Re: svn commit: r1592724 - in /subversion/trunk/subversion: include/svn_client.h svn/propget-cmd.c tests/cmdline/prop_tests.py tests/cmdline/svnadmin_tests.py

Julian Foad wrote on Mon, May 19, 2014 at 12:15:00 +0100:
> Daniel Shahaf wrote:
>
> > URL: http://svn.apache.org/r1592724
> > Log:
> > Follow-up to r3265 (r843339): In 'svn propget -r', error out on
> > non-existing properties.
>
> This makes 'propget --revprop' inconsistent with 'propget [versioned
> prop]' which still prints nothing. Did you intend to change both
> forms, for consistency with 'svnlook'?
>
> Previously, 'svn propget' would print nothing for a nonexistent prop
> (either versioned or revprop), while 'svnlook propget' would issue an
> error for a non-existent prop (either versioned or revprop), which was
> a bit more consistent than what we now have.

I've updated issue #4505 with answers to the concerns raised in the thread:

http://subversion.tigris.org/issues/show_bug.cgi?id=4505#desc2

To summarize, the answers are: effect on XML mode is the same as on non-XML
mode; there is no regression in -R mode because r1592724 only changed --revprop
mode, which ignores -R; and there are no other inconsistencies beyond 'svn
propget nodeprop' not erroring.

At this point we have two options: either revert the change (so 'svn
propget' doesn't error either in revprop mode or in nodeprop mode; this
is the 1.8 behaviour) or make 'svn propget nodeprop' (without -R) error
out too.

I'm attaching a patch implementing the latter.

Thoughts?

Daniel

---
Personally, I lean towards applying the patch.  Like any change, it may break
users who depend on the current semantics, but it's also likely to convert
silent failure modes to noisy failures for users who didn't realize 'propget
svn:lgmsg' (sic) doesn't error.  For scripted users, the current behaviour
means non-existent properties are indistinguishable from empty-valued ones
(unless 'proplist' is used to query existence).