Great change. However, I would propose eliminating the old print statements, since the same information is in the error message.

I have considered doing that, but I am not entirely comfortable with it because throw PSIEXCEPTION is not perfectly reliable when it comes to delivering the error message to the user. For example see #2997.

Until a reliable solution is found for the error messages getting swallowed up on some platforms, I am reluctant to remove the traditional error messages. But in principle I agree completely.

PS: In #2997 I added the band-aid workaround of printing to stderr for PSIO errors, since those are quite common due to disks getting full and whatnot, but doing that globally in PSIEXCEPTION itself seems inelegant.
Users who are not affected by the missing error message would get the entire stack trace, etc. twice, which would make the already quite noisy error exits worse.

The cause of missing error messages is likely the improper use of exit(). I have prepared a companion pull request to this one to get rid of them.

The cause of missing error messages is likely the improper use of exit(). I have prepared a companion pull request to this one to get rid of them.

I think there are two separate issues here. One is the improper use of exit() scattered around the codebase, but I still think that there is a separate issue here where not even throw PSIEXCEPTION(...) is delivering the error message, in certain environments. I think that would still be a problem even if every instance of exit() were to be eradicated from the repo.

In #2970, the part of PSIO that was raising the error was already using throw PSIEXCEPTION(...), it was not calling exit(). Unless removing all of the exit() calls elsewhere from the codebase magically changes how PSIEXCEPTION works, I cannot see how that would fix that.