Entering a "no entry" restriction in JOSM with more than one "from" way

The JOSM validator have a strict interpretation of the tagging schemes. This is nice for an editor as it might indicate that somthing is wrong. An application should use a loose interpretiation of the schemes as to fix a lot of the errors that editors make.

You are not guarantied that every application is able to interpret the tags the same way you do (or as the wiki sais). It may be that a routing application only considers one of the "from" way and disregard the others and this is what the JOSM validator is warning you about. This may be something that have to be fixed in the application however you have been warned by the validator.

The publicly editable wiki may also be wrong. It may be that applications and editors have different standards then the wiki have. Generaly the wiki have more thought behind, however it is not always the case that the wiki have all the answers.

The wiki is contradictory. If you look at:

http://wiki.openstreetmap.org/wiki/Relation:restriction#Members

it says you have one from way, one to way, then one via node or one or more via ways. The road signs table on the other hand contradict this just for no entry signs:

http://wiki.openstreetmap.org/wiki/Relation:restriction#Road_signs

The no entry ones were only added to the wiki on 11th October last year, and I can't remember seeing discussion about them. I personally would have suggested sticking with the same definition of the relation, and if necessary creating one for each "from" way.

As a reminder, no entry restrictions aren't needed for one way streets.

JOSM indeed has a strict validator. For good reasons.
I would disagree that application should be any less strict about interpreting data. That is what happened with html and the compatibility of websites with different browsers was a nightmare (with no more than 10 browsers). Because of this current browsers are slower and bigger than they could be. That is the reason why incorrectly formatted xml ought to be reported as such and no fixing attempts should be made.

History has shown already that standards should not be taken lightly when it comes to data for automated processing. It sure is possible to overcome most of the inconsistencies on big computers (eg. servers), not so much in mobile phones and other limited performance devices.

In short: listen to the most strict validator you can find. :)