Additional output from program built with the UB sanitizer of Clang - c++

On travis CI where I use clang version 3.4 (tags/RELEASE_34/final) which is already installed, I build my code with this:
clang++ main.cpp -m64 -fsanitize=undefined -Werror -std=c++98 -pedantic -pedantic-errors -fvisibility=hidden -fstrict-aliasing -Weverything -Qunused-arguments -fcolor-diagnostics -O3 -DNDEBUG
(this command is obtained from compile_commands.json which is generated from cmake)
And in the output after the program quits some numbers (like hex representation of memory) appear:
00 00 00 10 70 fb 01 00 00 00 00 10 70 fb 01
My guess is this is from the UB sanitizer because when I build with ASAN or no sanitizer at all these numbers aren't there.
So what do they mean? How do I diagnose my UB error (if this is indeed such)?
I thought when a sanitizer encounters an error it crashes the program and prints a big message with explanation. So what is this?
This is a deal breaker for me because I compare reference output in a text file with the output of the program from the current build and such additional output breaks everything.
I tried locally using Clang 3.6, which is the default for my Ubuntu using the same build command, but when I run the executable I get no errors or such additional output.
here is the failing build on travis - and I don't think my code is relevant because my problem is with the sanitizer output not being helpful at all.

I also enabled the builds with clang 3.5/3.6/3.7/3.8 and turns out clang 3.5 behaves the same way...
clang 3.6 however gives more output!
20 6c 98 01 00 00 00 00 20 6c 98 01 00 00 00 00 20 6c 98 01
^
/usr/lib/gcc/x86_64-linux-gnu/4.6/../../../../include/c++/4.6/bits/stl_tree.h:247:17: runtime error: upcast of address 0x00000115e090 with insufficient space for an object of type 'std::_Rb_tree_node<doctest::detail::TestData>'
0x00000115e090: note: pointer points here
00 00 00 00 00 00 00 00 00 00 00 00 20 6c 98 01 00 00 00 00 20 6c 98 01 00 00 00 00 20 6c 98 01
clang 3.7 and 3.8 give the same output as 3.6
I am using libstdc++ so I will switch to libc++ to hopefully remove this error (which I think is not from my code!)
I was using a simple TestData structure inside a std::set<>...

Related

Configure ocamlopt so gdb can get source location information

The native OCaml compiler exposes options that control whether debug information is emitted. For instance -g controls whether to record information needed to reconstruct exception backtraces. Is there an option to emit the debug information that would be needed for gdb to associate breakpoints with source information like file name and line number?
I don't think OCaml is, at present, a fully supported language for gdb and it isn't possible to pretty-print values or evaluate OCaml expressions. That is okay, I'm just wondering how to configure the ocamlopt compiler or gdb in such a way that gdb can find the source file. Ideally I'd like to be able to see both the OCaml sources and the C source files that implement OCaml runtime when they are present (i.e. when building a compiler by hand rather than through OPAM).
For instance,
(* hello.ml *)
let main () =
Printf.printf "hi there\n%!";;
let () = main ()
compiled using corebuild hello.native, produces a symlink to an executable, hello.native.
and then, when starting gdb:
(gdb) file hello.native
Reading symbols from hello.native...done.
(gdb) start
Temporary breakpoint 1 at 0x405580: file main.c, line 32.
Starting program: /home/g/ws/tmp/ocaml/hello/hello.native
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
Temporary breakpoint 1, main (argc=0x1, argv=0x7fffffffdaf8) at main.c:32
32 main.c: No such file or directory.
We can see that it was unable to determine where the file is.
When you hit C-x a and switch gdb to tui mode, then the message [ No Source Available ] is visible in the top pane.
Use the dir directive in gdb to point it to the place where is the source code of the OCaml runtime, e.g.,
(gdb) dir ~/warehouse/ocaml/byterun/
Source directories searched: /home/ivg/warehouse/ocaml/byterun:$cdir:$cwd
(gdb) l
27 #endif
28
29 CAMLextern void caml_main (char_os **);
30
31 #ifdef _WIN32
32 CAMLextern void caml_expand_command_line (int *, wchar_t ***);
33
34 int wmain(int argc, wchar_t **argv)
35 #else
36 int main(int argc, char **argv)
Few more tips. You can link your program with the debugging runtime, e.g.,
ocamlopt -runtime-variant x -g hello.ml -o hello
It won't have the source code embedded though.
Also, OCaml has quite a good support for gdb, you can step, backtrace, and even observe the source code. The only problem is that names are usually mangled, so it is hard to set up a breakpoint. However, you can use objdump to reverse engineer your file. It is easy if it is built with the -g option:
$ objdump -S hello | grep hello.ml -A 10
(* hello.ml *)
let main () =
404a70: 48 8d 1d 81 a6 24 00 lea 0x24a681(%rip),%rbx # 64f0f8 <camlHello__5>
404a77: 48 8d 05 da b3 24 00 lea 0x24b3da(%rip),%rax # 64fe58 <camlPervasives>
Printf.printf "hi there\n%!";;
404a7e: 48 8b 80 d0 00 00 00 mov 0xd0(%rax),%rax
404a85: e9 66 ae 01 00 jmpq 41f8f0 <camlPrintf__fprintf_1294>
404a8a: 66 0f 1f 44 00 00 nopw 0x0(%rax,%rax,1)
0000000000404a90 <camlHello__entry>:

