns out that performance is mostly on par with jemalloc. It is still much less configurable than jemalloc, but it is definitely usable if you do not need that configurability.

• (build) The small universal release binaries are now built with mimalloc instead of jemalloc, reducing their size by about 10%.

• (build) Old Clang compiler versions, such as those on Ubuntu 22.04, are no longer supported because they cannot use libstdc++'s std::expected implementation. DwarFS can still be built on Ubuntu 22.04 with GCC.

• (test) After profiling test execution, a few tests stood out as being particularly slow. These have been modified to run faster while still covering most of the same code paths.

Version 0.15.1 - 2026-03-21

• (fix) mkdwarfs did not correctly handle inputs where hardlinks had the same inode number on different devices. To run into this issue, you would have to make mkdwarfs scan files from multiple devices (e.g. the root of a directory tree with multiple mounted filesystems) and have files with the same inode number on different devices and have at least two of those files also have a link count greater than 1. While this is hopefully rare in practice, it is a serious bug that can lead to crashes (in the best case) or even data loss (in the worst case), as only the data of one of these files would be stored in the image. This has been fixed and a test has been added to cover this case.

• (fix) A missing dependency was causing linker errors with shared library builds on macOS. This has been fixed.

• (build) The static release binaries are now all built using Clang and link-time optimization. This was previously not the case for some architectures due to bugs in the toolchain. As a result, the binaries are now significantly smaller.

• (build) There is now a new set of binaries (dwarfs-universal-small) that are built without brotli support and without support for the performance monitor. The performance monitor is rarely used and brotli compression comes with a huge dictionary that bloats the binary size without offering much benefit over lzma or zstd in most cases. If you care about binary size, these new binaries are a good default choice.

Version 0.15.0 - 2026-03-18

• (fix) Commas in the filesystem image path were not escaped when passed to the FUSE driver as the fsname option. Because commas are FUSE argument separators, this caused mounting to fail with an irritating error message for paths containing commas. Fixes GitHub issue #323.

• (fix) Progress reporting in dwarfsextract was broken when extracting a subset of files using patterns, as it was computed relative to the total filesystem size rather than the total size of the files being extracted. There were several subtle issues that could lead to progress percentages either not reaching 100% or even exceeding it, all of which should now be fixed. Fixes GitHub issue #316.

• (fix) Fixed FUSE argument vector initialization in dwarfs_main, which could trigger an assertion inside libfuse when extra arguments were added after an uninitialized vector was passed to FUSE_ARGS_INIT.

• (fix) The Windows build of mkdwarfs no longer aborts with a fatal error when it encounters an empty file during scanning.

• (fix) Fixed a metadata lookup bug for parent_dir_entry in filesystems with format version 2.2 and earlier (DwarFS releases before v0.5.0), where the lookup required an additional level of indirection that was previously missing. Fortunately, this bug only affected the debug output of dwarfsck: the parent= field shown with -d directory_tree would show the inode number of the parent directory instead of the entry number. The only other code path that used parent_dir_entry effectively "fixed" the bug by applying the missing indirection.

• (fix) Recompressing a filesystem image with sparse files without also rebuilding the metadata would erroneously trigger an error stating that sparse file support cannot be disabled, even without --no-sparse-files. The root cause was an unchecked std::optional access, which has been fixed.

• (fix) When rewriting a filesystem image, the bytes_in and bytes_out progress counters were not updated simultaneously. bytes_in was updated before compression, while bytes_out was updated after compression. This could lead to incorrect compression ratios being reported in the progress display. This has been fixed by updating both counters simultaneously after compression.

• (fix) When using --format=newc and extracting a subset of hardlinked files, dwarfsextract would crash with an "unexpected deferred entry" error. This is due to a peculiarity of libarchive's newc implementation that was not handled properly before. This has been fixed, and a test has been added to cover this case.

• (fix) Corrected the license information in a few headers, changing them from GPL-3.0-or-later to MIT.

• (feat/build) Major dependency reduction by mostly "de-Meta-ing" the project. The fbthrift and folly libraries are no longer dependencies of DwarFS, and the submodules have been removed from the repository. fbthrift has been replaced by a new thrift_lite library that implements only the subset needed by DwarFS (the thrift compiler, compact protocol, JSON serialization, debug output, and frozen-layout support). The frozen library that is part of fbthrift has been forked and is now maintained as an internal component of DwarFS. The fork has been cleaned up and refactored to remove all folly dependencies. By using a new abstraction for accessing bit-packed data, the frozen code is now measurably faster than the original. The folly library has been replaced by a combination of standard C++23 facilities, Boost facilities, and a few new in-repo components. As a result, gflags, glog, double-conversion, and libevent are also no longer dependencies. On macOS, this also removes the libsodium dependency. Because this change uses much simpler abstractions where the complexity of the original libraries was unnecessary, there has also been a significant reduction in binary size, although some of that reduction is offset by the addition of new features. Note that the thrift_lite library is only just "compatible enough" with the original fbthrift to ensure forward and backward compatibility. The compact protocol is fully compatible, but the debug and JSON output formats are not identical. The debug output will look very familiar, but not identical. The JSON output leans toward the "simple" JSON serialization style, but uses lists of pairs instead of objects for maps. The latter difference also affects the --export-metadata option of dwarfsck, which will produce very different but much more easily digestible JSON output.

• (feat) mkdwarfs now automatically selects the progress display mode based on whether the output is connected to a terminal and whether the current locale uses UTF-8. Previously, the default was always unicode, which could produce garbled output in non-UTF-8 environments. Addresses GitHub issue #326.

• (feat) The project is now compliant with the REUSE specification. All source files have been annotated with SPDX license identifiers, a REUSE.toml configuration file has been added, and full license texts are provided in the LICENSES directory.

• (feat) New --hollow option for mkdwarfs that allows building "hollow" filesystem images. Instead of storing actual file data, each file is represented as an empty sparse file with the same size and metadata as the original. When accessing a hollow filesystem, file reads return all zeros. This can be useful when a realistic filesystem structure is needed for testing, but the actual file contents do not matter. Fixes GitHub issue #131.

• (feat) mkdwarfs now supports ZSTD long-distance matching (LDM) via a new long algorithm option for --compression. Enabling LDM can improve compression ratios when used with extremely large block sizes (typically blocks larger than 128 MiB), or when using smaller block sizes at lower compression levels. Additionally, all ZSTD compression parameters can be tuned if the ZSTD library is linked statically and does not expose a bug that leads to a division-by-zero error when setting the parameters manually (see facebook/zstd#4590). The statically linked release binaries contain a patched version of libzstd and support tuning all parameters. Fixes GitHub issue #322.

• (feat) New binary file categorizer (--categorize=binary) that can identify ELF (Linux/FreeBSD), PE (Windows), and Mach-O (macOS) executables and shared libraries and group them into separate categories by type and architecture. This can dramatically improve compression ratios in cases where lots of binaries from different platforms and architectures are mixed together. It can still help in other cases simply by grouping binaries and non-binary files into distinct streams.

• (feat) dwarfsextract has a new --num-disk-writers option to run multiple writer threads in parallel when extracting files to disk. This can improve throughput at the cost of higher CPU usage, especially when extracting large numbers of small files.

• (feat) dwarfsextract has new options --skip-devices and --skip-specials to skip device nodes and special files (sockets, FIFOs) during extraction.

• (feat) dwarfsextract now emits a warning when a pattern is provided but no matching files are found.

• (feat) dwarfsck can now export metadata to stdout instead of a file using --export-metadata=-.

• (feat) With mkdwarfs in --input-list mode, specifying --order=none now preserves the exact order of entries as given in the input file. Previously, none was not treated as a meaningful ordering guarantee in this mode.

• (feat) New --no-check option for mkdwarfs that skips the filesystem integrity check before recompression. This can speed up recompression workflows when the source image is assumed to be valid. The individual checks will still be performed during the rewrite. Addresses GitHub issue #322.

• (feat) When recompressing a filesystem image, blocks that are uncompressed in the source filesystem are no longer unnecessarily copied into memory. Instead, the mapped memory region is passed directly to the compressor. This saves a bit of memory and CPU time in this specific case.

• (feat) While profiling mkdwarfs, the memory usage code turned out to be a significant hotspot, as reading from /proc/self/smaps_rollup is surprisingly expensive. To mitigate this, the code now reads from /proc/self/status by default, accepting potentially less accurate memory usage data in exchange for much lower overhead. It is still possible to use /proc/self/smaps_rollup by setting DWARFS_ACCURATE_MEMORY_USAGE=1 in the environment. This can be useful alongside DWARFS_LOG_MEMORY_USAGE, which can be set to the name of a log file where memory usage is logged periodically during the mkdwarfs run. When /proc/self/smaps_rollup is inaccessible, the code will automatically fall back to /proc/self/status.

• (feat) Added FreeBSD support for memory usage tracking in mkdwarfs using the kinfo_proc interface.

• (feat) Memory usage during mkdwarfs rewrite operations is now properly limited by the -L/--memory-limit option, taking into account the memory used by both the queued blocks and the compression algorithm itself. Addresses GitHub issue #322.

• (feat) The similarity ordering option now uses a different hash mixing function. The main motivation was to ensure better distribution, since only a small number of hash bits are actually used. This may or may not improve compression ratios, but it will affect the resulting image size.

• (build) The project now requires C++23 compiler support (previously C++20). Care has been taken to ensure only widely available and long-supported C++23 features are used.

