[Patch, fortran] PR29786 - [4.1/4.2/4.3 Regression] Initialization of overlapping variables: Not implemented

[Brooks Moses]

Paul Richard Thomas wrote:
> Brooks, Steve, FX...
>> Getting it to regtest on 4.2 may be a bit of a trick, since I haven't
>> backported any of the simplify_transfer stuff (or its middle-end
>> prerequisites) to 4.2 yet!  :)
> 
> It was the regression wrt g77 that led me to propose backporting to
> 4.2.  I think that the voices against the backport are sufficiently
> convincing that they should have the day.

Fair enough.  :)  I was in the process of arguing with FX with regards 
to whether or not this is actually an "invasive" change [1] and what 
that meant for backporting it, but then I realized that regardless of 
whether it's invasive, (a) he's certainly right that it's not tested 
enough to be reliable and silent wrong-code bugs are much worse than 
"not implemented" errors, and (b) by the time we get everything settled 
out, it's likely to get lots more invasive.

Maybe once things get settled on trunk, we can change those error 
messages to say "Not implemented until version 4.3."  :)

Thanks for the thoughtful arguments!

- Brooks


[1] I don't think it's invasive despite the volume of code added, 
because in a reasonable backport most of the code would only be called 
in cases where we currently issue a "not implemented" error.