Here is what I puzzled together. Does this seem right? Is there a more efficient way to do this:

const fs = require('fs')
const tar = require('tar-stream')
const gunzip = require('gunzip-maybe');

function tar_index(path){
  const input = fs.createReadStream(path);
  const extract = tar.extract();
  let output = [];
  return new Promise(function(resolve, reject) {

    function process_entry(header, stream, next) {
      var offset = extract._buffer.shifted
      //console.log(stream)
      output.push({
        name: header.name,
        offset: offset,
        size: header.size
      });
      stream.on('end', function () {
        next() //read for next file
      })
      stream.on('error', reject);
      stream.resume();
    }

    function finish_stream(){
      resolve(output);
    }

    var extract = tar.extract({allowUnknownFormat: true})
      .on('entry', process_entry).on('finish', finish_stream).on('error', reject)
    input.pipe(gunzip()).pipe(extract);
  }).finally(function(){
    input.destroy();
  });
}

tar_index('myfile.tar.gz').then(console.log)

Should be easy to add yea. Feel free to PR that

@mafintosh sorry to pick you again, trying to understand this code.

Am I correct that extract._buffer.shifted global state contains the total offset within the tar file of where the files starts, at the time the entry callback is triggered?

I would track it independently, seems much simpler. ie a property that tracks byteOffset that is updated everytime we eat from the buffer. then add that to the header object we emit

I'll give that another try.

FWIW, the goal is to mmap the tar file in WASM using the emscripten packaging format. So we need the server to generate an index for the tar that looks like so: https://jeroen.r-universe.dev/bin/emscripten/contrib/4.4/jsonlite_1.8.9.js.metadata

I would track it independently, seems much simpler. ie a property that tracks byteOffset that is updated everytime we eat from the buffer. then add that to the header object we emit

Do you suggest we only have to add a line to _next() with byteOffset += size;? I am trying to understand how this byteOffset would be different from the existing shifted?