• (build) Ubuntu 26.04 (resolute) and Debian testing have been added to the continuous integration matrix.

• (build) Cleaned up lots of compiler warnings on different platforms.

• (docs) Documentation for the fits, hotness, and binary categorizers was added to the mkdwarfs manual page.

• (test) Test coverage has been significantly improved from 96.4% to 97.1%, with more than 10,000 lines of new test code.

Version 0.14.1 - 2025-10-25

• (fix) The new Windows code from the v0.14.0 release that implemented lstat()-like functionality to get file status information did not work correctly for reparse points that were not symbolic links (i.e. "real" symlinks or junctions). When dealing with such reparse points, the code would incorrectly determine the allocated size of the file to be zero, since it tried to determine that size from the reparse point and not from the target. This caused the progress percentage to be calculated incorrectly, since much more data was processed than was expected, leading to progress percentages well over 100%. Thankfully, @lunighty not only reported this issue, but also provided significant context by mentioning that this was on a Windows Server NTFS volume with deduplication enabled. Fortunately, this was also reproducible on "regular" Windows using a mounted WIM image. The fix will now correctly handle reparse points by re-opening the handle without FILE_FLAG_OPEN_REPARSE_POINT in order to open the target. This bug only affected Windows builds of mkdwarfs, and only when actually used on filesystems with reparse points that were not symlinks. Even then, the resulting DwarFS image would still be correct, except for a way too small total_allocated_fs_size metadata entry. However, this entry is purely informational. Huge thanks to @lunighty for both reporting and helping to debug this issue! Fixes github #307.

• (fix) Related to the above, the metadata_builder now recomputes all total sizes (total_fs_size, total_allocated_fs_size, and total_hardlink_size) as part of the build() function. This not only ensures that the totals are correct even if the allocated size changes between scanning and segmenting (which has been happening at least on ZFS volumes), but it also allows images affected by the above bug to be fixed by rebuilding the metadata.

• (fix) A bunch of changes were made to make the static (Linux) release binaries work on FreeBSD with the Linux emulation layer. While you can easily build DwarFS on FreeBSD natively, the static binaries might be bundled in Linux application images that you may want to run on FreeBSD as well. This required patching musl to fall back to FUTEX_WAKE if FUTEX_REQUEUE is not supported (which is the case on FreeBSD's Linux emulation); otherwise, the binaries would regularly hang when trying to shut down thread pools. Furthermore, the SFX stub did not work with the emulation layer due to lack of support for fexecve; so it now uses a fallback of temporary file + execve in case create_memfd and/or fexecve are not supported. This fallback is very similar to the one that was already in use for old Linux kernels, only that cleanup of the temporary file is now done using a child "janitor" process as soon as the execve call succeeds. Lastly, all static binaries are now branded as using the Linux ABI rather than the generic "System V" in the ELF header, which makes it unnecessary to explicitly brandelf the binaries on FreeBSD.

• (fix) The pcmaudio categorizer had two minor issues discovered by @lunighty when compressing a large number of WAV files. One was reporting an unsupported format: 3/0 or unsupported format: 65,534/3 warning, which isn't very useful for the end user. These format codes correspond to IEEE floating point formats, which are indeed unsupported. However, the format appears to be quite common, so the warning has been downgraded to an info message that explicitly mentions the floating point format. The second issue was an unexpected fmt chunk size of 20 bytes, which caused the file to be rejected as a PCM audio file (meaning it was added using a generic compressor instead of FLAC). It turns out that these non-conforming fmt chunks are also quite common in practice, so the code has been changed to accept the non-conforming file, but also logging an info message mentioning the non-conformance. Fixes github #309.

• (fix) The test_helpers library used for unit tests was not explicitly listing xxhash as a dependency, which caused linker errors on macOS. These unfortunately only showed up in the Homebrew CI. The dependency has been added.

• (fix) Instead of making the FUSE drivers fail hard when seeing the options that were removed in v0.14.0, they now just log a warning and ignore them. The options may still be fully removed in a future release. Fixes github #303.

• (fix) The help text for the mkdwarfs compress level option (-l) was misleading in combination with the manual page as neither mentioned that the table with details was shown only by -H / --long-help. Fixes github #312.

• (feat) Added shell completion for dwarfsck and dwarfsextract. (Thanks to Ahmad Khalifa for the contribution.)

• (feat) Added sample desktop unmount handlers. (Thanks to Ahmad Khalifa for the contribution.)

• (docs) Updated diagram in dwarfs-format.md to show new inode metadata fields.

• (build) Bumped libarchive, libressl, and libdwarf versions in the static build pipeline.

• (test) Added test for file_stat on symbolic links containing multibyte characters on Windows.

• (test) Added test for obsolete FUSE driver options being ignored and producing a warning.

• (test) Added test for the FUSE driver's imagesize option.

• (test) Added more tests for the --auto-mountpoint option in the FUSE driver.

Version 0.14.0 - 2025-10-21

• (fix) Leading dots in --input-list file paths were incorrectly treated as literal directory names instead of being expanded. This has been fixed. Fixes github #292.

• (fix) The SPDX license identifier in GPL-licensed source files was incorrectly specified as GPL-3.0-only instead of GPL-3.0-or-later. This has been corrected. Fixes github #275.

• (fix) Fixed an off-by-one error when recovering self_index fields in metadata, which could cause the sentinel directory to have a non-zero self_entry. While harmless by itself (since that entry is never actually used), this would cause the metadata consistency check to fail. The fix covers three aspects: correcting the off-by-one error; ensuring the self_entry recovery code does not run for the sentinel directory; and changing the metadata consistency check to only warn about a non-zero self_entry rather than fail. Running mkdwarfs with --rebuild-metadata will also reset a non-zero sentinel self_entry to zero.

• (fix) Fixed the implementation of the read operation in the FUSE driver to send positive error code values to libfuse. This was likely never triggered in practice, but in cases where parts of the filesystem image vanish while being accessed (which previously caused SIGBUS crashes), libfuse would not understand the negative error codes.

• (fix) Moved the FUSE driver binaries from sbin to bin and kept only the mount.dwarfs/mount.dwarfs2 symlinks in sbin. This better aligns with user expectations, other FUSE drivers, and the fact that the man pages are installed in section 1. (Thanks to Ahmad Khalifa for the fix.)

• (fix) The dwarfs2 binary was broken in builds using shared libraries. (Thanks to Ahmad Khalifa for the fix.)

• (fix) When setting CPU thread affinity for worker group threads via DWARFS_WORKER_GROUP_AFFINITY, the code did not CPU_ZERO the cpu_set_t structure before setting individual CPUs. This could pin threads to random CPUs in addition to the requested ones.

• (fix) The FITS categorizer would scan entire files for the end-of-header marker if their size was a multiple of 2880 bytes, causing significant slowdowns on large non-FITS files. Additional checks now ensure scanning only continues if the data truly looks like a standards-compliant FITS header.

• (fix) GCC caught a potential null-pointer dereference on error when opening a file in mkdwarfs. This has been fixed.

• (fix) Numerous fixes for 32-bit architectures, mostly related to integer overflows with file sizes larger than 4 GiB.

• (fix) Another off-by-one error caused the first regular file inode to be excluded from the file-size cache. This would be hard to notice unless that file was highly fragmented. The cache will be fixed when rebuilding the metadata.

• (fix) The FUSE driver’s enable_nlink option is now the default behavior and cannot be disabled. The previous optimization skipped building a table of hardlink counts, which produced inherently incorrect file status information (hardlinked files share an inode, so reporting a link count of 1 is wrong). The hardlink table is now stored in the metadata by default; if there are no hardlinks, it consumes no space. You can still omit the hardlink table with --no-hardlink-table, at the cost of building it on-the-fly when the filesystem image is loaded (typically fast — e.g., ~300 ms for 14 million files).

• (fix) Fixed a typo in dwarfs-format.md. (Thanks to Dennis Brakhane for spotting this and sending a PR.)

• (feature) New I/O layer abstraction that supports “classic” mmap-based file access, granular mmap-based access on 32-bit systems, and fully mmap-less access if desired. This applies to all DwarFS tools. By default, tools use the most efficient method—memory-mapping whole files on 64-bit systems and mapping file segments on 32-bit systems (to conserve address space). This can be controlled via the new DWARFS_IOLAYER_OPTS environment variable described in dwarfs-env(7).

• (feature) Full support for sparse files. mkdwarfs now detects and efficiently processes sparse files, skipping holes where possible and preserving them in the filesystem image. This is supported on all platforms. The FUSE driver implements lseek() where supported by the FUSE library (currently Linux and FreeBSD); Windows and macOS fall back to showing files as non-sparse. dwarfsextract extracts sparse files as such and preserves sparse representations when extracting to archive formats that support them (e.g., tar). Note: Sparse file support is not backwards compatible; images containing sparse files cannot be processed by DwarFS versions prior to 0.14.0. By default, mkdwarfs enables sparse file support if it detects sparse input. Use --no-sparse-files to disable it and ensure compatibility with older versions.

• (feature) Support for subsecond timestamp resolution. The default remains one second, but finer resolutions (down to nanoseconds) can be specified with --time-resolution. mkdwarfs will warn if the requested resolution is finer than the native filesystem resolution. This is fully backwards compatible: older DwarFS versions will handle such images but ignore the subsecond parts.

• (feature) Desktop integration for Linux. A new --auto-mountpoint option automatically creates or selects a mount-point directory, making it easier to mount DwarFS images from file managers. Desktop files and MIME type definitions are now installed to enable double-click mounting of .dwarfs files. (Thanks to Ahmad Khalifa for the implementation.)

