[Patch, Fortran] PR 40881: warn for obsolescent features

[Tim Prince]

Manfred Schwarb wrote:

> The description in the manual of -std=legacy is:
> The `legacy' value is equivalent (to `gnu') but without the warnings for obsolete extensions.
> 
> Note the word "extensions", i.e. support for some old (vendor/compiler)
> extensions not supported by a particular standards version. That's how 
> I read this, at least.
> 
> For me there is a huge difference beween "extension" and 
> "language feature", and between "supported", "obsolescent" and "deleted".
> 
> As long as a feature is supported by a standard, you should not
> get warnings with default compiler settings, at least for things which
> are only a style matter and do not bear the danger of unintentional
> coding errors.
> 
> In the "legacy" compartment some obscure features are included which
> very few people want to use, e.g. real do loops or implicit
> integer/logical conversion.
> 

real do parameters were specified in f77.  You just made a distinction 
between vendor-specific extensions and former standard syntax, and then 
  appeared to have changed your mind.  Of course, they were obsoleted on 
account of their potential for unintentional errors.
I find warnings about statement functions annoying.  Those could be 
divided into legacy (usage not permitted by a standard, if the compiler 
can diagnose it) and obsolescent. I doubt any compiler has tried to make 
a distinction, except by treating the non-standard usage as an error.
I don't know whether the non-standard usage influenced the committee in 
designating statement functions for obsolescence. Since that was done, 
some changes have been made in the support of external use of functions 
in CONTAINS, so we have to learn whether PRIVATE applies if we want to 
get the functionality of statement functions.