Issue #10561 has been updated by Eregon (Benoit Daloze).


There is an extra difference that #absolute_path will also try to canonicalize the path and resolve symlinks, so then even if #path is absolute, #absolute_path can be different.
See https://github.com/ruby/spec/blob/e829fb0bcb667013f3527142b9e29883f6904537/core/thread/backtrace/location/absolute_path_spec.rb#L39-L69

So in short #path is the same as `__FILE__`. #absolute_path is essentially File.realpath(#path) + it seems to work on CRuby even if the file was removed.
For eval(code, binding, file, line) code, #path and #absolute_path just return `file` and do not expand it, so it can be a relative path if `file` is relative.

----------------------------------------
Feature #10561: Improve function of Thread::Backtrace::Location #path and #absolute_path
https://bugs.ruby-lang.org/issues/10561#change-87668

* Author: sam.saffron (Sam Saffron)
* Status: Open
* Priority: Normal
----------------------------------------
I was working on this issue in Rails and hit an area where Backtrace Location can be improved

https://github.com/rails/rails/pull/17782


1. It is undefined in the documentation how #absolute_path should operate when #path is invalid (in case of instance eval)
2. There are a few conditions where #path and #absolute_path can return nil, this forces extra protection code when parsing paths to check for nil. (for example getting filename)  

Suggestions:

1. Instead of returning Qnil from location_path and location_absolute_path on invalid conditions, return the string "(unknown)" which is easier to parse and sticks out better in a big backtrace. There is precedent here with the string "(eval)"
2. If path is invalid have absolute_path return "(unknown)", define that in the documentation
3. (possible) add an additional method on caller_location called #filename so people stop parsing filename from #path and #absolute_path
4. Evaluate if it makes sense to have #path and #absolute_path in the API as both methods can return full paths so the semantic difference is subtle. 



-- 
https://bugs.ruby-lang.org/

Unsubscribe: <mailto:ruby-core-request / ruby-lang.org?subject=unsubscribe>
<http://lists.ruby-lang.org/cgi-bin/mailman/options/ruby-core>