GetFileVersionInfo equivalent in Linux

I'm porting a windows application to Linux, the application fetches version from a exe using Windows APIs GetFileVersionInfoSize() and GetFileVersionInfo(). But I'm unable to find something similar to that in Linux.
I'm aware that I can use readelf to read contents of application or object files in Linux as below
[root#Panch-Linux]# readelf <xyz.so> -h
ELF Header:
Magic: 7f 45 4c 46 02 01 01 03 00 00 00 00 00 00 00 00
Class: ELF64
Data: 2's complement, little endian
Version: 1 (current)
OS/ABI: UNIX - Linux
ABI Version: 0
Type: EXEC (Executable file)
Machine: Advanced Micro Devices X86-64
Version: 0x1
Entry point address: 0x4f7ec0
Start of program headers: 64 (bytes into file)
Start of section headers: 17823344 (bytes into file)
Flags: 0x0
Size of this header: 64 (bytes)
Size of program headers: 56 (bytes)
Number of program headers: 8
Size of section headers: 64 (bytes)
Number of section headers: 38
Section header string table index: 35
Shoud I use readelf to get the version information, if yes then the string "Version" is the one which actually represents the file version. Also please let me know how to use readelf in C++ programming on Linux environment.
If this is not the right way, please suggest me appropriate way of doing this.

