|
1 | 1 | # Change Log |
2 | 2 |
|
| 3 | +## Version 0.15.2 - 2026-03-31 |
| 4 | + |
| 5 | +- (fix) The image size was not correctly passed to an instance of the |
| 6 | + file system parser, which resulted in errors when trying to load an image |
| 7 | + embedded in a larger (e.g., multi-layer) file. Thanks to Ruan Formigoni |
| 8 | + for the pull request that fixed this issue. |
| 9 | + |
| 10 | +- (fix) The performance monitor timer for `op_lseek` in the FUSE driver was |
| 11 | + not set up correctly, which could lead to segmentation faults or bus errors |
| 12 | + due to an uninitialized index into a `std::deque`. This has been fixed, and |
| 13 | + an additional check has been added to make sure these kinds of errors are |
| 14 | + caught in the future. |
| 15 | + |
| 16 | +- (fix) A few test failures have popped up in openSUSE Tumbleweed native |
| 17 | + 32-bit builds. One was caused by glibc requiring `_FILE_OFFSET_BITS=64` for |
| 18 | + 64-bit file operations. This caused some sparse file tests that created |
| 19 | + files larger than 4 GiB to fail. This was never needed in musl-libc builds, |
| 20 | + so the problem was not caught before. The other failures were mostly |
| 21 | + cosmetic and only showed up in GCC builds. The affected tests were checking |
| 22 | + the formatting code for times and ratios, which relied heavily on |
| 23 | + floating-point arithmetic and was not deterministic across platforms and |
| 24 | + compilers. The code has been completely rewritten to use integer arithmetic |
| 25 | + and avoid any platform-specific behavior. The final issue was a bug in the |
| 26 | + test code that, interestingly, also surfaced only with GCC. Addresses |
| 27 | + GitHub issue #354. |
| 28 | + |
| 29 | +- (fix) The FUSE drivers (`dwarfs`, `dwarfs2`) have gone through several |
| 30 | + stages of refactoring to fix a large number of very subtle issues. For |
| 31 | + example, it is now guaranteed that you will see an error for a misspelled |
| 32 | + option rather than an error caused by the absence of the option you were |
| 33 | + trying to use. For example, if you used `-o image_size=1234` instead of |
| 34 | + the correct `-o imagesize=1234`, you would likely get an error saying that |
| 35 | + the file system could not be loaded rather than an error about the unknown |
| 36 | + option. Before the refactoring, the code was a horrible mess of |
| 37 | + preprocessor conditionals, and basically every flavor of the FUSE driver |
| 38 | + (high-level, low-level, Windows, FUSE v2, or FUSE v3) had a different code |
| 39 | + path during startup, exposing a different set of quirks. The code is now |
| 40 | + *much* more readable, a lot more consistent, and even better tested than |
| 41 | + before. |
| 42 | + |
| 43 | +- (fix) The manual pages shown with `--man` for all tools unintentionally |
| 44 | + contained the license header that had been added in XML comments at the top |
| 45 | + of the source files. The renderer has been fixed to ignore those comments. |
| 46 | + |
| 47 | +- (build) After benchmarking the latest `mimalloc` memory allocator, it turns |
| 48 | + out that performance is mostly on par with `jemalloc`. It is still much |
| 49 | + less configurable than `jemalloc`, but it is definitely usable if you do |
| 50 | + not need that configurability. |
| 51 | + |
| 52 | +- (build) The `small` universal release binaries are now built with |
| 53 | + `mimalloc` instead of `jemalloc`, reducing their size by about 10%. |
| 54 | + |
| 55 | +- (build) Old Clang compiler versions, such as those on Ubuntu 22.04, are |
| 56 | + no longer supported because they cannot use libstdc++'s `std::expected` |
| 57 | + implementation. DwarFS can still be built on Ubuntu 22.04 with GCC. |
| 58 | + |
| 59 | +- (test) After profiling test execution, a few tests stood out as being |
| 60 | + particularly slow. These have been modified to run faster while still |
| 61 | + covering most of the same code paths. |
| 62 | + |
3 | 63 | ## Version 0.15.1 - 2026-03-21 |
4 | 64 |
|
5 | 65 | - (fix) `mkdwarfs` did not correctly handle inputs where hardlinks had the |
|
0 commit comments