That is one option. I tend to prefer, after a few times doing this, to rewrite the app. The number of improvements and bugs fixed tend to outweigh the number of errors.

If you have buggy code that nobody understands then translation may be a better option.

In my case my clients were perfectly happy with their little language and the quality of their codebase. They were not happy with their PDP-11s - which were getting more expensive to maintain every year. So building a new compiler for a new box was the obvious decision. For all I know they're still tweaking their little expert system now.

If we'd gone the translation route then not only would the development time have been longer, the client would have also had to retrain their internal people to use a new language.

Also 20KLOC isn't really that much. When you have organisations with hundreds of programmer years invested in their code you know why they prefer to take compiler development in-house rather than manually convert everything to whatever the current language-of-choice is :-)

The number of improvements and bugs fixed tend to outweigh the number of errors. Granted, I have only worked with relatively young apps and rewriting them after they were implemented by junior developers. *shrugs* YMMV, as with everything

Indeed. Sometimes moving to another language is exactly the right thing to do. Context is everything.