Due to certain circumstances, there is a choice for me to either continue with what I am doing or to stop doing it. Whether to stop or not is a question I've asked myself over the years, and still don't have a concrete answer aside from "it depends" which I will now explain a bit.

Suppose you've been working on something for a couple hours, what this usually means is that you've built up lots of context about what you're doing and so acting on ideas will be much faster than as compared to when you started working a couple of hours ago, if you have a task that you know won't take that long to tack on then it can be more worth it to just do it then and there because you can finish it faster as compared to starting tomorrow when you have lost some of the mental context.

On the flip side if you continue working on what you're doing then you can potentially start losing sleep and not get other things done that you wanted to get done, so there's a tradeoff here, roughly you could have the following (informal) equation.

Also taking a higher level look at things, you won't always be able to work on things for continuous hours at a time, and sometimes you will only have 15 - 30 minutes to work on something, and now there's another problem which is that if what you're working on is complex there is a setup time before you can start working at your usual level where you regain the context that you had last time, and so that reduces your actual time to do stuff by like 15-30 minutes making shorter sessions almost useless sometimes.

solution

The way I deal with the above problems is that I spend a bit of time every session working on systems that will help bootstrap the context rebuilding phase, the way to do that is by writing things like "aboutodo.txt" where you write out everything that you were planning on doing right before you stopped, so that you don't have to spend time trying to remember what you were going to work on, also by focusing on creating self-describing things and simple to understand things it also helps you speed up this bootstrap process. When the thing you're working on is complex this could look like breaking it down into high level chunks that can be processed easily and then if you need more detail you look inside one of the chunks so that you're not cognitively overloaded, which is the idea of abstraction.