Re: Valarrary regression (was: Re: [PATCH] HP/SGI extensions...)

[Gabriel Dos Reis <gdr at codesourcery dot com>]

Paolo Carlini <pcarlini@unitus.it> writes:

| Paolo Carlini wrote:
| 
| > the reason why I'm CC-ing Gaby too is that today, after a dozen of patches to
| > the C++ front-end by Nathan Sidwell, I'm seeing a couple of regressions in
| > valarray tests, in particular, binary_closure.cc and valarray.cc (of course both
| > with and *without* the present patch).
| 
| Finally, it seems that someone has reproduced the problem, on a different platform
| too:
| 
|     http://gcc.gnu.org/ml/gcc-testresults/2002-01/msg00030.html
| 
| Gaby, what do you think?

I think that the Standard and I are being caught by GCC/g++ being more
and more standard conformant;  on one hand that is a good thing :-)

| I have real trouble understading what's going on.

That is an old issue in GCC for quite a long time: Historically
access check in the front-end wasn't right.  On the other hand, there
are access violations in the specification of valarray.  I made a
proposal to fix them, but the committee was reluctant.  Lately, Robert
Klarer arrived at the same conclusion as I and proposed to adopt half
of my proposal, but to date there is no concrete decision.  Benjamin
reported failures in some testcases, I thought I fixed them but
clearly I fixed only a tiny part.  Working on another round of
patches. 

| If we have to prepare a compact testcase for Nathan in order for him to fix the
| front-end I don't even know where to begin with: notwithstanding your tutorials
| around the world, to date template-metaprogramming-based code is still mostly
| black-magic for me... :-(

Hmmm.  At Oxford, I thought I convinced you that template metaprogramming isn't
black-magic :-)

-- Gaby