bad distributed roots specification causes odd error

Reported in EP1660, if decomposeParDict has distributed true, but has an empty roots () entry, we get very odd error messages.

@martin.werther

assigned to @mark

This should not really work anyhow (ie, should be fatal or some other recourse for this type of input), but it seems that the logic in argList causes the master to think that it is distributed, but sends across nroots to the sub-processes, which means that they then think they are running non-distributed. The cryptic error is actually caused by the fileOperation handler since the scattering of fileNames is not mismatched entirely (ie, master sends nothing, but the sub-procs are trying to receive it.

Can call it a misfeature or a bug, depending.

changed the description

After first discussion with @martin.werther - makes sense to make the handling of empty roots more robust (send/receive a bool instead of the roots size). However, still open if we wish to make the overall handling more robust, emit a warning or whatever else.

changed the description

mentioned in commit 504e8b9c33c7dedb177c31022151db86303f278f

mentioned in commit f815a12b

mentioned in commit f4ccdec8

mentioned in commit 6168f960

mentioned in commit 1ead1759

closed via commit f815a12b

mentioned in commit 5a121119

mentioned in merge request !506 (closed)