Recent version of mkpsxiso core dumps when creating certain ISOs

[lameguy64]

Recent version of mkpsxiso core dumps when creating certain ISOs

[marco-calautti]

By having a quick look at the code, I believe the problem lies in the fact that mkpsxiso launches a thread (std::launch::async), for each sector for the XA file, to generate edc/ecc data, for each sector, which for very big files this would exhaust the maximum number of thread that can be run. In fact, when running mkpsxiso with gdb, I can see a lot of threads spawn.

I do not see the point of using threading for such a very simple task like generating ecc/edc, considering psx bins are not going to be very large anyway.

-EDIT- in fact, by removing all the async non-sense, everything works fine as expected. If using threads is really important, then I would suggest limiting the amount of tasks you spawn. My laptop has 32GB RAM, and this was not enough to pack a 434 MB XA file.

[marco-calautti]

Probably you did not spot this issue, because the std library uses a thread pool in Windows. But on Linux, each std::async actually spawns a new thread. See https://stackoverflow.com/questions/29622179/async-uses-a-threadpool

[CookiePLMonster]

The recent rewrite has added a lot of changed code that I fear has not been properly tested, and thus regressed the tool, rather than improving it.

The previous version was not even capable of generating proper checksums so any and all files built through it were technically broken. I'm sorry that a regression crept in, but a rewrite was available for testing for a while before it got merged to the mainline repository and all reported issues were fixed. I'm sure you realize you can't fix bugs nobody spots, and it's not realistic to achieve 100% coverage of all possible discs.

Probably you did not spot this issue, because the std library uses a thread pool in Windows. But on Linux, each std::async actually spawns a new thread

That's quite disappointing. In this case I think the idea of using a thread pool as in the PR is fine.

[CookiePLMonster]

That's long fixed by #37.