Gallium
For now, the easiest way to use Gallium is through the examples/call.jl script.
julia> ARGS = ["1234"] # PID of the Julia process you wish to control
julia> include("call.jl")
This will add several REPL modes:
- Global target julia mode (activated via
\
). Evaluate julia code at global scope in the inferior - Host C++ mode (activated via
<
). Evaluate C++ code in the host. Useful to interact with LLDB's C++ API - Target C++ mode (activated via
>
). Evaluate C++ code in the target - LLDB mode (activated via
`
). Run an lldb command
The LLDB command prompt supports all your usual debugger commands (b
, bt
, s
, n
, finish
), as well as the following custom commands:
js
- Step through julia code. Including stepping through indirect dispatchjbt
- Obtain a julia backtracejp
- Run a julia expression, with current local variables in scope (to be integrated into the target julia REPL mode)jobj
- Retrieve a handle to the current frame's JIT object file
Debugging missing local variable and line number coverage
Local variable coverage can be missing at several level: julia, llvm or DWARF. To debug at the julia level, just use code_typed, etc. To debug at the llvm level, the following command sequence may be helpful:
julia> using DIDebug
julia> first(current_thread(Gallium.ctx(dbg))) |> Gallium.getASTForFrame
Target C++> $ans->functionObject
Target C++> GetBitcodeForFunction($ans)
julia> data = Gallium.retrieve(dbg,ans)
julia> DIDebug.parseBitcode(data)
C++> $ans->dump()
To debug at the DWARF, level the following may be helpful:
LLDB> jobj
julia> oh = ans
julia> collect(DWARF.DIETrees(ObjFileBase.debugsections(oh)))
Building Gallium
Gallium requires custom versions of julia, LLVM, Clang, LLDB and libuv. The easiest way to obtain these is to checkout the kf/gallium branch of julia, which will attempt to check out the correct branches and build everything from scratch. Note that this only works on a fresh install.
Alternatively, if you already have a version of llvm-svn checked out, you can manually go in and check out the kf/gallium branch on JuliaLang/{llvm, clang, lldb} and rebuild each of them.
After one of these two methods of building Julia with support for Gallium suceeds, you may need to apply the following patch to Cxx.jl:
diff --git a/src/bootstrap.cpp b/src/bootstrap.cpp
index 01ff792..8dca780 100644
--- a/src/bootstrap.cpp
+++ b/src/bootstrap.cpp
@@ -846,8 +846,8 @@ DLLEXPORT void init_clang_instance(C, const char *Triple) {
Cxx->CI->getLangOpts().Bool = 1;
Cxx->CI->getLangOpts().WChar = 1;
Cxx->CI->getLangOpts().C99 = 1;
- Cxx->CI->getLangOpts().RTTI = 1;
- Cxx->CI->getLangOpts().RTTIData = 1;
+ Cxx->CI->getLangOpts().RTTI = 0;
+ Cxx->CI->getLangOpts().RTTIData = 0;
Cxx->CI->getLangOpts().ImplicitInt = 0;
Cxx->CI->getLangOpts().PICLevel = 2;
Cxx->CI->getLangOpts().Exceptions = 1; // exception handling
Finally, you'll need to run the following series of commands at the julia prompt:
Pkg.clone("https://github.com/Keno/Cxx.jl.git")
Pkg.build("Cxx")
Pkg.add("Reactive")
Pkg.clone("https://github.com/Keno/JITTools.jl.git")
Pkg.clone("https://github.com/Keno/DIDebug.jl.git")
Pkg.clone("https://github.com/Keno/TerminalUI.jl.git")
Pkg.clone("https://github.com/Keno/Gallium.jl.git")
Pkg.clone("https://github.com/Keno/ObjFileBase.jl.git")
Pkg.clone("https://github.com/Keno/MachO.jl.git")
Pkg.clone("https://github.com/Keno/ELF.jl.git")
Pkg.clone("https://github.com/Keno/AbstractTrees.jl.git")
Pkg.clone("https://github.com/Keno/VT100.jl.git")
Common Troubles
unable to connect
on Linux: Make sure ptrace protection is disabled. You can do this manually by runningecho 0 > /proc/sys/kernel/yama/ptrace_scope
Make sure that you are connecting to a valid PID of a running Julia process (seeARGS
setting above).