[ruby-core:121116] [Ruby master Bug#20968] `Array#fetch_values` unexpected method name in stack trace

["Eregon (Benoit Daloze) via ruby-core" <ruby-core@ml.ruby-lang.org>]

Issue #20968 has been updated by Eregon (Benoit Daloze).


I think this is mostly a matter of getting used to `<internal:` backtrace entries, and so I think we should not do anything about this now but reevaluate in some time.
We already all agree this is not a bug.
As more methods are defined in Ruby, users are getting more used to this and won't find it surprising or unusual.
In fact these backtraces are more accurate and can help the user understand better what is going on.

As an example, I think semantically it is very nice to be able to see `Array#fetch_values` calls `Array#fetch` here.
I believe it is what users would expect (that `fetch_values` uses `fetch`), hiding that seems not good.

mame (Yusuke Endoh) wrote in #note-5:
> I agree that this is not a bug, but I think koic's expectations are clearly better than the current from a user benefit perspective, not from a consistency perspective.  
> I think it is significantly unlikely that `<internal:array>:211` is valuable information for user to debug a bug.
> It would be more convenient to combine `<internal>` backtraces into one and substitute the name of the calling Ruby source file and line number, as in the current C-implemented method.

FWIW we considered this (combine internal entries into one, or hiding internal entries, or [attribute all entries to the caller](https://github.com/oracle/truffleruby/issues/1414), etc) in TruffleRuby (it was discussed a few times on the truffleruby issue tracker, since truffleruby has many more methods defined in Ruby) but rejected this idea as it is clearly suboptimal for multiple reasons:
* Debugging an issue with a partial backtrace is always worse and more confusing (for both VM developers and users).
* Imagine that someone redefines `Array#fetch`. And when used from `fetch_values` it raises some unexpected exception. With the full backtrace like currently the user can easily understand what's going on. Without it, it's very hard. We might see the file which redefines `Array#fetch`, but not if it's redefined in C. And it would be weird for `fetch_values` to have a different number of backtraces entries based on which `fetch` method ends up being called.
* As an example for `Hash#[]=`, if that hides the `key.hash` call, and some `hash` method implementation raises, it's way more confusing if the `hash` is not shown in the backtrace.
Current:
```
$ ruby -e 'k=Object.new; def k.hash; raise "foo"; end; {}[k] = 1'
-e:1:in 'hash': foo (RuntimeError)
	from -e:1:in '<main>'
```
vs
```
$ ruby -e 'k=Object.new; def k.hash; raise "foo"; end; {}[k] = 1'
-e:1:in '<main>': foo (RuntimeError)
```
(both of these hide the `[]=` call, that's not great but due to having a special instruction for this, hard to fix without hurting performance)
* Another example could be `inject` calling some core method, if we replace the two entries by one it would be very confusing, leading to think that `inject` is raising when it's the core method being called which is raising an exception.
* `<internal:...>` is now a well-established format for this for some time already.
* The line information could be useful for users too, especially if the editor/IDE would support viewing the source file in such a case.
* It is important for reasoning that one Ruby call/frame = one backtrace entry.
* In some specific use cases it may make sense to skip `<internal:` entries (e.g. already done for `Kernel#warn`) or present them differently. That's fine, but in general we should not hide crucial information like that.

---

Also while Rubyists are used to backtraces like:
```ruby
def foo = [0].fetch(42)
foo
```
Giving:
```
fetch.rb:1:in 'Array#fetch': index 42 outside of array bounds: -1...1 (IndexError)
	from fetch.rb:1:in 'Object#foo'
	from fetch.rb:2:in '<main>'
```
They are actually "wrong" because `Array#fetch` is not defined in `fetch.rb:1`, it is defined in the core library.
I think that has BTW caused some confusion because people might not realize the location on the left should be the definition of the method mentioned on the right (since it doesn't hold for C methods).

----------------------------------------
Bug #20968: `Array#fetch_values` unexpected method name in stack trace
https://bugs.ruby-lang.org/issues/20968#change-112038

* Author: koic (Koichi ITO)
* Status: Open
* ruby -v: ruby 3.4.0dev (2024-12-19T04:44:56Z master 2783868de2) +PRISM [x86_64-darwin23]
* Backport: 3.1: UNKNOWN, 3.2: UNKNOWN, 3.3: UNKNOWN
----------------------------------------

It seems that the current Ruby implementation is displaying unexpected method name in stack trace.

## Expected

Similar to `Hash#fetch_values`, the method name `Array#fetch_values` is expected to be displayed in the stack trace.

```console
$ ruby -e '{k: 42}.fetch_values(:unknown)'
-e:1:in 'Hash#fetch_values': key not found: :unknown (KeyError)
        from -e:1:in '<main>'

$ ruby -e '[1].fetch_values(42)'
-e:1:in 'Array#fetch_values': index 42 outside of array bounds: -1...1 (IndexError)
        from -e:1:in '<main>'
```


## Actual

The stack trace displays the `Array#fetch` method, which user is not aware of, along with the `<internal.array>` stack trace.

```console
$ ruby -e '[1].fetch_values(42)'
<internal:array>:211:in 'Array#fetch': index 42 outside of array bounds: -1...1 (IndexError)
        from <internal:array>:211:in 'block in Array#fetch_values'
        from <internal:array>:211:in 'Array#map!'
        from <internal:array>:211:in 'Array#fetch_values'
        from -e:1:in '<main>'
```


It likely requires an approach such as implementing it in C, as suggested in https://github.com/ruby/ruby/pull/11555.



-- 
https://bugs.ruby-lang.org/
 ______________________________________________
 ruby-core mailing list -- ruby-core@ml.ruby-lang.org
 To unsubscribe send an email to ruby-core-leave@ml.ruby-lang.org
 ruby-core info -- https://ml.ruby-lang.org/mailman3/lists/ruby-core.ml.ruby-lang.org/