Quincey Koziol

No - the execution environment (i.e. what happens when a floating-point exception occurs, etc) can be different.

>> Please give a concrete example. Actually, this issue needs to work the opposite way - the new way needs to prove that it will work in all existing cases....

> @qkoziol can you think of any example where a compiler doesn't know, at compile-time, `sizeof(long double)` for the target CPU? Agree here, yes.

> > Agree here, yes. > > OK great. (Indeed I don't see how a compiler could generate assembly for `long double` use if it didn't know how big it...

Those two changes are indeed related. :-) It was a bug to include pushing the function stack in the *_NOINIT macros, because they aren't supposed to perform actions that initialize...

> Those two changes are indeed related. :-) It was a bug to include pushing the function stack in the *_NOINIT macros, because they aren't supposed to perform actions that...

> I'm still unsure about making the H5FD API calls reentrant-only. Is this change needed to fix something or is it only to remove the overhead associated with the init...

> > > I'm still unsure about making the H5FD API calls reentrant-only. Is this change needed to fix something or is it only to remove the overhead associated with...

Following up on our discussion at the community meeting earlier today, I reviewed the difference between the FUNC_*_API and FUNC_*_API_REENTER macro pairs. It's not just the API context that is...

I've been working on the locking protocol, API context, and error reporting more over the weekend and the approach in this PR may need to change. I'll give it some...