Re: Stack traces not displaying correctly on macOS

[Paul Rockwell]

Just got finished downloading SVN 1752 and recompiling.

The Backtrace::show_item() code seems to work fine on both Linux (Fedora 39 x64) and macOS (Sonoma 14.3.1 Apple Silicon).

macOS:

      ⎕FIO ¯16 

==============================================================================

Assertion failed: 0 && "Simulated Assert() (aka. ⎕FIO ¯16)"

in Function:      eval_B

in file:          Quad_FIO.cc:1191

C/C++ call stack:

*** useless apl.lines (no CXXFLAGS=-rdynamic -gdwarf-2)

----------------------------------------

-- Stack trace at Assert.cc:75

----------------------------------------

0x1026570bc main

0x1028759e8  Workspace::immediate_execution(bool)

0x1026c8660   Command::process_line()

0x1026c8964    Command::process_line(UCS_string&, std::__1::basic_ostream<char, std::__1::char_traits<char>>*)

0x1026cc128     Command::do_APL_expression(UCS_string&)

0x1026e0954      Command::finish_context()

0x102700600       Executable::execute_body() const

0x1027fbc28        StateIndicator::run()

0x102751e70         Prefix::reduce_statements()

0x10274ece8          Prefix::reduce_MISC_F_B_()

0x1027a4134           Quad_FIO::eval_B(Value_P) const

0x10267d240            do_Assert(char const*, char const*, char const*, int)

========================================

SI stack:

Depth:      0

Exec:       0x600001860000

Safe exec:  0

Pmode:      ◊  ⎕FIO ¯16 

PC:         3 (4) RETURN_STATS

Stat:       ⎕FIO ¯16

err_code:   0x0

Fedora 39 x64:

      ⎕FIO ¯16

==============================================================================

Assertion failed: 0 && "Simulated Assert() (aka. ⎕FIO ¯16)"

in Function:      eval_B

in file:          Quad_FIO.cc:1191

C/C++ call stack:

*** useless apl.lines (no CXXFLAGS=-rdynamic -gdwarf-2)

----------------------------------------

-- Stack trace at Assert.cc:75

----------------------------------------

0x7f1971a4620b __libc_start_main

0x7f1971a4614a  

0x472a6a   main

0x60aa25    Workspace::immediate_execution(bool)

0x4d517d     Command::process_line()

0x4d5961      Command::finish_context()

0x4e2087       Executable::execute_body() const

0x59a610        StateIndicator::run()

0x5210b1         Prefix::reduce_statements()

0x51e3d5          Prefix::reduce_MISC_F_B_()

0x55e65e           Quad_FIO::eval_B(Value_P) const

0x4884dc            do_Assert(char const*, char const*, char const*, int)

========================================

SI stack:

Depth:      0

Exec:       0xbb6d40

Safe exec:  0

Pmode:      ◊  ⎕FIO ¯16

PC:         3 (4) RETURN_STATS

Stat:       ⎕FIO ¯16

err_code:   0x0

One small item. Appreciate the credit on the copyright notice in Backtrace.cc for SVN 1752. But I'd suggest not including "(Apple)" on the notice as I am not associated with Apple.

I would not want them to have any copyright claim on the code I submitted to you. That code was engineered by examining the output of the macOS system library function (which, by the way they did not document the string format), and not by referencing any Apple-provided examples.

- Paul

On Feb 13, 2024, at 9:35 AM, Paul Rockwell <paulrockwell@mac.com> wrote:

Jürgen, 

I’ll download and test SVN 1752 on both macOS and a Fedora Intel system I have hanging around.  I figured you don’t use a Mac, so that’s why I took it upon myself to exercise some dormant coding skills and provide the diffs. 

The macro __APPLE__ is a system defined macro for macOS. I will find which system defined header file it resides in. It was recommended by some research to provide conditional compilation of code for the platform without the need for additional compiler flags. It shouldn't exist on non Apple platforms. 

- Paul

On Feb 13, 2024, at 8:21 AM, Dr. Jürgen Sauermann <mail@jürgen-sauermann.de> wrote:

 Hi Paul,

thanks. I have incorporated your diffs, SVN 1752.
Please check on your Apple since I don't own one (⎕FIO ¯16 
to provoke a stack dump).

I was also wondering where macro _APPLE_ is defined. In a header file
(which ?) or with -D on the command line? In the latter case we could let
./configure set the macro (for that I would need to know the output of
the uname command on your machine).

Best Regards,
Jürgen

On 2/12/24 23:40, Paul Rockwell wrote:

I've found that stack traces don't display correctly with GNU APL on macOS platforms (Intel and Apple Silicon). Examining the source for Backtrace.cc, I've found that function Backtrace::show_item() is expecting the string passed to it to look like this:

  // ./apl(_ZN10APL_parser9nextTokenEv+0x1dc) [0x80778dc]

  // │││││                             │││││   │││││││││

  // └┴┴┴┴─── program name             │││││   └┴┴┴┴┴┴┴┴───── abs_addr

  //                                   └┴┴┴┴───────────────── asm_offset

The macOS system function backtrace_symbols() - called by Backtrace::show() returns the strings in this format:

  //

  // 14  apl                                 0x00000001000a93f0 _ZN9Workspace19immediate_executionEb + 68

  // ││  └┴┴─── program name                 ││││││││││││││││││ ││││││││││││││││││||||||||||||||||||   └┴─── asm_offset

  // └┴─── stack level                       ││││││││││││││││││ └┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴─── function name

  //                                         └┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴┴───── abs_addr

  //

These strings don't match what the code is looking for, so the stack trace doesn't contain any useful information.

I prototyped some changes to Backtrace::show_item() to be able to recognize that GNU APL is being built on Apple macOS, and parse the macOS backtrace_symbols() string format. The resulting code seems to compile and work correctly on Intel Linux and macOS (both on Intel and Apple Silicon CPUs).

I've attached the diffs between SVN 1745 and my changed code. (I also made some other changes to the code to reflect use of nullptr and '\0' instead of the number 0 to improve the contextual clarity of some of the code. There are also a couple of changes to remove some compiler warnings on macOS for unreachable code)

You're welcome to use them if you find them useful.

- Paul Rockwell