Visual Leak Detector - File and line number not available

I am debugging an MFC program and have come to a point where I am trying to find and resolve memory leaks. First I tried the built-in leak detector in VC, but once I redefined the new operator I got lots of problems with redefinitions of "new" and errors.
Moved on to Visual Leak Detector. I have got it up and running and it seems to find the leaks, but it cannot seem to find the file and line numbers. This is what it looks like:
WARNING: Visual Leak Detector detected memory leaks!
---------- Block 2 at 0x0343F718: 67 bytes ----------
Call Stack:
0x00B88E2F (File and line number not available): program.exe!for__get_vm + 0x1F bytes
0x00B8FC28 (File and line number not available): program.exe!for_rtl_init_wrap_ + 0x118 bytes
f:\dd\vctools\crt_bld\self_x86\crt\src\crt0dat.c (873): MSVCR100D.dll!_initterm
f:\dd\vctools\crt_bld\self_x86\crt\src\crtexe.c (473): program.exe!__tmainCRTStartup + 0xF bytes
f:\dd\vctools\crt_bld\self_x86\crt\src\crtexe.c (371): program.exe!wWinMainCRTStartup
0x7540338A (File and line number not available): kernel32.dll!BaseThreadInitThunk + 0x12 bytes
0x77819F72 (File and line number not available): ntdll.dll!RtlInitializeExceptionChain + 0x63 bytes
0x77819F45 (File and line number not available): ntdll.dll!RtlInitializeExceptionChain + 0x36 bytes
Data:
43 3A 5C 50 72 6F 6A 65 63 74 5C 50 72 6F 70 63 ........ ........
61 6C 63 5C 53 65 6B 6F 6E 64 65 6E 54 72 75 6E ........ ........
6B 5C 77 42 6C 61 64 65 45 64 5C 2E 5C 44 65 62 .Random. .Symbols
75 67 5C 77 42 6C 61 64 65 45 64 2E 65 78 65 00 ........ ........
CD CD CD ........ ........
Having checked the FAQ on the Visual Leak Detector (VLD) website, there is a section stating that if this error occurs, then it may be so that VLD cannot find the debug symbols located in programname.pdb, since it must be situated alongside the programname.exe file. I have checked (and rebuilt), making sure that my project has the "Generate Debug Symbols" option turned on and that the programname.exe file gets created and put next to the programname.exe file in the debug folder.
Still, I refuses to show me the file and line numbers.
Also, at the very top of my "output window" in Visual Studio i can see the following line:
'Program.exe': Loaded 'C:\Project\Program\Debug\Program.exe', Symbols loaded.
So, the symbols are loaded?
Any ideas?
Thanks!

Linking to 32bit shared object file in 32 bit enviornment generates ELFCLASS64 error

System : I'm running 32 bit Ubuntu 12.04 on a i32 chipset.
Build Info : I'm using C++ 11 with Qt 4.8.1 and GCC.
Problem : I am developing a project that depends on in-house developed libraries hereafter mentioned to as slug. These libraries are built into .so files via the traditional "cmake ." then "make all" process. After the .so files are generated from cmake, I copy the .so files into a sub-directory of my project so that I can test that they are working before I install them into /usr/lib. Then I provide Qt (in my .pro file) with a absolute path to each library and build my application. My application finds these local libraries and compiles without errors or warnings. However, when I run my application (via command line) I receive the following error:
error while loading shared libraries: libslugSpec.so: wrong ELF class: ELFCLASS64
However, when I build the slug libraries on a 64 bit environment and link to them in a 64 bit environment my application runs perfectly. So, I figured I built the .so files for a 64 bit environment,and proceeded to edit the CMakeLists.txt file used to build the libraries. I add numerous 32 bit compilation and linking flags, shared object module flags, in addition to cuda (which slug relies on) bitness flags . No matter what combination of flags I tried, the libraries generated kept causing the error when running my application executable. So I decided to run the readelf -h command on each shared object file and check to see if they were really 64 bit. I found out that none of them are in fact, all producing logs similar to the following: (output for one of 3 .so files)
readelf -h libslugSpec.so
ELF Header:
Magic: 7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
Class: ELF32
Data: 2's complement, little endian
Version: 1 (current)
OS/ABI: UNIX - System V
ABI Version: 0
Type: DYN (Shared object file)
Machine: Intel 80386
Version: 0x1
Entry point address: 0x10150
Start of program headers: 52 (bytes into file)
Start of section headers: 511960 (bytes into file)
Flags: 0x0
Size of this header: 52 (bytes)
Size of program headers: 32 (bytes)
Number of program headers: 7
Size of section headers: 40 (bytes)
Number of section headers: 38
Section header string table index: 35
So now I am completely at a loss and can't seem to find an answer anywhere. Any illumination into how this could be happening would be a huge help.
For reference, here are the flags I am providing in the qt .pro file that builds my application:
QMAKE_CXXFLAGS += "-std=gnu++0x"
QT += core gui opengl
TARGET = source
TEMPLATE = app
CONFIG+= wwwidgets
#linux specific settings
unix:!macx{
LIBS += -lGLEW -lGLU /home/alex/Dropbox/brain-viz/source/lib/libslugUtil.so /home/alex/Dropbox/brain-viz/source/lib/libslugSpec.so /home/alex/Dropbox/brain-viz/source/lib/libslugSim.so
}
#windows specific settings
win32{
LIBS += -lglew32 -lwwwidgets4d
}
This error message:
error while loading shared libraries: libslugSpec.so: wrong ELF class: ELFCLASS64
Is not consistent with this output:
readelf -h libslugSpec.so
ELF Header:
Magic: 7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
Class: ELF32
Conclusion: you load a different version of libslugSpec.so from the one you ran readelf -h on.
Find out which libslugSpec.so is loaded at runtime by doing this:
LD_DEBUG=files,libs /path/to/your/executable
Then run readelf -h on that library, and confirm that it is in fact ELF64.
Finally replace it with the version you've built (or set LD_LIBRARY_PATH to pick up your version before "system" one), and the problem will be solved.