• (feature) Shell completion for mkdwarfs (bash and zsh). (Thanks to Ahmad Khalifa for the contribution.)

• (feature) Improved error handling when DwarFS tools encounter SIGBUS (usually caused by accessing memory-mapped files on unreliable or faulty storage like network shares or flaky USB drives). When SIGBUS is caught, tools now print an error suggesting switching from mmap- to read-based I/O via DWARFS_IOLAYER_OPTS.

• (feature) dwarfsck now checks metadata consistency by default (unless --no-check is given), improving detection of filesystem image corruption.

• (feature) If sparse files are supported by the FUSE library, the FUSE driver exposes new options cache_sparse and no_cache_sparse to control whether sparse files should be cached in the kernel page cache. See dwarfs(1) for details.

• (feature) The JSON output from dwarfsck now contains a complete raw metadata dump when the detail level includes metadata_full_dump.

• (feature) dwarfsck no longer artificially limits string sizes when dumping metadata. (Thanks to Dennis Brakhane for the contribution.)

• (feature) Accelerated search for the start of a DwarFS image in files with custom headers; the new code is about four times faster, scanning at more than 6 GiB/s on a modern CPU.

• (feature) The cache size can now be configured for dwarfsck, useful with the --checksum option.

• (feature) Both dwarfsck and dwarfsextract now limit the amount of data requested from the filesystem image at once to avoid exhausting memory (and virtual address space on 32-bit systems).

• (feature) Improved self-extracting binary stub with better compatibility for qemu, binfmt_misc, and old kernels. The stub now works on Linux kernels as old as 2.6.21 (and possibly older), and it now uses nanoprintf to further reduce binary size.

• (feature) The accepted minor version for the DwarFS image format has been incremented. Release v0.16.0 will also increment the written minor version. This means images produced with v0.16.0 will not be readable by DwarFS tools prior to v0.14.0. See the “Features” section in dwarfs-format(7) for details.

• (feature) The FUSE driver will now show the name of the mounted file system image in the mount point listing (e.g., in df or mount output) on Linux, FreeBSD and macOS, as well as the filesystem subtype (dwarfs) on Linux and FreeBSD.

• (compat) The (no_)cache_image option has been removed from the FUSE driver.

• (docs) Added documentation on manual FSST decoding to dwarfs-format.md. (Thanks to Dennis Brakhane for the PR.)

• (docs) Several cleanups and additions to dwarfs-format.md, including a glossary of terms, clarification of blocks vs. sections, and descriptions of compatibility handling via features, plus details on the representation of sparse files and hardlinks.

• (docs) New manual page dwarfs-env(7) documenting DwarFS-specific environment variables.

• (build) Removed the dependency on Boost.Iostreams and the hard dependency on Boost.System.

• (build) Removed the hard dependency on the date library, which caused build issues on distributions that no longer bundle it (e.g., SUSE).

• (build) The build system now creates symbolic mount links at install time rather than in the build directory.

• (test) Significantly improved test coverage.

Version 0.13.0 - 2025-08-29

• (fix) The linker configuration for the release binaries was broken. The symptom of this was that very occasionally, tests would fail in the CI with std::terminate being called after the exception handling code failed to unwind the stack. The root cause was that in clang builds, code from libunwind and code from libstdc++ was rather arbitrarily mixed, which, depending the order in which individual threads were scheduled in the unit tests, could lead to stack unwinding working flawlessly, or not at all. This was one of the hardest bugs to track down this year, fortunately the fix was quite simple. In any case, there is potential for this issue being present in the previously released binaries, although there haven't been any reports.

• (fix) Made section index discovery more robust. Fixes github #264.

• (fix) A recent kernel change (https://lkml.org/lkml/2025/5/5/2868) caused the tools_test to fail on Linux 6.14 and later. This has been fixed by accepting both EPERM and ENOSYS as valid error codes for link() calls.

• (feature) Support for FreeBSD. Everything that works on Linux should also work on FreeBSD. There are no static binaries, but the build should work out of the box once all dependencies are installed.

• (feature) Support for big-endian architectures. This is still experimental, even though all unit tests pass with QEMU, and the benchmark suite runs fine on real hardware. This currently requires forked versions of folly and fsst. The changes are small and the pull requests will hopefully be merged upstream soon.

• (feature) Experimental support for 32-bit architectures. While DwarFS should mostly "just work" on 32-bit when using small images (a few hundred megabytes), the limited address space is a problem for the extensive use of memory-mapped files inside DwarFS. There will be changes to limit the use of mmap in the future (mainly due to other issues), which should help 32-bit compatibility as a side-effect. Fixes github #268.

• (feature) Support for a wide range of CPU architectures. The static binary releases (including the universal binaries) are now available for x86_64, aarch64, i386, arm, ppc64, ppc64le, riscv64, s390x, and loongarch64. Trying to build the new release binaries has uncovered a few bugs in clang, binutils, mold, and upx, not all of which have been fixed yet. As a result, the binaries use slightly different toolchains and configurations depending on the architecture. Fixes github #266, #268.

• (feature) Custom self-extracting binary stub for universal binaries. This aims for simplicity and portability and should work on most Linux systems. This is used if upx support for an architecture is not available, or if the binaries extract much faster than the upx compressed version. The stub also supports a --extract-wrapped-binary <file> option to extract the embedded binary.

• (feature) The category metadata for categorized blocks is now stored in the metadata block by default. This allows re-compressing the blocks with a metadata-dependent algorithm (e.g. FLAC) even if they were previously compressed using a metadata-independent algorithm. This can be disabled using the --no-category-metadata option. See the mkdwarfs man page for more details.

• (feature) The --no-category-names and --no-category-metadata options can be used to reduce the size of the metadata. However, this will make it impossible to use metadata-dependent compression algorithms (e.g. FLAC), or even select category-specific compression, when recompressing the image.

• (feature) Metadata rebuilding is now supported in mkdwarfs using the --rebuild-metadata option. Previously, the metadata could only be recompressed, but it was impossible to change it. With the new option, it is now possible to change metadata packing and apply operations like --set-owner, --set-group, --set-time, --time-resolution, --chmod, or --no-create-timestamp. Note that these are potentially lossy operations that may be irreversible. By default, the history of metadata rebuilds is tracked in the metadata itself, but this can be disabled using --no-metadata-version-history.

• (feature) In addition to metadata rebuilding, it is now also possible to change the block size of an existing image using the --change-block-size option. This implies --rebuild-metadata and --recompress=all. This can be useful for tuning the performance of an existing image without having to re-create it from scratch.

• (feature) mkdwarfs now shows its current memory usage while running. Note that -L/--memory-limit still only limits the memory used for the block queue, not the overall memory usage. Fixing this is on the roadmap, there's no need to file an issue.

• (feature) dwarfsextract has new options to control the output format: --format-options and --format-filters. There is also --format=auto to automatically "guess" the format and filters based on the output file name. (Thanks to @oxalica for the pull request.)

• (feature) dwarfsck has a new frozen_details detail level that will show the frozen_analysis content ordered by memory location instead of memory usage and also shows the address range of each section.

• (feature) Replaced folly’s EvictingCacheMap with a simple in-repo LRU cache implementation. Reduces external dependencies and binary size while not sacrificing performance.

• (feature) The pxattr utility now supports all extended attribute operations, including setxattr and removexattr, on Windows. Also, error handling/reporting for extended attributes on Windows has been improved.

• (docs) Updated dwarfs-format.md with more info on section types, compression metadata, and Frozen2 binary metadata layout. (Thanks to @oxalica for asking the right questions, reporting bugs, and ultimately releasing a Rust library to read/write DwarFS images.)

• (docs) Added notable users section to README. (Thanks to Vitaly Zdanevich for the PRs.)

• (docs) Updated mkdwarfs docs with more info on worker threads and requirements for bit-identical images.

• (docs) Major README overhaul. Added quick start section, added more links, fixed typos and wording.

Version 0.12.4 - 2025-05-14

• (fix) Segfault on bad_compression_ratio_error. When recompressing a filesystem where some blocks cannot be compressed using the selected algorithm because of a bad_compression_ratio_error, the resulting block was left empty after the refactoring done in 06f8728cc.

• (fix) Add history unless --no-history is given when rewriting a file system image.

• (fix) Allow dumping frozen_layout w/o frozen_analysis in dwarfsck.

• (fix) Logging timestamps should show local time.

• (fix) Workaround a weird MSVC bug.

• (fix) Remove useless cast causing compiler warning.

• (feat) More complete breakdown of metadata in dwarfsck.

• (feat) Add schema_raw_dump flag to dwarfsck --detail.

• (build) Switch static build to libressl on Windows.

• (build) Update static build libraries.

• (build) Update folly/fbthrift/fsst.

• (refactor) Use unordered_set::contains to simplify check.

• (refactor) Introduce and use safe_localtime() to prevent issues with fmt deprecating fmt::localtime.

• (test) Speed up a few slow tests on Windows.

Version 0.12.3 - 2025-04-21

• (fix) Automatic image offset detection (for images using a custom header) did not work correctly if the header contained a string that would be identified as the start of a v1 section header (these were only used before dwarfs-0.3.0). If there was either "DWARFS\x02\x00" or "DWARFS\x02\x01" in the header, offset detection would fail. The check has been modified to peek further into the data and ensure this really is a v1 section header, and also checking if the next section header position can be derived from the length field. It is still possible to construct a file system image where offset detection will ultimately fail, but it is much less likely with the change.

