Re: [patch] Fix fortran/25031, multiple allocations

[Paul Thomas <paulthomas2 at wanadoo dot fr>]

Thomas,

> Because I would have had to handle the error conditions in the user
> program, as well, and I didn't want to bloat the user's program with a call
> to runtime_error() for each allocate statement.
>
>  
The overhead is actually quite small but you are quite right. I had 
written a TREE_SSA version that does suffer from the problem that you 
mention. It is intimately mixed up with a patch to deliver zero length 
arrays in response to request for negative lengths, in various 
conditions. I will separate the two tonight or tomorrow morning; having 
done that, I will review your patch. It could stand having one part of 
my patch incorporated; I replaced the gfc_ref* argument in gfc_allocate 
array by the gfc_expr*. I'll get the mods to you tomorrow morning.

Best regards

Paul