On Thu, Jun 16, 2011 at 10:38:12PM +0200, Johan Corveleyn wrote: > Are you sure? Maybe there are different definitions of how patience > diff works (it seems to be rather undocumented :-)). I'm not really > into the details, but from what I've read on this thread on the git > mailinglist [1], I thought it worked recursively, more or less like > this: > > (1) Find the unique lines, and perform an LCS on them. > > (2) For each in-between section, again perform step (1) (finding > unique lines for this section, which may turn up new unique lines, > unique to this section). > > Repeat this recursively, until there are no more unique lines in a > section. Then fall back to the regular diff algorithm (a "standard > minimal diff" on the remaining section).
As far as I can tell the recursion was dropped later: http://bramcohen.livejournal.com/73318.html "I've previously described it with the ordering a bit different and a recursive step at the end..."