• (build) The build process for the release binaries has been further tweaked to reduce binary size. The dwarfs-fuse-extract binary now again supports extracting files by pattern; I didn't realize that this was actually a widely used feature before dropping it in the last release. dwarfs-universal is now linked against LibreSSL's libcrypto instead of OpenSSL's. This significantly reduces the size at the expense slightly slower cryptographic hash functions. However, this will likely only be perceivable when using --tool=dwarfsck with either --check-integrity or --checksum. The binaries from the release tarballs are still linked against libcrypto from OpenSSL.

Version 0.12.2 - 2025-04-16

• (fix) A performance regression introduced in v0.12.0 was fixed. This caused pcmaudio compression in mkdwarfs to take more than twice as long as in the previous releases.

• (build) A few small refactoring changes to further reduce the size of the fuse-extract binary. In particular, the performance monitor and the history feature are now fully removed. Also, the functionality to extract in different archive formats as well as to extract only files matching a pattern have been removed, so the image can only be fully extracted to disk.

Version 0.12.1 - 2025-04-10

• (fix) Attempt to fix linking issue in Homebrew build.

• (feature) Added --memory-limit=auto to mkdwarfs to use a more reasonably (hopefully) default for the block queue. The old default of 1 GiB was quite arbitrary and definitely not suitable for low-end systems. The new auto default will determine the limit based on the number of workers (which in turn is based on the number of CPUs), the block size, and the amount of physical memory of the system.

• (feature) Replace vector_byte_buffer with malloc_byte_buffer, which is internally based around a simple buffer that doesn't incur the cost of initializing each element like std::vector. Especially for large blocks which are known to be overwritten immediately, this can save a few CPU cycles.

• (feature) In the x86_64 release binaries, use an optimized memcpy implementation if supported by the CPU.

Version 0.12.0 - 2025-04-08

• (fix) Build release binaries against an up-to-date libfuse. Fixes github #252.

• (fix) Changes for compatibility with Boost.Process v2.

• (feature) Re-licensed all libraries required for reading DwarFS images under the MIT license. The source of all tools that just read DwarFS images (i.e. everything except for mkdwarfs) are also under the MIT license now. Everything else is still GPL-3.0. Addresses github #255.

• (feature) Significantly reduced binary size in the static release builds. This is the result of refactoring code that unconditionally pulled in code-heavy dependencies such as libcrypto, as well as optimizing the build pipeline (e.g. building dependencies with only the necessary set of features) and turning on link time optimization.

• (feature) A new kind of "universal" binary dwarfs-fuse-extract is part of the release now. This combines the FUSE driver (dwarfs) and dwarfsextract into a single binary, but does not include the mkdwarfs and dwarfsck tools that are also part of the regular universal binary. dwarfs-fuse-extract is much smaller than the regular universal binary and especially suitable to AppImage-like applications.

• (feature) New hotness categorizer in mkdwarfs that allows a list of "hot" files to be stored in distinct file system blocks.

• (feature) New explicit ordering mode in mkdwarfs that allows files to be ordered accoring to the order in a given list file.

• (feature) dwarfs now shows the version of the FUSE library used.

• (feature) New dwarfs options preload_all and preload_category to populate the block cache immediately after mounting.

• (feature) New dwarfs option analysis_file that can be used for profiling and as input to mkdwarfs new hotness categorizer and explicit ordering mode.

• (feature) New dwarfs option block_allocator that allows the user to switch from a malloc-based block allocator to an mmap-based one. This can help with returning memory back to the system if the blocks are evicted from the cache.

Version 0.11.3 - 2025-03-31

• (fix) Handle absolute paths in --input-list. Fixes github #259.

• (fix) Don't prefetch blocks that are already in the active list within the block cache.

• (fix) Ensure that statistics for block tidying are correctly updated in the block cache.

• (build) A few build fixes, mainly to simplify building on Alpine.

Version 0.11.2 - 2025-03-20

• (fix) macOS Ventura's version of clang appears to be missing an implementation of std::hash<std::filesystem::path, making it hard to define an unordered_map<filesystem::path. Work around by simply using an unordered_map<string> instead.

• (fix) Installing the binaries using cmake did not honor the CMAKE_INSTALL_BINDIR or CMAKE_INSTALL_SBINDIR variables. Fixes github #253.

Version 0.11.1 - 2025-03-18

• (fix) macOS Ventura's version of clang appears to be missing the <source_location> header, despite Apple claiming otherwise. Fix this by shipping a wrapper and providing a fallback implementation.

Version 0.11.0 - 2025-03-17

• (fix) Remove the access implementation from the FUSE driver. There's no point here trying to be more clever than FUSE's default. This makes sure DwarFS will behave more like other FUSE file systems. See github discussion #244 for details.

• (fix) Limit the number of chunks returned in inodeinfo xattr. Highly fragmented files would have megabytes in inodeinfo, which not only breaks the xattr interface, but can also dramatically slow down tools like eza who like to read xattrs for no apparent reason.

• (fix) Avoid nested indentation due to ronn-ng bug. Fixes github #249.

• (fix) Don't link library against jemalloc. This fixes both issues with pydwarfs and issues building with jemalloc support on macOS. Only the binaries are now linked against jemalloc, which should be sufficient.

• (feat) Support case-insensitive lookups. Fixes github #232.

• (feat) Allow setting image size in FUSE driver. Fixes github #239.

• (feat) Support extracting a subset of files with dwarfsextract using the new --pattern option. The same glob patterns can be used as for the filter rules in mkdwarfs. Fixes github #243.

• (feat) Allow overriding UID / GID for the whole file system when using the FUSE driver on non-Windows platforms. See github discussion #244.

• (feat) Expose more LZMA options (mode, mf, nice, depth).

• (feat) Improve filter patterns, which now support ranges and complementation.

• (feat) Improve speed of filesystem walk / walk_data_order calls by 80% / 40%. The impact of this will largely depend on what the code is being run for each inode, but, for example, the speed of listing more than 14 million files with dwarfsck will take about 16 seconds compared to 17 seconds with the previous release.

• (feat) Added an inode size cache to the metadata to speed up file size computation for large, highly fragmented files. The configuration is currently fixed using a conservative default. Only files with at least 128 chunks will be added to the cache, so in a lot of cases this cache may be completely empty and not contribute to the size of the file system image at all.

• (feat) Use bit-packing for hardlink, shared files, and chunk tables. This will consume less memory when loading a DwarFS image.

• (feat) Show total hardlink size in dwarfsck output.

• (feat) Library: return a dir_entry_view from readdir and find. This is more consistent, but was previously not easily possible due to the lack of a "self" dir entry in the metadata. The "self" entry has been added and will only impact the size of the metadata if directories metadata is not packed.

• (feat) Library: prefer std::string_view over char const*.

• (feat) Library: add directory iterator to directory_view.

• (feat) Library: support for maxiov parameter in readv call.

• (refactor) Lots of internal refactoring to improve overall code quality.

Version 0.10.2 - 2024-12-02

• (fix) Gracefully handle localized error message on Windows. These error messages can contain characters from a Windows (non-UTF-8) code page, which could cause a fatal error in fmt::print in the logging code. Call sites that log such error messages now try to convert these from the code page to UTF-8 or, if that fails, simply replace all characters that are invalid from a UTF-8 point-of-view. Partial fix for github #241.

• (fix) Handle invalid wide chars in file names on Windows. For some reason, Windows allows invalid UTF-16 characters in file names. Try to handle these gracefully when converting to UTF-8. Partial fix for github #241.

• (fix) Workaround for new boost versions which have a process component.

• (fix) Workaround for a deprecated boost header.

• (fix) Support for upcoming Boost 1.87.0. io_service was deprecated and replaced by io_context in 1.66.0. The upcoming Boost 1.87.0 will remove the deprecated API. (Thanks to Michael Cho for the fix.)

• (fix) Disable extended output algorithms (shake(128|256)).

• (fix) Install libraries to CMAKE_INSTALL_LIBDIR. Fixes github #240.

• (fix) mode/uid/gid checks were expecting 16-bit types.

• (fix) stricter metadata checks and improved error messages.

• (fix) Various fixes for filesystem_extractor to prevent memory leaks, correctly handle errors during extraction, and prevent creation of invalid archive outputs due to padding.

• (fix) Various minor fixes: non-virtual dtors, missing includes, std::move vs. std::forward, unused code removal.

• (test) More test cases for stricter metadata checks. Also enable the strict checks in in unit tests by default.

• (docs) Fix typos in man pages.

Version 0.10.1 - 2024-08-17

• (fix) Allow building utils_test against a non-compatible, system-installed version of gtest. This is a common issue when trying to integrate dwarfs into a package manager, as these generally disallow fetching external dependencies at build time.

• (fix) dwarfsck was always reporting a block size of 1 byte rather than the actual block size of the image.

• (fix) DWARFS_HAVE_LIBBROTLI was not set correctly in the config file, causing build errors if the library was built without brotli.

• (fix) Several small fixes for building with Homebrew.

Version 0.10.0 - 2024-08-14

• (fix) Fixed a race condition identified by ThreadSanitizer in the root node name processing.

• (fix) The terminal abstraction code did not check any errors when trying to determine the terminal width, leading to a random terminal width value. This caused the manual page tests to occasionally crash.

