"`port -pf upgrade outdated` is incredibly unsafe" ??

[René J.V. Bertin]

On Thursday May 28 2015 19:54:27 Kurt Pfeifle wrote:

> I've seen the upgrade of 200+ packages stopping just because one of the
> first five package upgrade fails.

So have I, but I rarely if ever use even -p when I do launch a bunch of updates. I rather tend to isolate the ones I know will be time-consuming, and yes, I do use -n quite often nowadays in order to update ports without updating their dependencies (if I even updated their port directories).

I would be nice however if there were a flag that works more like make's -k flag. It would do something like

- keep working until an upgrade fails, building only ports that don't depend *) on a failed port
- add the failed port to a list of ports that failed until now
- skip to the next port that needs updating, and try to build it if it doesn't depend *) on a failed port
- repeat

*) with -k, this is a recursive depend; with -kf (or some other variation) this is a direct depend

It should be safe to build ports that do not depend directly on a failed port, but it seems not impossible that the ABI-changing effect of the update of an indirect dependency permeates to a direct dependency. At (probably) worst that would mean that you have to rebuild some ports again after getting the failed port(s) to build, so I think it should be fine to allow a less strict skipping mechanism in case of upgrade failure.

R.