On Tue, 17 Aug 2021, Sven Schreiber wrote:
Am 17.08.2021 um 12:07 schrieb Riccardo (Jack) Lucchetti:
>
> I realised that one of the new features was not documented in the xml
> file, so I added a paragraph. If someone could take a look to what's
> in git, that'd be great.
Looks good to me. We only have to remember to adjust the 1.25 seconds
value in this reference whenever it is changed in the source ;-)
I have just also added the notes to the funcerr and errorif() docs that
in MPI mode execution will not be stopped. To be precise, I added this
sentence: "An exception is the special MPI mode for parallelized program
execution, where only the associated string is printed."
Not sure whether gretl's actual behavior already matches this intention,
though, but I guess it's coming soon.
It's more complicated than I thought. I don't think we should make
any changes to the code until we've thought through the whole thing
carefully. As things stand the user is probably not going to see a
string set by funcerr or errorif under MPI. That could be arranged,
but at a cost in terms of relaying MPI output in real time (as when
a time-consuming MPI block is run in the GUI program). There's a
trade-off here.
Sorry, I don't have time to explain this properly right now but we
can get to it after releasing 2021c. For now it might be better to
be even less committal in the doc, just saying that funcerr/errorif
are unlikely to work as expected under MPI. ("Unlikely" because it
depends on the exact context: running via the gretl GUI, or
gretlcli, or gretlmpi.)
Allin