[patch] fix wrong gimple code with ptr to forward vla

[Olivier Hainque]

Joseph S. Myers wrote:
> In any case excluding TYPENAME shouldn't make things worse since we have 
> no logic at present to ensure evaluation always happens at exactly the 
> right time; it simply happens to be done at about the right time in most 
> cases.

 OK. From this and previous comments (e.g. possible issues with PARM you
 pointed out), it seems to me that restricting the artificial TYPE_DECL
 creation to the FIELD and NORMAL cases would represent a reasonable
 compromise.

 Does this sound right to you ?

 I'll be happy to test that.

 Thanks for your feedback,

 Olivier