Re: "restrict" implementation bug ?

[Dan Nicolaescu <dann at godzilla dot ICS dot UCI dot EDU>]

"Joseph S. Myers" <jsm28@cam.ac.uk> writes:

  > On Sun, 1 Apr 2001, Dan Nicolaescu wrote:
  > 
  > > According to http://wwwold.dkuug.dk/JTC1/SC22/WG14/www/docs/n897.pdf  
  > > (pointed to by a link from http://gcc.gnu.org/readings.html)
  > > (and BTW there is a n937.pdf file in the same place that seems to be
  > > a newer version)
  > > global variables with a restrict qualifier should act "as if it were
  > > declared as an array". 
  > > 
  > > It seems that there is a bug when using both a restricted global var
  > > and a pointer obtained from "malloc" call. 
  > > As shown in the example bellow when using either of them individualy
  > > "restrict" works correctly. 
  > 
  > Can you please send actual bug reports (as opposed to questions) for
  > non-bootstrap bugs to our GNATS bug tracking database, rather than to the
  > gcc list, so they are properly tracked for when someone looks at the
  > problem, quite likely months later?  

I did that, the bug ID is c/2462.
I wasn't sure that this was a bug in the implementation, that is why I
was asking a question. Now I am quite convinced it is. (or a missing
feature, given that "restrict" is a pure optimization with no effects
on correctness). 


  > Also, when dealing with subtle areas of the language, it would be helpful
  > to reference the actual standard wording (with subclause and paragraph
  > numbers) rather than the draft Rationale.  Especially here, where I don't
  > think the Rationale has been effectively updated for the changes in
  > "restrict" between the public C9X drafts and the FDIS and final standard.

Sorry, but I don't have a copy of the standard, I didn't even know
that there is a difference between the Rationale, FDIS and the standard,
given that yesterday was the first time when I read the Rationale...


  > I think category "c", class "pessimizes-code" would be appropriate for the
  > PR.  Note 6.7.3.1 paragraph 6 and 6.7.3 paragraph 7: a conforming C99
  > implementation could entirely ignore "restrict" apart from checking type
  > compatibility rules.

Thanks for the advice, I did that.