Stephen Butler wrote on Tue, Dec 20, 2011 at 14:08:17 +0100:
>
> On Dec 20, 2011, at 12:51 , Eric S. Raymond wrote:
>
> > Still working on it as I await restoration of commit access. Here's the
> > 0.4 version. Note the two open questions and the buglet.
>
> [[[
> 1. Is it true that a replace operation on a directory *must*
> have a copyfrom source?
> ]]]
>
> No. A replace operation can create a "fresh" item without prior
> history. For example, the following node record is acceptable to
> 'svnadmin load', assuming there's already a file at ^/omega.
>
And this is how to produce such a thing:
svnadmin create r
svn checkout file://$PWD/r wc
cd wc
mkdir A; touch iota;
svn add A iota
svn commit -madd
svn up
do-replace() {
svn rm A iota
mkdir A; touch iota;
svn add A iota
}
do-replace
svn commit -mreplace
svnadmin dump --deltas --incremental -r2 ../r
Some other definitions that work are:
do-replace() {
svn rm A iota
mkdir iota; touch A;
svn add A iota
}
do-replace() {
svn rm A iota
svn commit -m rm
svn up
svn copy ^/A@1 A; svn copy ^/iota@1 iota;
}
And even:
do-replace() {
svn rm A iota
svn cp ^/@0 A; svn cp ^/@1 iota;
# well-defined; does not cause a loop
}
> [[[
> Node-path: omega
> Node-kind: file
> Node-action: replace
> Prop-content-length: 30
> Text-content-length: 9
> Text-content-md5: a2f613b8081a42cc0ff004f1191fe59e
> Text-content-sha1: 772ec56b2bff8db0bb4adb065ce9f212838a5143
> Content-length: 39
>
> K 3
> foo
> V 7
> new foo
> PROPS-END
> new text
>
>
> ]]]
>