• (feat) Two sets of universal binaries and binary tarballs are provided for Linux platforms: one without any debug symbols, the other with minimal debug symbols and support for stack traces. For the universal binary, only the version without debug symbols will be UPX-compressed, as the stack trace functionality doesn't work with a compressed binary.

• (feat) Symbolic links to the universal binary may now be suffixed with a version (i.e. any part of the name starting with - and followed by a digit will be ignored, e.g. the symlink could be mkdwarfs-0.10 and it would be treated as mkdwarfs).

• (feat) Introduced support for extended attributes on Windows, including a new utility for cross-platform xattr manipulation (pxattr, for portable xattr).

• (feat) Enhanced file system API, adding error-code based and exception-safe versions for getattr, access, and similar functions.

• (feat) Filter rules now consistently use Unix path separators, even for the root path component. Addresses a comment in github discussion #228.

• (refactor) Extensive refactoring to improve code modularity, maintainability and to provide proper libraries. The library code has been moved to different namespaces to make it easier to understand the role of different components (e.g. reader, writer, extractor).

• (refactor) Replaced all folly library dependencies in the public DwarFS library interface with alternatives from libraries like e.g. boost or nlohmann::json which are more broadly available. folly and fbthrift are still used as implementation details, but no longer leak into the public library interfaces.

• (refactor) A much smaller subset of folly is now used in DwarFS and only the necessary components are built, significantly reducing the number of compilation units when building DwarFS.

• (build) It is now possible to do modular builds in addition to the default monolithic build, i.e. you can build and install just the DwarFS libraries and later build/install the tools (mkdwarfs, ...) and/or the FUSE driver against these libraries. This is particularly useful for packaging (e.g. in Homebrew, which has removed all FUSE support from the core formulae).

• (build) Shared library builds are now explicitly supported. This fixes issues such as github #184.

• (build) The source tarball now contains all auto-generated code, e.g. manual pages or generated thrift code. This reduces the number of build-time dependencies (e.g. ronn or mistletoe are no longer required) and significantly reduces the build steps (it is no longer necessary to build the thrift compiler). The build is now roughly twice as fast as in the 0.9.x releases.

• (build) The parallel-hashmap, xxHash and zstd submodules have been removed from the git repo and are no longer added to the source tarball. Both xxHash and zstd are now widely available. If a suitable version of parallel-hashmap is found on the system, this will be used, otherwise it will be fetched during the build. Being a header-only library and only used internally, there's no need for it to be installed.

• (build) A lot of GCC warnings have been fixed and upstreamed to folly / fbthrift.

• (test) Fixed some flaky tests, e.g. unmounting the FUSE driver on macOS, or the manpage test that used to crash occasionally.

Version 0.9.10 - 2024-05-30

• (fix) When cloning LZMA compressor objects, the LZMA filter options of the cloned instance would still point to an options object in the original instance. This could lead to LZMA errors when initializing a new compressor. Fixes github #224.

• (fix) Fetch range-v3 if no suitable version is found. Fixes github #221.

• (fix) Filter rules did not work correctly when input is root dir

• (fix) duf reports odd sizes due to using bsize instead of frsize

Version 0.9.9 - 2024-04-30

• (fix) A bug introduced by an optimization to skip hashing of large files if they already differ in the first 4 KiB could, under rare circumstances, lead to an unexpected "inode has no file" exception after the scanning phase. This bug did not cause any file system inconsistency issues; mkdwarfs either crashes with the exception, or its output will be correct. Fixes github #217.

• (feat) Add sequential access detector and block prefetching to the block cache. This improves sequential read throughput roughly by a factor of two. Can be configured / disabled using -o seq_detector.

• (feat) Add tracing support in FUSE driver and dwarfsextract, which allows simple performance analysis using chrome://tracing. Traces can be enabled using -o perfmon_trace and --perfmon-trace.

• (feat) Add performance monitoring and tracing support for the block cache.

• (perf) Significantly improve speed of dwarfsck --checksum.

Version 0.9.8 - 2024-04-14

• (fix) Build custom version of libcrypto to link with the release binaries in order for them to run properly on FIPS-enabled setups. Fixes github #210.

• (fix) When mounting a DwarFS image on macOS and viewing the volume in Finder, only the directories were shown, but no files. The root cause was that a non-existent extended attribute is reported via a different error code in macOS (ENOATTR) compared to Linux (ENODATA) and the wrong error code was returned for certain Finder-related attributes. Fixes github #211.

• (fix) macOS builds using jemalloc were crashing when calling mallctl("version", ...). The root cause of the crash is still unclear, but as a workaround, the jemalloc version is compiled in from a preprocessor constant rather than using mallctl.

Version 0.9.7 - 2024-04-10

• (fix) Handle root uid correctly in access() implementation. Fixes github #204.

• (feature) Show and track library dependencies. Dependencies will be displayed in the command line help; they will also be tracked in the history metadata of a DwarFS image. See also github #207.

• (doc) Describe nilsimsa ordering algorithm more accurately.

• (perf) Reorder branches to improve ricepp speed with real world data.

• (perf) Some tweaks to improve segmenter speed.

Version 0.9.6 - 2024-02-24

• (fix) Add workaround for new glog release breaking folly build. Fixes github #201.

• (perf) Improve ricepp decoding speed by about 25% on x86 and arm and up to 100% on Windows. Also improve encoding speed on Windows by 25%. No more need for special Clang build.

Version 0.9.5 - 2024-02-13

• (fix) Windows path handling was wrong and didn't work properly for e.g. network shares. This is hopefully fixed for all tools now.

Version 0.9.4 - 2024-02-12

• (fix) Prevent installation of ricepp headers/libs. Fixes github #195.

• (fix) Don't fetch googletest in ricepp build if the targets are already available. Fixes github #194.

• (feature) Added blocksize option to the FUSE driver, which allows the st_blksize value to be configured for the mounted file system. Increasing this value can improve throughput for large files.

• (feature) Added experimental readahead option to the FUSE driver. This can potentially increase throughput when performing sequential reads.

Version 0.9.3 - 2024-02-11

• (fix) v0.8.0 removed the implementation of the null decompressor under the assumption that it was no longer used; it was, however, still used when recompressing an image with null-compressed blocks. The change to remove the implementation was reverted and a new test case was added. Fixes github #193.

• (perf) Some more ricepp compression speed improvements. Also, the universal binaries for x86_64 now automatically choose a ricepp version based on CPU capabilities.

Version 0.9.2 - 2024-02-09

• (fix) v0.9.0 introduced an optimization where large files of equal size were only fully hashed for deduplication if the first 4K of their contents also produced the same hash. This introduced a bug causing an exception to be thrown when processing large hard-linked files. The root cause was that the data structure intended to be used for exactly this case was just never populated, and the fix was adding a single line to fill the data structure. The test cases didn't cover large hard-linked files, so this slipped through into the release. A new test case has been added as well.

• (fix) On Windows, when using Power Shell, the error message dialog for a missing WinFsp DLL was not shown when running dwarfs.exe. The workaround is to use the same delayed loading mechanism that's already used for the universal binary and show the error in the terminal. See also the discussion on github #192.

• (feature) Added a --list option to dwarfsck. This lists all files in the files system image. When used with --verbose, the list also shows permissions, size, uid/git and symbolic link information. Fixes github #192.

• (feature) Added a --checksum option to dwarfsck. This produces output similar to the *sum programs from coreutils and can be used to check the contents of a DwarFS image against local files.

Version 0.9.1 - 2024-02-06

• (fix) Invalid UTF-8 characters in file paths would crash mkdwarfs if these paths were displayed in the progress output. A possible workaround was to disable progress output. This fix replaces any invalid characters before displaying them. Fixes github #191.

• (fix) The CMakeLists.txt would bail out as soon as it discovered --as-needed in the linker flags. However, --as-needed is only a problem when combined with BUILD_SHARED_LIBS=ON. The check has been changed to only trigger if both conditions are met.

• (perf) Minor speed improvements in ricepp compression.

Version 0.9.0 - 2024-02-05

• (feature) Experimental macOS support. Fixes github #132.

• (feature) New ricepp compression algorithm for raw images as well as a categorizer for the FITS image format. This is quite limited at the moment, as only two-dimensional, 16-bit integer FITS is supported. However, this covers the majority of astro camera images, which is the primary use case at the moment. This can likely be extended to other raw image formats in the future.

Version 0.8.0 - 2024-01-22

• (fix) Allow version override for nixpkgs. Fixes github #155.

• (fix) Resize progress bar when terminal size changes. Fixes github #159.

• (fix) Add Extended Attributes section to README. Fixes github #160.

• (fix) Support 32-bit uid/gid/mode. Also support more than 65536 uids/gids/modes in a filesystem image. Fixes gh #173.

• (fix) Add workaround for broken utf8cpp release. Fixes github #182.

• (fix) Don't call check_section() in filesystem ctor, as it renders the section index useless. Also add regression test to ensure this won't be accidentally reintroduced. Fixes github #183.

• (fix) Ensure timely exit in progress dtor. This could occasionally block command line tools for a few seconds before exiting.

• (fix) --set-owner and --set-group did not work properly with non-zero ids. There were two distinct issues: (1) when building a DwarFS image with --set-owner and/or --set-group, the single uid/gid was stored in place of the index and the respective lookup vectors were left empty and (2) when reading such a DwarFS image, the uid/gid was always set to zero. The issue with (1) is not only that it's a special case, but it also wastes metadata space by repeatedly storing a potentially wide integer value. This fix addresses both issues. The uid/gid information is now stored more efficiently and, when reading an image using the old representation, the correct uid/gid will be reported. Unit tests were added to ensure both old and new formats are read correctly.

