Debug File Formats

Sentry differentiates four kinds of debug information:

  • Debug Information: Provides function names, paths to source files, line numbers, and inline frames. The process of resolving this information from instruction addresses is called "symbolication". This information is relatively large compared to the executable and usually put into a separate file. In Sentry, these files are designated as debug companions and show the debug tag.

  • Symbol Tables: If debug information is not available for a certain library, Sentry can use symbol tables as a fallback to retrieve function names. Symbol tables are usually included in both the executable and debug companion files. However, they do not contain sufficient information to resolve inline functions or file names and line numbers. The symtab tag indicates symbol tables.

  • Source Code: Conventionally, source code is not part of regular debug information files. Sentry CLI can bundle source code of your application and upload it to display source context in stack traces in Sentry. These bundles show up with the sources tag.

  • Unwind Information: Enables Sentry to extract stack traces from Minidumps and other binary crash formats of optimized builds. This process is referred to as "stack unwinding" or "stack walking". Since this is also required when throwing exceptions in C++, this information is often included in the executable or library. If an uploaded file contains this information, it shows the unwind tag. Note that on some platforms no actual unwinding takes place. For instance, WebAssembly currently does not have the equivalent of minidumps which means we do not require that type of information in such cases.

Compilers place the above debug information in different files passed on the target platform, architecture, build flags or optimization level. Consequently, Sentry might not need all of the above information to process crash reports. Still, it is always a good idea to provide all available debug information.

sentry-cli can be used to list properties of supported debug files and validate their contents. See Debug Information Files in sentry-cli for more information.

Mach-O and dSYM

Executables, dynamic libraries and debug companions on all Apple platforms use the Mach Object, or short Mach-O, container format. This applies to iOS, iPadOS, tvOS, watchOS, and macOS.

  • Executables do not carry a file extension. For desktop applications, they are often placed in app bundle structures with the .app suffix. Unless stripped manually, executables contain unwind information and a symbol table. Debug information is never stored in executables.

  • Dynamic Libraries use the .dylib extension, and otherwise behave exactly the same as executables.

  • Debug Companions are placed in a folder structure with the .dSYM extension, and are located at <name>.dSYM/Contents/Resources/DWARF/<name>. They usually contain a symbol table and debug information, but rarely unwind information.

When building an application with Xcode or with the clang compiler, debug information is automatically placed in a dSYM file. When linking manually, however, the dSYM file must be created using the following command:

Copied
dsymutil /path/to/output[.dylib]

Executable and Linkable Format (ELF)

On Linux distributions, executables and debug information are stored in ELF containers. Unlike other platforms, there is no dedicated container or specifier for debug companion files.

Debug information is part of the binary (executable or library) and stripped when generating release builds due to their size. However, there is a way to retain them in a separate file (either in a different location or with .debug extension):

Copied
# There is an executable called "binary" in the CWD
objcopy --only-keep-debug binary binary.debug
objcopy --strip-debug --strip-unneeded binary
objcopy --add-gnu-debuglink=binary.debug binary

This results in the following structure:

  • Executables do not carry a file extension. If stripped like above, executables contain a symbol table, but no debug information. If the build run omits frame pointers, unwind information will also be retained. Both can be further stripped using flags like --strip-all.

  • Shared Libraries use the .so extension, and otherwise behave exactly the same as executables.

  • Debug Companions do not carry a standard file extension, but are often named .debug. If stripped like above, these files contain unwind information, debug information and a symbol table.

Shared libraries installed via package managers usually provide their debugging information in separate *-dev packages and put it in locations like /usr/local/debug/.... To receive symbolicated stack traces from those libraries, make sure to also upload their symbols in addition to your app’s symbols.

ELF supports the compression of debug information which can significantly reduce the time required to upload debug information files to Sentry and thus improve build times. gcc (version 5 or newer) and clang (version 5 or newer) support this by passing the -gz flag to both the compiler and linker. The common way to compress debug information, however, is when stripping the executable:

Copied
# Note the --compress-debug-sections option
objcopy --only-keep-debug --compress-debug-sections=zlib binary binary.debug

This can be verified by checking for the C flag in readelf, corresponding to SHF_COMPRESSED:

Copied
readelf -S path/to/file
  ...
  [21] .debug_info       PROGBITS         0000000000000000  00000370
       000000000000e133  0000000000000000   C       0     0     1

PE and PDB

On Microsoft Windows, executables and dynamic libraries use the Portable Executable container, in short PE. Debug information is stored in Program Database files, commonly referred to as PDB.

  • Executables use an .exe file extension. Only when compiled for 64-bit architectures, they contain unwind information. Otherwise, they do not contain any usable information and will be omitted for the upload to Sentry.

  • Dynamic Libraries use the .dll file extension, and otherwise behave exactly the same as executables.

  • Debug Companions are stored in .pdb files. They usually contain debug information and in most cases symbol tables. For 32-bit programs, they also contain unwind information. In rare cases, they might have different file names than their corresponding executable.

Breakpad Symbols

The Google Breakpad library has established a platform-independent ASCII format to store debug information. Such files are usually generated for applications using Breakpad, Crashpad, or the Electron Framework.

The Breakpad repository includes dump_syms tools for each platform that can convert from native debug files to Breakpad symbols. These converters bundle all available information into one file, such that only one file needs to be uploaded.

In contrast to native debug files, Breakpad symbols discard a lot of information that is not required to process minidumps. Most notably, inline functions are not declared, such that Sentry is not able to display inline frames in stack traces.

WASM

For WebAssembly, we support DWARF in WASM containers. Note that we do not support source maps, which are also a format used for WASM debugging, but have shortcomings that make them impractical for a crash reporting tool like Sentry.

Since WASM does not specify debug/build IDs yet, we provide a separate tool to add build IDs and split files called wasm-split to help you create a debug companion file ready for uploading to Sentry while removing all debug information from the release binary.

ProGuard Mappings

ProGuard mapping files allow Sentry to resolve obfuscated Java classpaths and method names into their original form. In that sense, they act as debug information files for Java and Android applications.