Install dSYM debug information for macOS

[dylanahsmith]

Problem

On macOS, ruby reads the debug information from a *.dSYM directory corresponding to get a C source file and line number on a backtrace. Although ruby.dSYM was getting generated in the build directory, it wasn't getting installed along with the ruby binary. Also, native extensions neither had the *.dSYM directories generated or installed.

For example, try using this from an installed version of ruby on macOS

ruby -r bigdecimal -e 'BigDecimal.save_exception_mode { Process.kill("SEGV", Process.pid) }' 2>&1 | grep -A1 BigDecimal_save_exception_mode

/Users/dylants/.rubies/ruby-master/lib/ruby/3.1.0/x86_64-darwin20/bigdecimal.bundle(BigDecimal_save_exception_mode+0x6b) [0x11200323b]
/Users/dylants/.rubies/ruby-master/bin/ruby(vm_call_cfunc_with_frame+0x14f) [0x10e73d0af]

Gems installed through gem install or bundle install had the same problem.

Solution

I used the feature detection configure.ac does to detect if the dsymutil program is present on the darwin platform to conditionally generate this debug information (as part of POSTLINK) and install the generated *.dSYM directories.

I tested on macOS and linux for installing ruby and a gem with a native extension through both gem install and bundle install and testing cleaning make clean and gem uninstall. It seems to work as expected with these changes.

For example, I get the following output now from the above SEGV test within BigDecimal.save_exception_mode

/Users/dylants/.rubies/ruby-dev/lib/ruby/3.1.0/x86_64-darwin20/bigdecimal.bundle(BigDecimal_save_exception_mode+0x6b) [0x107b7323b] ../../../ext/bigdecimal/bigdecimal.c:3239
/Users/dylants/.rubies/ruby-dev/lib/ruby/3.1.0/x86_64-darwin20/bigdecimal.bundle(BigDecimal_save_exception_mode) (null):0
/Users/dylants/src/ruby/build-dev/./ruby(vm_call_cfunc_with_frame+0x14f) [0x1042b50af] ../vm_insnhelper.c:2924

[dylanahsmith]

I had to remove these compiler stubs, since it would otherwise try to use dsymutil on the empty files created from touch.

[nobu]

Rubygems might not assume the compiler, why not remove CONFIG['dsymutil'] instead?

[nobu]

The failures seem coming from codesign, and this file won't need changes after 8a89dd2.

[dylanahsmith]

Rubygems might not assume the compiler, why not remove CONFIG['dsymutil'] instead?

Handling CC not referencing a C compiler when compiling a C compiler seems overly contrived. There should be no need to remove CONFIG['dsymutil'], since the configure script should leave it blank on platforms where it isn't available. Besides, the CI tests are passing on all platforms as it is now.

The failures seem coming from codesign, and this file won't need changes after 8a89dd2.

By failures I guess you mean from you running the tests locally with the RUBY_CODESIGN environment variable set. Thanks for catching that, I didn't realize that would cause the test to fail.

[dylanahsmith]

This isn't needed and was causing problems without ruby installed, since it wasn't able to find the ruby/config.h header, since it wasn't finding it where it gets installed. So I removed it to avoid breaking this test.