Re: [ublas] broken coordinate_matrix::sort with gcc 4.7 (was: patches for #7297, #7296, #6514, #6511 on trunk - please verify)

Subject: Re: [ublas] broken coordinate_matrix::sort with gcc 4.7 (was: patches for #7297, #7296, #6514, #6511 on trunk - please verify)
From: Gunter Winkler (guwi17_at_[hidden])
Date: 2012-09-04 19:34:30

Hello,

thanks for the valuable comments.

Am Tuesday 04 September 2012 schrieb Ungermann, Jörn:
> Hi,
> 
> test_inplace_solve fails mostly, because it uses a coordinate_matrix.
> *Some* errors are be unrelated, though.
> I had a closer look at the failures:
> 
> a) The includes in lines 10-12 of test_inplace_solve.cc I supplied
> are unnecessary and break on some (for me exotic) systems. (They
> were originally included for timing and I forgot to remove them.
> Sorry).
> b) The const int n = 10; in line 87 is superfluous, which produces
> some warnings.

I removed these headers and the const.

> c) The vacpp testcases fail, because it cannot instantiate a
> mapped_vector_of_vector on that system. There does not seem to be any
> other testcase for this type. std::map seems to not provide a
> data_value_type member type.
> One might want to add a simple testcase using a
> mapped_vector_of_vector for verification of this hypothesis before
> proceeding. I have not access to the vacpp compiler.

I fixed two issues in mapped_vector_of_vector - maybe they caused the 
trouble.

> d) Some systems fail, because there is no triangular.hpp.new. I
> cannot fathom, how this is correct on some systems, but not on
> others. The original testcase provided by me had this error, but
> according to the bug entry, it was fixed, so it should be OK on all
> of them...

The revisions are too old. We have to wait until the tests run again.

> d) The remaining errors fail due to the coordinate_matrix problem.

I have split some of the failing tests.

mfg
Gunter