[PPL-devel] ppl 1.0pre24 make check

[Roberto Bagnara]

Jack Howarth wrote:
>    You might consider creating alternative 'make check'
> for ppl 1.0 which doesn't take as long to run. Compared
> to ppl 0.9, the 1.0 make check takes forever. On a 3 GB,
> MacBook Pro Core2Duo 2.33GHz machine running Leopard,
> the make check takes at least 3 hours

Hi Jack,

we have been thinking about that.  There are two things
to be considered:

1) The PPL is a rather complex library that provides
    several numerical abstractions, several of which
    in the form of C++ templates.  While it is true
    that our testsuite is huge, no part of it is useless.
    Indeed, it happens all the time that a bug affects
    only one or two tests of the thousands that are
    run by `make check'.

2) It is true that our `make check' may be perceived
    as too heavy by some users.

So we have come up with the following compromise:

a) we have added a configure option --enable-check=X
    where X can be `thorough' or `quick';

b) the default for the CVS version is `thorough',
    so that developers and our build machines
    play safely;

c) the default for the tarballs created with `make dist'
    is `quick' so that the users not particularly concerned
    with safety are not annoyed;

d) in README.configure we will encourage all users to
    configure with --enable-check=thorough.

What is missing is a decision of what to put in the
`quick' checks.  We need more time to thing about that
(we will certainly cover all what is used by CLooG though).

Anyway, do you think the above is an acceptable compromise?

> at which time
> it hits the memory1 test which seems to go into VM
> and execute extremely slowly. All of the make check
> passes up to that point.

This is not normal.  On my machine (which is less powerful
than yours, both in terms of CPU power and of physical memory)
I get:

$ time memory1

real    0m0.059s
user    0m0.040s
sys     0m0.022s

Can you please investigate your observation further?
Alternatively, would it be possible to have access to
the machine you mention for just an hour or so?
All the best,

    Roberto

-- 
Prof. Roberto Bagnara
Computer Science Group
Department of Mathematics, University of Parma, Italy
http://www.cs.unipr.it/~bagnara/
mailto:bagnara at cs.unipr.it