A 256 KB input file for Custom Invocations is many times not enough to cover large tests for problems.
The whole point of custom invocation is that it allows contestants to see how their code behaves on the machine used by the judge. Particularly handy when checking for time limits, since you cannot know how fast your code will run on the target machine.
With full feedback this isn't really a problem, since you'll just submit your code and get the "time limit" verdict and you'll know you have to optimize your code. But with partial feedback, you can test your code on your local machine, see that it runs in reasonable time, submit it, pass the pretests with plenty of time to spare and get time limit on the main tests. This tends to happen particularly when problems run in ~10^8 steps with strict time bounds. I feel like when this happens, it isn't really the programmer's fault since they cannot know for sure how fast their code will run on the judging machine. This is where Custom Invocation comes in, but is heavily limited by the 256 KB input file size limit (especially since checking for time limits usually involves large cases).
I was thinking we could have input generators for custom invocations. The system is already in place for hacks, so why not add it to invocations as well? Doesn't seem like it should be much of a hassle, does it? What are your opinions on this?
TL;DR Since input generators are already in place for hacks, why not add them for Custom Invocations as well?