When we talk about rewrites I absolutely agree. As for example there have been many attempts in the financial sector to rewrite million lines of COBOL code run

When we talk about rewrites I absolutely agree. As for example there have been many attempts in the financial sector to rewrite million lines of COBOL code running on mainframes for decades. Many of those attempts failed and they probably will run COBOL indefinitely / until the code isn't useful anymore. To throw away the entire code base and start from scratch in a completely different and non compatible language, isn't a smart idea in many cases, agreed. But we don't talk about a rewrite. One very obvious difference is, that you are able to migrate slowly, by writing python code that runs on both python 2 and 3. It's possible to slowly increase the percentage of python 3 compatible code in your codebase until it reaches 100%. Many huge code bases have been successfully migrated this way, and there is no obvious reason to me against such an slow migration, again except you want to shutdown your company in 2020 anyway.