I think that this tool still has a lot of potential, and doesn't deserve to be abandoned, so I decided to rewrite traverse_xiso, the most difficult function to read and understand in the code, and split it into two pieces to make it easier to understand.

Judging from the changelog on top of the file, it appears that it once used recursion, but it has been rewritten as the spaghetti code is it now to avoid sporadic stack overflows with highly unoptimized ISOs written by some tools.

It looks like whomever is making xiso creation tools out there
has never heard of a binary tree and is sticking every
directory entry off the right subnode (at least on all the iso's
I've found so far). This causes extremely deep recursion for
iso's with lots of files (and also causes these iso's, when
burned to DVD, to behave as a linked list for file lookups, thus
providing worst case lookup performance at all times).

Not only do I find this annoying and extremely bad programming,
I've noticed that it is causing sporadic stack overflows with
my (formerly) otherwise good tree traversal code.

In order to combat such bad implementations, I've re-implemented
the traverse_xiso() routine to get rid of any non-directory
recursion. Additionally, I've made a few extra tweaks to
conserve even more memory. I can see now that I'm going to need
to write xiso creation as well and do it right. (rev to v1.5 beta)
NOT FOR RELEASE

I honestly don't think the cause of the overflows was the recursion: each iteration uses very little memory and the same ISOs are processed on the OG Xbox anyways. Probably it was some other bug that manifested itself only in some cases.

I tried to keep the memory usage as low as possible, by freeing up unused memory early, but it will probably use more memory than before. I guess that's the tradeoff for code legibility.

Anyways, the code has been tested to work and produce functionally identical ISOs. They are not hash-identical to the previous ones because of the order the files are processed. As as side note, it looks like that now it generates node trees identical to retail, so I guess it is more accurate? Anyways, if someone knows what tools produce such linked list ISOs, it would be awesome to test them and see if the stack overflow problem is fixed.

Of course it needs extensive testing, but I think that with traverse_xiso rewritten, it would be much easier to update the tool and/or solve bugs.

The next step would be to review all the open issues, and see if they still apply. In particular, I'm using 64-bit builds and they produce files identical to 32-bit ones, so maybe this particular issue is solved.