• (fix) mkdwarfs is now much better at handling inaccessible or vanishing files. In particular on Windows, where a successful access() call doesn't necessarily mean it'll be possible to open a file, this will make it possible to create a DwarFS file system from hierarchies containing inaccessible files. On other platforms, this means mkdwarfs can now handle files that are vanishing while the file system is being built.

• (fix) mkdwarfs progress updates are now "atomic", i.e. one update is always written with a single system call. This didn't make much of a difference on Linux, but the notoriously slow Windows terminal, along with somewhat interesting thread scheduling, would sometimes make the updates look like a typewriter in slow-motion.

• (fix) utf8_truncate() didn't handle zero-width characters properly. This could cause issues when truncating certain UTF8 strings.

• (fix) A race condition in simple progress mode was fixed.

• (fix) A race condition in filesystem_writer was fixed.

• (fix) The --no-create-timestamp option in mkdwarfs was always enabled and thus useless.

• (fix) Common options (like --log-level) were inconsistent between tools.

• (fix) Progress was incorrect when mkdwarfs was copying sections with --recompress.

• (fix) Treat NTFS junctions like directories.

• (fix) Fix canonical path on Windows when accessing mounted DwarFS image.

• (fix) Fix slow sorting in file_scanner due to path comparison.

• (fix) On Windows, don't crash with an assertion if the input path for mkdwarfs is not found.

• (remove) Python scripting support has been completely removed.

• (feature) Categorizer framework. Initially supported categorizers are pcmaudio (detect audio data & metadata and provide context for FLAC compressor) and incompressible (detects "incompressible" data). Enabled using the --categorize option.

• (feature) Multiple segmenters can now run in parallel and write to the same filesystem image in a fully deterministic way. Currently, a segmenter instance will be used per category/subcategory. This can makes segmenting multi-threaded in cases where there are multiple categories. The number of segmenter worker threads can be configured using --num-segmenter-workers.

• (feature) The segmenter now supports different "granularities". The granularity is determined by the categorizer. For example, when segmenting the audio data in a 16-bit stereo PCM file, the granularity is 4 (bytes). This ensures that the segmenter will only produce chunks that start/end on a sample boundary.

• (feature) The segmenter now also features simple "repeating sequence detection". Under certain conditions, these sequences could cause the segmenter to slow down dramatically. See github #161 for details.

• (feature) FLAC compression. This can only be used along with the pcmaudio categorizer. Due to the way data is spread across different blocks, both FLAC compression and decompression can likely make use of multiple CPU cores for large audio files, meaning that loading a .wav file from a DwarFS image using FLAC compression will likely be much faster than loading the same data from a single FLAC file.

• (feature) Completely new similarity ordering implementation that supports multi-threaded and fully deterministic nilsimsa ordering. Also, nilsimsa options are now ever so slightly more user friendly.

• (feature) The --recompress feature of mkdwarfs has been largely rewritten. It now ensures the input filesystem is checked before an attempt is made to recompress it. Decompression is now using multiple threads. Also, recompression can be applied only to a subset of categories and compression options can be selected per category.

• (feature) mkdwarfs now stores a history block in the output image by default. The history block contains information about the version of mkdwarfs, all command line arguments, and a time stamp. A new history entry will be added whenever the image is altered (i.e. by using --recompress). The history can be displayed using dwarfsck. History timestamps can be disabled using --no-history-timestamps for bit-identical images. History creation can also be completely disabled using --no-history.

• (feature) All tools now come with built-in manual pages. This is valuable especially on Windows, which doesn't have man at all, or for the universal binaries, which are usually not installed alongside the manual pages. Running each tool with --man will show the manual page for the tool, using the configured pager. On Windows, if less.exe is in the PATH, it'll also be used as a pager.

• (feature) New verbose logging level (between info and debug).

• (feature) Logging now properly supports multi-line strings.

• (feature) Show compression library versions as part of the --help output. For dwarfsextract, also show libarchive version.

• (feature) --set-time now supports time strings in different formats (e.g. 20240101T0530).

• (feature) mkdwarfs can now write the filesystem image to stdout, making it possible to directly stream the output image to e.g. netcat.

• (feature) Progress display for mkdwarfs has been completely overhauled. Different components (e.g. hashing, categorization, segmenting, ...) can now display their own progress in addition to a "global" progress.

• (feature) mkdwarfs now supports ordering by "reverse path" with --order=revpath. This is like path ordering, but with the path components reversed (i.e. foo/bar/baz.xyz will be ordered as if it were baz.xyz/bar/foo).

• (feature) It is now possible to configure larger bloom filters in mkdwarfs.

• (feature) The mkdwarfs segmenter can now be fully disabled using -W 0.

• (feature) mkdwarfs now adds "feature sets" to the filesystem metadata. These can be used to introduce now features without necessarily breaking compatibility with older tools. As long as a filesystem image doesn't actively use the new features, it can still be read by old tools. Addresses github #158.

• (feature) dwarfsck has a new --quiet option that will only report errors.

• (feature) dwarfsck with --print-header will exit with a special exit code (2) if the image has no header. In all other cases, the exit code will be 0 (no error) or 1 (error).

• (feature) The --json option of dwarfsck now outputs filesystem information in JSON format.

• (feature) dwarfsck has a new --no-check option that skips checking all block hashes. This is useful for quickly accessing filesystem information.

• (feature) The FUSE driver exposes a new dwarfs.inodeinfo xattr on Linux that contains a JSON object with information about the inode, e.g. a list of chunks and associated categories.

• (feature) Don't enable readlink in the FUSE driver if filesystem has no symlinks. This is mainly useful for Windows where symlink support increases the number of getattr calls issued by WinFsp.

• (feature) As an experimental feature, CPU affinity for each worker group can be configured via the DWARFS_WORKER_GROUP_AFFINITY environment variable. This works for all tools, but is really only useful if you have different types of cores (e.g. performance and efficiency cores) and would like to e.g. always run the segmenter on a performance core.

• (doc) Add mkdwarfs sequence diagram.

• (doc) Document known issues with WinFsp.

• (doc) Update README with extended attributes information.

• (doc) Add script to check if all options are documented in manpage.

• (build) Factor out repetitive thrift library code in CMakeLists.txt.

• (build) Use FetchContent for both fmt and googletest.

• (build) Use mold for linking when available.

• (build) The CI workflow now uploads coverage information to codecov.io with every commit.

• (test) A ton of tests were added (from 4 kLOC to more than 10 kLOC) and, unsurprisingly, a number of bugs were found in the process.

• (test) Introduced I/O abstraction layer for all *_main() functions. This allows testing of almost all tool functionality without the need to start the tool as a subprocess. It also allows to inject errors more easily, and change properties such as the terminal size.

• (other) The universal binaries are now compressed with a different upx compression level, making them slightly bigger, but decompress much faster.

Version 0.7.5 - 2024-01-16

• (fix) Fix crash in the FUSE driver on Windows when tools like Notepad++ try to access a file like a directory (presumably because this works in cases where the file is an archive). This is a Windows-only issue because the Linux FUSE driver uses the inode-based API, whereas the Windows driver uses the string-based API. While parsing a path in the string-based API, there was no check whether a path component was a directory before trying to descend further.

Version 0.7.4 - 2023-12-28

• (fix) Fix regression that broke section index optimization introduced in v0.7.3. Fixes github #183.

• (fix) Add workaround for broken utf8cpp release. Fixes github #182.

Version 0.7.3 - 2023-12-05

• (feature) Support forward-compatibility. Fixes github #158.

Version 0.7.2 - 2023-07-24

• (fix) Fix locale fallback if user-default locale cannot be set. Fixes github #156.

Version 0.7.1 - 2023-07-20

• (fix) Fix potential division by zero crash in speedometer.

• (other) New tool header.

• (other) Source code cleanups.

• (other) Updated static build procedure (see README).

Version 0.7.0 - 2023-07-11

• (fix) FUSE/WinFsp driver now handles Unicode characters in the file system image name (the file system itself would already work properly with Unicode file names).

• (fix) Fixed heap-use-after-free when using a file system image built with brotli compression. This was caught last minute by ASAN.

• (fix) Catch errors from locale-setting at startup. These errors will only be reported now, but will no longer cause the program to abort.

• (feature) mkdwarfs command-line options have been reorganized into groups to make them easier to find and to make the default help message less intimidating. The full help can now be accessed using -H or --long-help.

• (feature) Symbolic links to the universal binary now also work as aliases on Windows.

• (test) Test universal binary in both --tool and symlink modes.

• (other) CI pipeline tweaks & fixes.

Version 0.7.0-RC6 - 2023-07-09

• (feature) Support delayed loading of WinFsp DLL for universal binary. This makes the mkdwarfs, dwarfsck and dwarfsextract tools of the universal binary usable without the WinFsp DLL.

• (perf) Optimized the offset cache to improve random read latency as well as sequential read latency. This gave a 100x higher throughput for a case where DwarFS was used to compress raw file system images. Fixes github #142.

• (fix) Fix building with make instead of ninja. Also fix builing in Debug mode. Fixes github #146.

• (fix) Fix ninja clean.

• (fix) Fix symlink creation for mount.dwarfs/mount.dwarfs2.

• (other) Added CI pipeline.

• (other) Don't write versioning files to source tree.

