Re: write recursive call

[Brooks Moses <bmoses at stanford dot edu>]

Craig Powers wrote:
> Pierre-Matthieu anglade wrote:
> > So, If I fixe this, it should be disabled when strict f77 or F90 or
> > F95 is asked, enabled for F2003 and enabled by default?
>
> At the risk of saying something blindlingly obvious, note that recursive 
> I/O is not, generally, something you can detect at compile time, and I 
> don't recall whether I've seen that the runtime library has switchable 
> standardization.

A good point -- thus, it may not be possible to disable it when strict 
f95 compliance is requested.

However, it's worth noting that the standard does not require the 
compiler to diagnose this problem -- merely that the standard does not 
permit standard-conforming code to use recursive I/O.  Thus, it is 
perfectly permissable for a standard-conforming compiler to "correctly" 
compile non-standard code that has recursive I/O in it.

I would further note that the idea of having the runtime library aware 
of the "standardization" options is a very tricky thing to define, much 
less implement.  Suppose that I write a code A that has a write 
statement that calls a function in code B which has an internal write in 
it.  Further suppose that I compile code A with -std=f95 (which 
hypothetically disallows recursive I/O), and compile code B with 
-std=gnu (which hypothetically allows it), and then link them.  Should 
the recursive I/O call produce an error, or not?  What if I reverse the 
flags?  I don't think there's a clear or meaningful answer there.

- Brooks