Source Map Support in Node.js

source-map-example.md

Programatic Access of Source Maps

To begin collecting source-maps, node must be run with --enable-source-maps (or with NODE_V8_COVERAGE set).

At which point, the new built-in module source_map can be used to interact with the source-map cache:

// tooling has one interface into the source-map cache, URIs, any type of module that we
// wish to attach source-maps to (ESM, CJS, evals, data-urls) needs to have a methodlogy
// for setting a URI; Two different source-maps cannot be stored at the same URI.
const {cache} = require('source_map');
const sm1 = cache.sourceMapFromURI(`file://${require.resolve('./my-module')}`);

These methods return the same data-structure that is written to disk when executing with NODE_V8_COVERAGE:

{
  "url": "data:abc123", // original source-map URL.
  "data": {} // parsed source-map or null.
}

Stack Traces When Source Maps are Enabled

When executing with --enable-source-maps, stack traces will be rewritten if source-maps are found corresponding to files represented in the stack-trace.

The hope is to intitially engage folks rom the https://github.com/evanw/node-source-map-support project, to support stack-trace rewritting in Node.js. And, in the future, work with the V8 team to move more of this behavior into V8 itself.

@ljharb are you picturing an API more like this:

const {fetchSourceMap} = require('source_map');
const sm1 = fetchSourceMap(require.resolve('./my-script'));

1. if the source-map is already in the cache, return the object:

   {
     "url": "file://original-source-map-url",
      "data": {} // maybe loaded Source Map V3.
   }

2. if the Source Map was not yet loaded, we read the file from disk and look for a source map potentially; returning the same format.

What's your reasoning for a Buffer rather than returning the object, as suggested above?

This is fantastic and a real pain for a lot of Node.js developers - thank you for working on it.

Making source maps work out of the box in stack traces would be great and it's mostly blocked on someone actually doing the work on making it happen.

@bcoe yes, the object would work fine. I'd initially thought of a Buffer so it would directly return the content (the source being irrelevant), but I'd forgotten about remote URLs (which would not be good for node to implicitly request).