Branko Čibej wrote on Sat, Nov 24, 2012 at 13:38:39 +0100: > On 24.11.2012 13:30, Daniel Shahaf wrote: > > Branko Čibej wrote on Sat, Nov 24, 2012 at 12:33:01 +0100: > >> I'm also considering requiring --force if one tries to use a revision > >> property name as a node property, and vice versa. > >> > > +0 > > > >> $ svn ps svn:barfoo x . > >> svn: E195011: 'svn:barfoo' is not a valid svn: property name; did you mean > >> 'svn:group'? > >> (To set the 'svn:barfoo' property, re-run with '--force'.) > >> > >> In this case the suggestion is clearly bogus. > > It seems the code should filter svn:user and svn:group from > > SVN_PROP_ALL_NODE_PROPS? > > That wouldn't make suggestion any less silly. :) >
That would make the suggestion of 'svn:group' not happen at all. > In fact, with the change I made just now (to not consider the prefix > when sorting the properties by similarity), trying to set "svn:barfoo" > results in the suggestion to use "svn:mergeinfo" instead, which is just > as wrong. > Maybe it shouldn't suggest anything? IIRC the code just does a qsort() and suggests the first option --- it doesn't check that the similarity score of that option exceeds some minimum threshold. > Regarding svn:group, svn:user and svn:unix-mode, these are defined in > svn_props.h as reserved, user-visible properties and I see no reason to > require --force to set them on /this/ level. This is purely a syntactic > filter; it would be wrong IMO to infuse any kind of awareness of > semantics into it. > I could go either way about whether it should ever suggest 'svn:mergeinfo'. > -- Brane > > -- > Branko Čibej > Director of Subversion | WANdisco | www.wandisco.com >
