Comment by toast0
7 hours ago
> But you can do the same thing in PHP or with CGI without sacrificing much reproducibility
PHP is similar, but not the same. You can't (or at least I can't) stop a request in progress and change its code; but you can rapidly change the code for the next request. Make a change in the editor, hit reload in the browser is a productive short loop, but stop at a breakpoint, inspect the state and change the code is a more powerful loop. Stopping at a breakpoint is challenging in systems with communication though, and I've learned to live without it for the most part.
Yes, I agree. I meant that you can edit code on the production web server so that the customer you're on the phone with sees their application start working. PG had the advantage that he could tweak the Lisp process for that specific user, reducing the risk of breaking the site for other concurrent users.
Database transactions bridge some of the gap between "change the code for the next request" and "stop at a breakpoint and change the code": as long as your request handler code keeps failing, it will abort the transaction, so the database is unchanged, so you can restart the transaction as many times as you want to get to the same point in execution, at least if your programming language is deterministic. By providing a snapshot you can deterministically replay from, it allows you to add log entries before the point where the problem occurred, which can be very useful.
Stopping at a breakpoint can be more productive, especially with edit-and-continue, but often it isn't. A breakpoint is a voltmeter, which you can use to see one value at every node in your circuit; logs are a digital storage oscilloscope with a spectrum analyzer, where you can analyze the history of millions or billions of values at a single node in your circuit.