Teach emacs recognize Boost.Test errors

This is an output of Boost.Test when test case failes:
bjam toolset=msvc
...patience...
...found 1287 targets...
...updating 4 targets...
compile-c-c++ ..\bin\test\Function.test\msvc-8.0\debug\link-static\threading-multi\Function.obj
Function.cpp
msvc.link ..\bin\test\Function.test\msvc-8.0\debug\link-static\threading-multi\Function.exe
msvc.manifest ..\bin\test\Function.test\msvc-8.0\debug\link-static\threading-multi\Function.exe
testing.capture-output ..\bin\test\Function.test\msvc-8.0\debug\link-static\threading-multi\Function.run
====== BEGIN OUTPUT ======
Running 1 test case...
Function.cpp(26): fatal error in "FunctionConstruction": critical check pf->Name() == "F13" failed [F1 != F13]
*** 1 failure detected in test suite "foo_test"
Detected memory leaks!
Dumping objects ->
{235} normal block at 0x003A7C88, 32 bytes long.
Data: 00 00 00 00 CD CD CD CD 54 31 00 CD CD CD CD CD
{234} normal block at 0x003A7E00, 96 bytes long.
Data: 00 00 00 00 CD CD CD CD 54 31 00 CD CD CD CD CD
{233} normal block at 0x003A7D88, 76 bytes long.
Data: F4 D9 45 00 00 00 00 00 CD CD CD CD 00 7E 3A 00
Object dump complete.
EXIT STATUS: 201
====== END OUTPUT ======
MSVC parses this errors correctly so i can double click and jump to place in code. But emacs can't parse this output. How to teach it?
The solution will involve customizing the variables: 'compilation-error-regexp-alist, 'compilation-error-regexp-alist-alist, 'compilation-directory-matcher.
The first, 'compilation-error-regexp-alist is just a list of symbols telling the compilation mode what to look up in the second variable `'compilation-error-regexp-alist-alist', so you'll probably just add something for boost:
(add-to-list 'compilation-error-regexp-alist 'boost)
Then, to make that work, you need to add a list to the second variable, 'compilation-error-regexp-alist-alist. This is where it starts to get tricky. You'll need to read the documentation for the first variable to get the regexp right, but it'll be something like:
(add-to-list 'compilation-error-regexp-alist-alist
'(boost
"^\\(.*\\)(\\([0-9]+\\)): fatal error in" 1 2))
The regexp matches the error line, and the 1 and 2 specify the sub expression specifying the filename and line number respectively. There are other things you can specify (see the documentation).
Though, to be honest, the above two settings are probably unnecessary as I'm pretty sure one of the existing regexps will match the format. The problem really exists with the directory tracking.
The last variable, `'compilation-directory-matcher' is the one that lets the next-error track where to find the files. So it needs to be updated appropriately. It doesn't look like the boost test spits out the somewhat standard "Entering directory ..." that Emacs looks for, but the information seems to be there in the compile line...
You might also try asking on the boost user mailing list to see if someone there has solved this problem. The mailing list can be found here.