On Mon, 2011-05-16 at 17:34 +0200, Daniel Shahaf wrote: > Julian Foad wrote on Mon, May 16, 2011 at 15:49:52 +0100: > > Daniel Shahaf wrote: > > > Daniel Shahaf wrote on Mon, May 16, 2011 at 15:36:25 +0200: > > > > [[[ > > > > % wc-format.py > > > > .: 11 > > > > > > It said '28' originally. > > > > > > > % svn --version -q > > > > 1.5.1 > > > > % svn st -q > > > > When I try with 1.5.9 and 1.6.16, they say "svn: warning: '.' is not a > > working copy" which is different but is still misleading. > > > > Yes, I copied the wrong part of my shell session. > > > > > % echo 11 > .svn/format > > > > % echo 11 > .svn/entries > > > > % svn st -q > > > > svn: This client is too old to work with working copy '.'. You need > > > > to get a newer Subversion client, or to downgrade this working copy. > > > > See http://subversion.tigris.org/faq.html#working-copy-format-change > > > > The URL is now > > <http://subversion.apache.org/faq.html#working-copy-format-change> and > > Fixed on trunk.
Thanks. > > we should update the FAQ answer given there. > > > > How? To mention that 1.7 doesn't autoupgrade and doesn't provide > a downgrade script; anything else? That sounds fine. > > > > for details. > > > > zsh: exit 1 svn st -q > > > > % > > > > ]]] > > > > > > > > So, RFC: > > > > > > > > * wc-ng working copies shall contain an "entries" file (and/or > > > > a "format" file) containing the text "11\n", inside all .svn > > > > folders they have (if any). > > > > Should be SVN_WC__VERSION (curently 28), not 11. > > > > >From IRC: > > 18:20:39 <@julianf> Curious: why on earth would we not want to provide the > correct value? > 18:21:38 <@Bert> julianf: We would have to maintain it later > 18:21:50 <@danielsh> julianf: because we'd have to create/maintain it at the > right value, > 18:21:52 <@Bert> (not sure if it is a good reason, but it is a reason) > 18:21:55 <@danielsh> and what would need it? > 18:21:57 <@julianf> OK > 18:22:01 <@danielsh> wc1 clients need ">10" > 18:22:05 <@danielsh> wc-ng clients use sqlite > 18:22:18 <@danielsh> wc-ng-ng clients read "11" as an indication of "this is > sqlite backed working copy" > > Julian/all, I'll assume "11" is okay if I don't hear otherwise. > > > When I try with just a 'format' file it doesn't work. It has to have > > the 'entries' file (as well or instead) to produce the useful > > diagnostic. > > > > I don't remember which (or both) of them do old clients look for. > Having both of them is safe though. > > > This only helps when the target of "svn st" is the WC root folder > > (except in some cases it works for an immediate child of the root). > > > > True. > > > Also I would guess there are some other tools (such as Subversion > > plug-ins for IDEs) that would give a more helpful diagnostic if they > > find a WC version number there. I wonder if such tools would > > > > > 18:13:12 <@julianf> I meant, "wonder if such tools would like to find a > > 'format' file, so we ought > > to provide both 'entries' and 'format'." > > I see no harm in providing both files. As to old tools that request our > format number via the API, we'll have our API's implemented (the > internal API's too) by looking in wc.db. > > > Given all that, and despite only working for the WC root, I think that > > is a useful enough result to be worth doing. I can't think of any > > practical objection to having these one or two extra files per WC if it > > helps smooth the transition for users. > > > > Thanks. Working on a patch... > > (I'll create a workqueue 'file-install-string' operation for the > svn_wc_upgrade() use case, hence the delay) Sounds good to me. - Julian