Version 0.7.0-RC5 - 2023-07-04

• (feature) Windows support. All tools can now be built and run on Windows, including the FUSE driver, which makes use of WinFsp.

• (feature) Build a "universal" binary that combines mkdwarfs, dwarfsck, dwarfsextract and dwarfs in a single binary. This binary can be used either through symbolic links with the proper names of the tool, or by passing --tool=<name> as the first argument on the command line.

• (feature) Bypass the block cache for uncompressed blocks. This saves copying block data to memory unnecessarily and allows us to keep all uncompressed blocks accessible directly through the memory mapping. Partially addresses github #139.

• (feature) Show throughput in the scanning and segmenting phases in mkdwarfs.

• (feature) Show how much of a file has been consumed in the segmenting phase. Useful primarily for large files.

• (feature) dwarfs and dwarfsextract now have options to enable performance monitoring. This can give insight into the latency of various file system operations.

• (feature) Added inode offset cache, which improves read() latency for very fragmented files.

• (fix) Use folly::hardware_concurrency(). Fixes github #130.

• (fix) Handle ARCHIVE_FAILED status from libarchive, which could be triggered by trying to write long path names to old archive formats.

• (fix) Properly handle unicode path truncation.

• (doc) Update file system format documentation to cover headers and section indices.

• (test) Lots of new tools tests.

• (test) Remove dependency on tar and diff binaries.

• (other) Switch to C++20.

Version 0.7.0-RC4 - 2022-12-24

• (feature) Add --compress-niceness option to mkdwarfs.

Version 0.7.0-RC3 - 2022-11-20

• (fix) Fix heap-use-after-free in dwarfsextract.

• (fix) Fix dwarfs benchmark binary.

• (feature) Add --stdout-progress option to dwarfsextract. Fixes github #117.

• (test) Reduce amount of test data to speed up compiles and avoid timeouts on travis.

Version 0.7.0-RC2 - 2022-11-17

• (fix) Fix linking against compression libs. Fixes github #112.

• (fix) Default FUSE driver debuglevel to warn in background mode. Fixes github #113.

• (feature) Add --chmod option. Fixes github #7.

• (feature) Add unreadable files as empty files. Fixes github #40.

• (doc) Document how to produce bit-identical images

• (doc) Update internal operation section of mkdwarfs manpage

• (doc) Add more documentation details for --file-hash option

• (test) Test image reproducibility for path and similarity ordering

Version 0.7.0-RC1 - 2022-11-08

• (fix) Fixed extract_block.py, which was incorrectly using printf instead of print.

• (fix) Support LZ4 compression levels above 9.

• (feature) Added --filter option to support simple (rsync-like) filter rules. This was driven by a discussion on github #6.

• (feature) Added --input-list option to support reading a list of input files from a file or stdin. At least partially fixes github #6.

• (feature) The compression code has been made more modular. This should make it much easier to add support for more compression algorithms in the future.

• (feature) Added support for Brotli compression. This is generally much slower at compression than ZSTD or LZMA, but faster than LZMA, while offering a compression ratio better than ZSTD. Fixes github #76.

• (feature) Added support for choosing the file hashing algorithm using the --file-hash option. This allows you to pick a secure hash instead of the default XXH3. Also fixes github #92.

• (feature) Improved de-duplication algorithm to only hash files with the same size. File hashing is delayed until at least one more file with the same size is discovered. This happens automatically and should improve scanning speed, especially on slow file systems.

• (feature) Added --max-similarity-size option to prevent similarity hashing of huge files. This saves scanning time, especially on slow file systems, while it shouldn't affect compression ratio too much.

• (feature) Honour user locale when formatting numbers.

• (feature) Added --num-scanner-workers option.

• (feature) Added support for extracting corrupted file systems with dwarfsextract. This is enabled using the --continue-on-error and, if really needed, --disable-integrity-check options. Fixes github #51.

• (test) Added unit tests for progress class.

• (other) Lots of internal cleanups.

Version 0.6.2 - 2022-10-24

• (fix) Fix github #91: image creation reproducibility. Add --no-create-timestamp option, produce deterministic inode numbers and fix fsst bug that causes symbol tables to be non-deterministic. Images built while omitting create timestamps will now be bit-identical.

• (fix) Fix github #93: only overwrite existing output file when --force option given on command line.

• (fix) Fix github #104: extracting large files was causing dwarfsextract to OOM. This was fixed by extracting large files in chunks rather than all at once.

• (fix) Fix github #105: handle strrchr() return NULL.

• (fix) Fix out-of-bounds access (PR #106).

• (fix) Fix swapped-out cached block detection (PR #107).

• (fix) Fix data race in cached block that was triggered by statistics collection and could cause the process to crash.

• (fix) Fix heap-use-after-free when writing section index.

Version 0.6.1 - 2022-06-11

• (fix) Fix binary installation

Version 0.6.0 - 2022-06-11

• (fix) Fix and simplify static builds as much as possible. Document how to set up a static build environment. This also fixes github #75 and github #54. Huge shoutout to Maxim Samsonov for implementing most of this!

• (fix) Fix github #71: driver hangs when unmounting

• (fix) Fix github #67: dwarfs I/O hangs if call to to fuse_reply_iov fails

• (fix) Fix github #86: block size bits config issues

• (fix) Various build fixes.

• (feature) Add support for cache tidying, which releases cache memory when the mounted file system is unused.

• (feature) Section index support for speeding up mount times (fixes github #48).

Version 0.5.6 - 2021-07-03

• (fix) Build fixes for gcc-11

• (fix) Use REALPATH in version.cmake to fix building in symbolically linked repositories (fixes github #47).

Version 0.5.5 - 2021-05-03

• (feature) If a filesystem block cannot be compressed to less than the uncompressed size, it will be stored uncompressed. This feature actually fixes the bug described below.

• (fix) When building a filesystem from high entropy input data (e.g. already compressed files), and when using LZMA compression with block sizes >= 25, the LZMA algorithm could be unable to pack a block into the worst-case allocated size. This behaviour was not expected and crashed mkdwarfs, and seems to me like a bug in LZMA's lzma_stream_buffer_bound() function. The issue has been fixed by not compressing blocks at all if the compressed size matches or exceeds the uncompressed size. This fixes part of github #45.

• (fix) Filesystems created such that after segmenting the total data size was a multiple of the block size (i.e. the last block was completely filled) had the last block written to the image twice. Such a filesystem image is perfectly usable, but the repeated block uses space unnecessarily. This is highly unlikely to happen with real data.

• (fix) Filesystems created with -P shared_files, but no shared files in the source tree, were created correctly, but could not be loaded. This has been fixed and the filesystems can now be loaded correctly.

• (test) Add tests for binaries and FUSE driver.

• (other) Minor code cleanups.

Version 0.5.4 - 2021-04-11

• (fix) FUSE driver hangs when accessing files and the driver is not started in foreground or debug mode. This bug is present in both the 0.5.2 and 0.5.3 releases. Fixes github #44.

Version 0.5.3 - 2021-04-11

• (fix) Add PREFER_SYSTEM_GTEST for distributions (like Gentoo) that have a gtest package.

• (fix) Make sure the source tarball can be built inside a git repo. The version file generation code would attempt to pull information from any outside git repository without checking if it's actually the DwarFS repo.

Version 0.5.2 - 2021-04-07

• (fix) Make FUSE driver exit with non-zero exit code if filesystem cannot be mounted. Fixes github #41.

Version 0.5.1 - 2021-04-06

• (fix) fsst library was built with -march=native, which caused the static binaries not to work on non-AVX platforms. The fsst library is now being built with no extra flags.

Version 0.5.0 - 2021-04-05

• (fix) Disable multiversioning on non-x86 platforms, which broke the ARM build.

• (fix) Due to a bug in the bloom filter code, only half of each 64-bit block in the bloom filter was utilized, which reduced the efficiency of the filter. The bug was spotted thanks to ubsan. With the fixed filter being twice as effective, the default size of the bloom filter has now been halved.

• (fix) When exporting metadata using --export-metadata, dwarfsck was not truncating the output file, which could lead to a corrupt metadata export.

• (perf) Scanning has been significantly optimized and is now up to three times faster on average.

• (perf) Digest computation has been parallelized in both mkdwarfs and dwarfsck giving better performance on multi-core systems.

• (perf) A set of micro-benchmarks has been added to evaluate the performance of different filesystem operations. This can be build by enabling the -DWITH_BENCHMARKS=1 cmake option.

• (perf) Zstd contexts are now reused during compression, which seems to give some minor speedup.

• (feature) New metadata format (v2.3). This includes a number of changes:

  • Correct hardlink preservation. With older metadata formats, all duplicate files would appear hardlinked. The new format preserves hardlinked files exactly as present in the input data, and performs additional deduplication at a lower level.

  • The new format offers a lot of customization for additional packing of metadata. You can use these to trade off metadata size, mounting speed, etc. Especially for filesystems with millions of files, the metadata size can be reduced significantly.

  • In particular, filename and symlink data can be stored in a format that reduces the size by roughly a factor of two, but still allows for random access, so the compressed data can be mapped into memory and decompressed on the fly.

• (feature) DwarFS now directly supports images using a custom header. The header can be completely arbitrary. mkdwarfs can write, replace or remove such headers, and all other tools can either skip to a specified offset, or determine this offset automatically. This fixes github #38.

• (feature) dwarfsck has been improved to perform extensive metadata checks. Also, checksumming is now done in a thread pool, which significantly speeds up dwarfsck for large file systems.

