Re: sqp failures

[Julien Bect]

On 09/11/2013 22:57, Mike Miller wrote:

I see that both of the bug reports you are looking at are older than this bug report and the change you bisected to. Are you sure you're tracing the same error that those reports are about, or are you seeing a new error due to the change to the new glpk API?

The two bug reports that I am pointing at clearly describe the same symptoms, but...

https://savannah.gnu.org/bugs/?32717 : this bug report merely indicate that 1) an error should be raised earlier when qp fails with INFO=6, instead of letting sqp fail later with a cryptic error message (this is easy to fix, but hasn't be done yet) and 2) that sqp should be able to recover from qp failures (much more vague and harder to fix).

https://savannah.gnu.org/bugs/?38875 : this bug report provides an example, that currently exhibits exactly the same failure (sqp: operator *: nonconformant arguments (op1 is 7x14, op2 is 0x0) at line 450) that I see in all my examples. But you're right, the bug report is older, the error message is different in 3.6.4, so perhaps the real cause of the problem is different.

If the error you're seeing is something that used to work in 3.6.4, then that should be marked as a regression.

Yes, I will file a bug report as soon I can provide at least one simple example of what I mean...