About internal fs: what if we exchange HDD modules (if possible, I don't know if whole firmware goes to RAM and if it is possible to replace anything from there on the fly) to support something normal, and different than APA and PFS like i.e GPT and i.e F2FS. Isn't this get rid of all PFS limitations? Of course it is titanic work but it is just an idea.
Such modules are provided by the software itself, hence the capabilities of any software is determined by its own design. The ROM of the PS2 is mostly just for booting up.
Yes, it is possible to drop APA & PFS entirely, but that will be at the cost of no backward compatibility. Neither with existing homebrew software nor with Sony games.
PFS is likely good to use. It's a 64-bit filesystem, similar to the types used with *nix.
The problem is APA. But I must say that it's still somewhat meeting our needs. It's only going to be truly left behind, should we (somehow) desire to move beyond 2TB.
Similar to external storage. Forgetting of MBR with FATs and focusing on i.e exFAT or F2FS (without partition table). There was some sort of experiment in OPL few years ago with EXT2.
You are the experts, but for me, from raw point of view, this could free us from ancient limitations.
MBR is okay IMO. It's because APA is different from MBR, in very bad ways.
In APA, each application will have a parent partition, and may have a child partition. Each of them may have 1 main partition, as well as up to 64 sub-partitions. The maximum partition size on a disk is 1/32 of the disk's capacity, limited to 32GB.
So if you have 500 titles, for example, you could have something like 1000 partitions. We also try to save space by mixing and matching partitions (i.e. for a 4.1GB game, it might consume 2 partitions, one 4GB and one 128MB).
It's poor for large disks because it has no partition table, requiring the driver to scan through all partitions. Each partition is also very spread out, so the HDD's seek time gets ramped up (if there are 1000 partitions, 5ms will become 5000ms). It's likely why using a SSD will magically solve all its problems.
Also because there's no partition table, the cache system of APA involves storing the partition's data in memory. That's 1KB per partition, which results in high memory usage. Since that's not possible, so the cache remains small and hence becomes insufficient for large disks.
When we comes to MBR, you can have up to 4 partitions, but nothing really forces you to have a specific number of partitions. You can have any number of applications on a partition, up to the limits of the filesystem.
I was skeptical about the capabilities of E2OPL (if I remember right) because I did try to make a micro FAT32 driver, and failed to make it work very well (the old commits are there).
But because I had a lot to deal with (in real life) during that period of time, I did not look at what its author did and did not involve myself with his progress. So I don't know for myself, whether it really worked as well as claimed, which is why I still have my doubts.