Issue #17930 has been updated by mame (Yusuke Endoh).


I brought this ticket to the agenda of today's dev-meeting, and confirmed that matz has already accepted this.

The accepted implementation approach is (1) to extend `RubyVM::AST.of` to accept `Thread::Backtrace::Location` to get the node and column information and (2) to create a new built-in MRI-specific gem (rather than did_you_mean gem) to extend `Exception#message`.

I'm prototyping the new gem, but I found that it is not trivial where to underline depending on context. So I will have to adjust the underline location, according to bug reports. After the specification is almost fixed, the implementation may be exposed as `NameError#code_location` or something in future, but at first, I would like to start this as a built-in feature.

----------------------------------------
Feature #17930: Add column information into error backtrace
https://bugs.ruby-lang.org/issues/17930#change-92565

* Author: mame (Yusuke Endoh)
* Status: Assigned
* Priority: Normal
* Assignee: mame (Yusuke Endoh)
----------------------------------------
Consider the following code and error.

```
data["data"].first["field"] #=> undefined method `[]` for nil:NilClass
```

There are two possibilities; the variable `data` is nil, or the return value of `first` is nil. Unfortunately, the error message is less informative to say which.

This proposal allows to help identifying which method call failed.

```
$ ruby -r ./sample/no_method_error_ext.rb err1.rb
err1.rb:2:in `<main>': undefined method `[]' for nil:NilClass (NoMethodError)

data["data"].first["field"]
                  ^^^^^^^^^
```

## Proposal

I'd like to propose a feature to get column information from each `Thread::BacktraceLocation`. Maybe it is good to provide the following four methods:

* `Thread::BacktraceLocation#first_lineno`
* `Thread::BacktraceLocation#first_column`
* `Thread::BacktraceLocation#last_lineno`
* `Thread::BacktraceLocation#last_column`

These names came from `RubyVM::AbstraceSyntaxTree::Node`'s methods.

## Implementation

Here is a proof-of-concept implementation: https://github.com/ruby/ruby/pull/4540

See https://github.com/ruby/ruby/pull/4540/commits/6ff516f4985826e9f9c5606638001c3c420f7cad for an example usage.
(Note that, currently, you need to build ruby with `./configure cflags=-DEXPERIMENTAL_ISEQ_NODE_ID` to enable the feature.)

To put it simply, this PR provides only a raw API, `Thread::BacktraceLocation#node_id`. To get actual column information, you need to manually identify `RubyVM::AbstractSyntaxTree::Node` that corresponds to `Thread::BacktraceLocation#node_id`.
But it would be arguable to expose "node_id", so I will wrap it as the above four methods if this is accepted.

Credit: the original implementation was done by @yui-knk.

## Drawback

To use this feature, we need to enable `-DEXPERIMENTAL_ISEQ_NODE_ID` to add "node_id" information (a subtree ID of the original abstract syntax tree) into each byte code instruction. If we provide this feature, the option should be enabled by default. However, the option increases memory consumption.

I performed a simple experiment: I created a scaffold app by `rails new`, and measured the memory usage after `rails s`. The result was 97 MB without `-DEXPERIMENTAL_ISEQ_NODE_ID`, and 100 MB with the option enabled.

In my opinion, it is not so large, but requiring more gems will increase the difference. I will appriciate it if anyone could provide the actual memory increase in a more practical Rails app.

Do you think this feature deserves the memory increase?

---Files--------------------------------
image.png (73.3 KB)


-- 
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>