[gfortran] patch PR 20788

[François-Xavier Coudert]

> I think it's pointless to add avoid_mmap to fd_to_stream's interface, and
> would prefer if you explicitly checked for the affected three units in that
> function.  Or is it likely that there will be other cases where this is necessary?

There are cases where a simple check on the fd number (is that what you 
suggest) is not enough: if someone wants to dup the file descriptor for 
stdin/stdout/stderr in the library in the future, or if we discover other 
cases where mmap should be avoided.

In brief: I think a more general mechanism is not something we should 
replace with an _ad hoc_ check.

> The patch is of course correct, and you can leave it installed if you don't
> agree with me.

I'm sorry it got applied (I thought I had checked which files would be 
commited, but it looks like monday will always be monday).

FX