[Pkg-scicomp-devel] Bug#439147: Bug#439147: libglpk0: some routines emit messages even with GLP_MSG_OFF set

[Peter T. Breuer]

Hi (I'm the bug reporter) ...

"Also sprach Rafael Laboissiere:"
> 
> Thanks for your comments.  I think that the GLPK manual should be adjusted
> in order to reflect the above.  At two different places, I read:
> 
>     Message level for terminal output:
>       GLP_MSG_OFF --- no output;
>       GLP_MSG_ERR --- error and warning messages only;
>       GLP_MSG_ON  --- normal output;
>       GLP_MSG_ALL --- full output (including informational messages).
> 
> Could you please propose a patch to be applied to the Debian package such
> that the bug report can be closed?

I had an additional message from Andrew saying that the bug would be
fixed in the next version. I presume the interchange is working its way
through the debian system (I think I cc'd my message to Andrew to
 439147 at bugs.debian.org, but I don't see a cc on his reply). Anyway,
here's the relevant part of Andrew's reply to me (with a fragment of my
message to him quoted incidentally):

  > It says "this paragraph describes all control parameters currently
  > used in the simplex solver".

  > and ... "message level for terminal output".

  > and .. "GLP_MSG_OFF - no output".

  > So one would naively conclude that that was definitive!

  You are right. This bug will be fixed in the next version of the
  package.

I gave a minor source analysis too. The message seems to happen because
it is conditional on LPX_K_MSGLEV >= 3, not on GPL_*, and 3 is the
default value. Setting GPL_* doesn't do anything for it.

The lpx_adv_basis() routine in which the message occurs is called by the
gpl_simplex() algorithm, not by me directly, when the presolver is
enabled, so I'd have no a priori reason to suspect that it was involved.

Trying to set the LPX_K_MSGLEV value directly on the lp problem involved
doesn't seem to help at present either, but I don't know why. I presume
the lp problem is given default LPX_ values internally when the
presolver is called.

It looks as though he just needs a more general cleanup of messages, but
I guess he could set the LPX_ values in the problem to the prevailing
GPL_ ones when the lpx_* routines are called from gpl_ ones.


Peter