• (feature) dwarfsck now shows a detailed breakdown of metadata memory usage, which can be used to optimize metadata packing options.

• (feature) Added ENABLE_COVERAGE cmake option.

• (test) Compatibility testing with older filesystem versions has been improved.

• (test) A new test suite has been added to check detection of corrupted DwarFS images.

• (doc) Added some high level internals documentation for mkdwarfs.

• (doc) Documented the filesystem and metadata formats.

• (other) Lots of internal cleanups.

Version 0.4.1 - 2021-03-13

• (fix) Linking against libarchive was fixed so that it also works for shared library builds. (fixes github #36)

• (fix) mkdwarfs didn't catch certain exceptions correctly, which would cause a stack trace instead of a simple error message. This has been fixed.

• (fix) The statically linked executables were unable to handle any exceptions at all due to duplicate stack unwinding code. This has (hopefully) been fixed now.

• (perf) GCC builds have traditionally been much slower than Clang builds, though it was unclear why that was the case. It turns out the reason is simply that CMake defaults to -O3 optimization, which is known to cause performance regressions in some cases. The build has been changed to always build with -O2 when doing an optimized GCC build. The Clang build is unaffected. (fixes github #14)

• (perf) The segmenting code now uses a bloom filter to discard unsuccessful matches as early and quickly as possible. While this only gives a minor speedup when using a single lookback block, as you increase the number of lookback blocks speed is barely affected whereas before it would slow down significantly. The bloom filter size (relative to the number of values) can be tuned by using --bloom-filter-size, though increasing it any further from the default is likely not going to make a difference.

• (perf) Nilsimsa similarity computation has been improved to make use of different instruction sets depending on the CPU architecture, speeding up the process of ordering files by similarity by almost a factor of 2.

• (doc) Added comparison with lrzip, zpaq. Updated wimlib comparison.

Version 0.4.0 - 2021-03-06

• (feature) New dwarfsextract tool that allows extracting a file system image. It also allows conversion of the file system image directly into a standard archive format (e.g. tar or cpio). Extracting a DwarFS image can be significantly faster than extracting a equivalent compressed archive.

• (feature) The segmenting algorithm has been completely rewritten and is now much cleaner, uses much less memory, is significantly faster and detects a lot more duplicate segments. At the same time it's easier to configure (just a single window size instead of a list).

• (feature) There's a new option --max-lookback-blocks that allows duplicate segments to be detected across multiple blocks, which can result in significantly better compression when using small file system blocks.

• (compat) The --blockhash-window-sizes and --blockhash-increment-shift options were replaced by --window-size and --window-step, respectively. The new --window-size option takes only a single window size instead of a list.

• (fix) The rewrite of the segmenting algorithm was triggered by a "bug" (github #35) that caused excessive memory consumption in mkdwarfs. It wasn't really a bug, though, more like a bad algorithm that used memory proportional to the file size. This issue has now been fully solved.

• (fix) Scanning of large files would excessively grow mkdwarfs RSS. The memory would have sooner or later be reclaimed by the kernel, but the code now actively releases the memory while scanning.

• (perf) mkdwarfs speed has been significantly improved. The 47 GiB worth of Perl installations can now be turned into a DwarFS image in less then 6 minutes, about 30% faster than with the 0.3.1 release. Using lzma compression, it actually takes less than 4 minutes now, almost twice as fast as 0.3.1.

• (perf) At the same time, compression ratio also significantly improved, mostly due to the new segmenting algorithm. With the 0.3.1 release, using the default configuration, the 47 GiB of Perl installations compressed down to 471.6 MiB. With the 0.4.0 release, this has dropped to 426.5 MiB, a 10% improvement. Using lzma compression (-l9), the size of the resulting image went from 319.5 MiB to 300.9 MiB, about 5% better. More importantly, though, the uncompressed file system size dropped from about 7 GiB to 4 GiB thanks to improved segmenting, which means less blocks need to be decompressed on average when using the file system.

• (build) The project can now be built to use the system installed zstd and xxHash libraries. (fixes github #34)

• (build) The project can now be built without the legacy FUSE driver. (fixes github #32)

• (other) Several small code cleanups.

Version 0.3.1 - 2021-01-07

• (fix) Fix linking of Python libraries

• (fix) Fix missing brace in version generator code

• (fix) Ensure the code builds fine without libdwarf

• (fix) Silence a warning and remove an unused definition

Version 0.3.0 - 2020-12-30

• (fix) File system images created with versions 0.2.2 and before did store symlinks incorrectly. While this was fixed in 0.2.3, old images could still not be read correctly. This has now been fixed and symlinks on all 0.2.x images will work correctly when using the 0.3.0+ FUSE driver.

• (fix) There was no error if the output file could not be written, mkdwarfs would just fail silently. This has now been fixed.

• (fix) When corrupted compressed blocks in either format (LZ4, ZSTD, LZMA) were detected, the FUSE driver would actually show the file contents as all zero bytes instead of signaling an I/O error. This has been fixes and verified for all formats.

• (fix) Better (hopefully) auto-detection of terminal settings to avoid using features like unicode or color when terminals don't support them. Fixes github #20.

• (fix) A number of checks has been added to make sure that corrupt file system images will not crash the binaries. In order for this to be most efficient, old images should be rewritten in the new format using:

  mkdwarfs -i old.dwarfs -o new.dwarfs --recompress none
  

• (compat) The metadata format has changed and new file system images can no longer be read by old FUSE drivers.

• (perf) Lots of tweaks and optimizations have resulted in an even better compression ratio while at the same time taking less time to build file system images. On the 48 GiB Perl dataset, for example, the compression improved from 555.7 MiB in 15m12s with 0.2.3 to 471.6 MiB in 13m59s with 0.3.0.

• (perf) Replace the cyclic hash function with the one used by rsync. The rsync hash produces similar results, but it's faster.

• (perf) mkdwarfs will now make use of hard link and inode data to avoid scanning the same inode multiple times.

• (perf) Segmenting performance has been improved by re-using data structures and thus avoiding extra memory allocations.

• (perf) All binaries now use jemalloc by default, which uses significantly less memory than glibc or tcmalloc, especially in the FUSE driver.

• (feature) New file system image format adds integrity checking as well as features for easier recovery in case of corruption. While currently there is no way to recover a corrupt file system, it is important to have the data in place sooner rather than later.

• (feature) New Python scripting support completely replaces Lua scripting. The new interface offers a lot more options and should be much easier to use.

• (feature) New nilsimsa similarity algorithm. This has become the default, as it's significantly better on my test data than the "simple" similarity algorithm.

• (feature) New option --keep-all-times to keep atime and ctime in addition to just keeping mtime.

• (feature) New option --time-resolution that allows to configure the resolution with which time stamp are stored.

• (feature) Device, FIFO and socket inodes can now be stored in DwarFS file system images. This has to be enabled with the new --with-devices and --with-specials options.

• (feature) The FUSE driver can now optionally expose correct hard link counts.

• (feature) mkdwarfs now has an option --remove-empty-dirs to remove empty directories.

• (feature) The FUSE driver has 4 new options to control caching. no_cache_image will explictly try to release compressed blocks from the file system image back to the kernel after reading. cache_image will keep them in the cache. no_cache_files will cause decompressed files not to be cached by the kernel. cache_files will cause them to be cached. The defaults are no_cache_image and cache_files.

• (feature) The FUSE driver now has a readonly option that will prevent any entries in the mounted file system to show up as writeable. This is not the default, because it interferes with setting up overlays.

• (feature) dwarfsck can now dump metadata as JSON blob.

• (feature) dwarfsck can now also export raw metadata as JSON. The difference to the --json option is that this JSON export could be used to fully reconstruct the metadata for a DwarFS image.

• (feature) More detailed logging and better error handling.

• (test) Added backwards compatibility tests.

• (build) Added zstd as a submodule.

• (build) There is now a binary package with statically linked binaries available.

• (doc) More accurate list of dependencies.

• (doc) Document how to add /etc/fstab entry for DwarFS image.

• (doc) Comparison with wimlib.

• (doc) Comparison with Cromfs.

• (doc) Comparison with EROFS.

• (doc) Updated benchmarks.

Version 0.2.4 - 2020-12-13

• Fix --set-owner and --set-group options, which caused an exception to be thrown at the end of creating a file system. (fixes github #24)

Version 0.2.3 - 2020-12-01

• Fix link handling. There were two bugs introduced with the new metadata format, one in file system creation and another in the fuse driver. You will have to re-create a file system created with dwarfs < 0.2.3 if it contained links. If you can absolutely not re-create the file system and the data is precious, let me know, there's actually a way to recover the missing data. EDIT: There will be a fix available in the 0.3.0 release, so you don't have to rebuild old file systems.

Version 0.2.2 - 2020-11-30

• Remove read-only masking as it prevents writable overlays

• Throw an error in mkdwarfs if unrecognized command line arguments are encountered (github #5)

• Various build fixes (github #2. #3)

• More documentation

Version 0.2.1 - 2020-11-29

• Replace --no-owner and --no-time with more flexible --set-owner, --set-group and --set-time options

• Update man pages

Version 0.2.0 - 2020-11-29

• Complete rewrite of the file system metadata storage using fbthrift's frozen library

Version 0.1.1 - 2020-11-23

• Test and fix Debian Buster and Ubuntu Focal builds

• Migrate from folly::StringPiece to std::string_view

• Documentation updates, list Debian/Ubuntu dependencies

Version 0.1.0 - 2020-11-22

• Initial release