Migrated a small-ish project from ASP.NET Framework (.NET 4.8) to ASP.NET Core (i.e. .NET 8) including EF6 to EF Core last year. Even though it was pretty smooth sailing (the application was already using MVC), it was still a few weeks work.
Rewrite queries that are not yet supported on EF Core, create some new DTOs, create drop-in replacements for stuff that isn't there anymore, create plumbing to connect some old stuff to some new stuff. Figure out what's the new replacement for XYZ and use it correctly to not break anything. Replace nuget packages that are not maintained anymore or have so few downloads that they are a supply chain risk.
You need to test every endpoint, every query in your application.
And in the Microsoft World you're still pretty ok, overall, it's quite stable.
God forbid you put your money on the wrong web framework (not even talking frontend) in the Java space. There's still Tapestry/Struts/Grails/JSF/GWT/Play etc. applications serving your request out there. Saw some banking website with ".do" urls recently (indicative of Struts).
Unless we get superhuman AI that can do such migrations all on their own, those zombies might have a longer working life left than lots of developers. Many companies just can't afford to upgrade - there's so many systems built over decades that just do what they should do. Why shell out tons of money to just get the same functionality?
> Copilot agent mode can create apps from scratch, perform refactorings across multiple files, write and run tests, and migrate legacy code to modern frameworks.
I wonder if the relative lack of training data makes this harder for AI. Most of these legacy apps are in private repositories, so there’s nothing for GitHub to scrape to identify the before/after.
And if Microsoft is creating synthetic data, that won’t account for the truly awful ways in which enterprise software evolves (or devolves).
Rewrite queries that are not yet supported on EF Core, create some new DTOs, create drop-in replacements for stuff that isn't there anymore, create plumbing to connect some old stuff to some new stuff. Figure out what's the new replacement for XYZ and use it correctly to not break anything. Replace nuget packages that are not maintained anymore or have so few downloads that they are a supply chain risk.
You need to test every endpoint, every query in your application.
And in the Microsoft World you're still pretty ok, overall, it's quite stable.
God forbid you put your money on the wrong web framework (not even talking frontend) in the Java space. There's still Tapestry/Struts/Grails/JSF/GWT/Play etc. applications serving your request out there. Saw some banking website with ".do" urls recently (indicative of Struts).
Unless we get superhuman AI that can do such migrations all on their own, those zombies might have a longer working life left than lots of developers. Many companies just can't afford to upgrade - there's so many systems built over decades that just do what they should do. Why shell out tons of money to just get the same functionality?