context-mode output sample

gistfile1.js

const fs=require('fs');
const t=fs.readFileSync('captures/test/index.html','utf8');
const r=fs.readFileSync('captures/reference/index.html','utf8');
const idxs=[];
for(let i=0;i<t.length;i++) if(t[i]==='�') idxs.push(i);
console.log('replacement count',idxs.length);
for(const i of idxs){
  const pre=t.slice(Math.max(0,i-40),i);
  const post=t.slice(i+1,i+40);
  const marker=pre+post;
  let pos=-1;
  // try find using trimmed context pieces
  for(let k=20;k>=6 && pos<0;k-=2){
    const key=t.slice(Math.max(0,i-k),i)+t.slice(i+1,i+1+k);
    pos=r.indexOf(key);
  }
  let expected='?';
  if(pos>=0){
    // find char in ref between matching halves
    // brute around pos
    const around=r.slice(Math.max(0,pos-5),pos+90);
    // estimate expected at split point: char at pos + leftLen maybe unknown; just inspect nearest non-ascii around there
    const near=r.slice(Math.max(0,pos-10),pos+60);
    const m=near.match(/[\u0080-\uFFFF]/);
    if(m) expected=m[0];
  }
  console.log('---');
  console.log('test ctx',JSON.stringify(pre+'[�]'+post));
  if(pos>=0){
    console.log('ref ctx ',JSON.stringify(r.slice(Math.max(0,pos-40),pos+80)));
  } else {
    // fallback by searching pre and post separately
    const p=r.indexOf(pre.slice(-20));
    const q=r.indexOf(post.slice(0,20));
    console.log('match fail',p,q);
  }
}