In the unlikely case that this is a problem, it's fairly straightforward to switch ...

I think we are not so far apart on this.

Switching involves abandoning one approach to solving the problem (and I might more clearly have expressed the idea that 'this approach' meant 'read the entire file to a string, operate on the string, and write the string back to a file') for a significantly different approach, perhaps one of those outlined in the balance of your reply. This is what I mean when I write that the approach does not scale well.

I agree that, given what one can divine about the OPer's data from the OP, the problem is not likely to arise. In fact, I would go further and say that, in my experience, problems like this never happen – until they happen, and then you have to go back and change a bunch of stuff and re-test everything; which, while straightforward, still costs time, effort and risk; hence, the caution about lack of scalability.