All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

• PR#146 changed windows::MinidumpWriter::dump_crash_context to take crash_context by reference.
• PR#155 refactored a large portion of the code.

• PR#150 fixed undefined behavior in the Linux crash dump writing.

• PR#143
  • turn many errors that are currently treated as critical (and thus prevent minidump generation) into non-critical "soft" errors
  • collect non-critical errors and serialize them into a new JSON stream in the minidump

• PR#145 updated dependencies.

• PR#129 added checking of additions to ensure invalid memory offsets are gracefully handled.
• PR#135 resolved #134 by attempting to read the executables build id from the mapped file if it could not be retrieved from process memory.
• PR#136 changed to an older type to avoid requiring a semi-recent rust version.

• PR#118 resolved #72 by adding support for reading process memory via process_vm_readv and /proc/{pid}/mem, in addition to the original PTRACE_PEEKDATA. This gives significant performance benefits as memory can now be read in blocks of arbitrary size instead of word-by-word with ptrace.
• PR#128 and PR#133 updated the lockfile.

• PR#127 resolved #27 by allowing programs to pass the needed auxv information, still falling back to reading /proc/{pid}/auxv to fill missing information, and being more permissive by still writing a dump if some or all of the auxv information could not be retrieved successfully rather than completely failing to write the minidump.
• PR#131 resolved #124 by reinjecting non-SIGSTOP signals after ptrace::attach so that the thread would be in the correct state after ptrace::detach.

• PR#117 resolved #79 by enabling reading of a module's build id and soname directly from the mapped process rather than relying on file reading, though that is still used as a fallback.

• PR#126 updated minidump-common -> 0.22.

• PR#110 changed it so that SIGCONT is sent regardless if the process was not able to be SIGSTOPed quickly enough.
• PR#113 fixed a segfault(!) on linux if it was compiled with rustc 1.77.0 in release mode.

• PR#108 resolved #28 by sending a SIGSTOP to the process that is about to be dumped to (hopefully) increase the robustness of the dumping process by reducing the chance of errors, particularly with regard to threads. This is done as a best effort, and will perform the old behavior if the process has not stopped within a timeout (by default 100ms), which can be overriden by the user.

• PR#106 bumped minidump-common, minidump, minidump-processor, and minidump-unwind -> 0.21.

• PR#104 slightly tweaked .so version parsing in the case of more "exotic" versions such as libdbus-1.so.3.34.2rc5. Previously this was parsed as 3.34.25 but would cause ambiguity if there was ever an actual .25 patch/age in the future. Now, the last version is parsed as 1-2 numbers, ignoring non-digit characters if the last component has them. If 2 numbers are parsed, the last number is now placed in VS_FIXEDFILEINFO::product_version_lo so that it is distinct from the patch/age component placed in VS_FIXEDFILEINFO::product_version_hi.

• PR#103 added .so file versions as additional metadata to minidumps, resolving this Mozilla bug. There is no true standard for .so file versions, so this is a best effort to pull what version information we can from the .so filename. The version components are major.minor.release similarly to semver, where major -> VS_FIXEDFILEINFO::file_version_hi, major -> VS_FIXEDFILEINFO::file_version_lo, and release -> VS_FIXEDFILEINFO::product_version_hi
  • libmozsandbox.so -> 0.0.0
  • libstdc++.so.6.0.32 -> 6.0.32
  • libcairo-gobject.so.2.11800.0 -> 2.11800.0
  • libm.so.6 -> 6.0.0
  • libabsl_time_zone.so.20220623.0.0 -> 20220623.0.0
  • libdbus-1.so.3.34.2rc5 -> 3.34.25

• PR#97 bumped goblin -> 0.8.
• PR#99 bumped minidump-common -> 0.20, scroll -> 0.12, memmap2 -> 0.9.

• PR#94 added support for writing file information for every file open in the process the dump is being performed for into the MINIDUMP_HANDLE_DATA_STREAM stream.
• PR#90 added support for including the /proc/<pid>/limits file in the MozLinuxLimits stream. This information can be used together with the file information described above to diagnose situations where the process was killed by the kernel due to file handle limits being hit. Thanks @lissyx!

• PR#94 updated several dependencies to align with minidump-common, which was also bumped.

• PR#86 added support for i686-android-linux.

• PR#85 removed the dependency on chrono.
• PR#89 resolved #88 by merging ranges that were mapped, but had 1 or more unmapped ranges in between them.

• PR#87 updated some dependencies.

• PR#70 resolved #8 by adding support for writing MemoryInfoListStream on Linux/Android targets, this allows minidump consumers to process minidumps more easily without needing to parse and understand Linux-specific information. Thanks @afranchuk!
• PR#81 stabilized arm and aarch64 support for unknown-linux and linux-android, as well as adding support for x86_64-linux-android.

• PR#70 replaced the custom reading of procfs information used when generating a minidump on Linux to use the procfs crate instead, removing a bunch of code.
• PR#80 along with PR#84 replaced procfs with procfs-core, removing unneeded dependencies such as windows-sys.

• PR#78 resolved #24 by ignoring guard pages when dumping the stack to the minidump in the event of a stack overflow.
• PR#83 resolved #82 by correctly aligning a structure.

• PR#77 removed the dependency on winapi, all bindings are either part of minidump-writer or crash-context now.

• PR#77 closed #67 by allowing the user to specify the MinidumpType flags when creating a minidump.

• PR#68 resolved #29 by ignoring the bening ESRCH error when detaching pthreads. Thanks @afranchuk!
• PR#74 resolved #73 by ensuring the NT_GNU_BUILD_ID section had the proper correct GNU name before using it as the build identifier.

• PR#65 updated crash-context to 0.5, which has support for a custom capture_context to replace RtlCaptureContext on Windows, due to improper bindings and deficiencies, resolving #63.
• PR#65 replaced most of the custom bindings from PR#60 with bindings from either crash-context or winapi.

• PR#60 removed the dependency on windows-sys due the massive version churn, resolving #58. This should allow projects to more easily integrate this crate into their project without introducing multiple versions of transitive dependencies.
• PR#62 replaced MDExceptionCodeLinux with minidump_common::ExceptionCodeLinux.
• PR#64 updated dependencies.

• PR#53 made the mem_writer and dir_section modules public. Thanks @sage-msft!
• PR#55 bumped nix, minidump-common, minidump, minidump-processor and dump_syms. Thanks @sfackler!
• PR#57 bumped windows-sys to 0.42.

• PR#56 removed the writing of the HandleOperationListStream stream, as it was essentially untested and was of dubious usefulness.

• PR#50 updated minidump-common and crash-context.

• PR#50 resolved #33 by encoding the full exception info in the exception_information field of the exception stream.
• PR#50 resolved #34 by unwrapping EXC_CRASH exceptions to retrieve the wrapped exception.

• PR#47 resolved #46 by handling the special case of dyld.

• PR#42 resolved #41 by capping the VM read of task memory to avoid a syscall failure, as well as made it so that if an error does occur when reading the module's file path, the module is still written to the minidump, as the file path is less important than the UUID in terms of module identification.
• PR#44 resolved #43 by correctly calculating the base address of each loaded module. The bug was inherited from Breakpad.
• PR#44 and PR#45 resolved #37 by making the crash_context::CrashContext optional on MacOS and Windows, to make creating a minidump without necessarily having an actual crash more convenient for users.

• PR#32 resolved #23 by adding support for the thread names stream on MacOS.

• PR#21 added an initial implementation for x86_64-apple-darwin and aarch64-apple-darwin

• Initial release, including basic support for x86_64-unknown-linux-gnu/musl and x86_64-pc-windows-msvc