Make sure commits are connected

[purpleKarrot]

Looking at the tree view, there are lots of commits that are not connected to any branch. Probably some merge information is missing.

[dabrahams]

Please add links to examples so I can understand what you mean

[purpleKarrot]

Here is a picture taken from: https://bitbucket.org/boostorg/smart_ptr/commits/all/6c79ff4e07d1

[dabrahams]

This appears to be missing branch info rather than missing merge info. Is it possible this is due to missing "recurse" rule functionality?

[dabrahams]

Many of the disconnected commits seem to be "manufactured by cvs2svn." I notice there is special-case handling in svn2git for that case, but I haven't yet had a chance to analyze it. Do you think it might make sense to disable the special case?.

[dabrahams]

FYI, I tried disabling the special case for cvs2svn and it made zero noticeable difference, down to generating the exact same SHAs for the commits I checked.

[dabrahams]

So, it looks like this is happening when a branch is created and then (the part of it directed at a given repository) is deleted in one revision. For example:

$ svnadmin dump ~/src/boost/svnrepo -q --incremental -r 7699
Revision-number: 7699
Prop-content-length: 149
Content-length: 149

K 8
svn:date
V 27
2000-09-09T10:20:26.000000Z
K 7
svn:log
V 75
This commit was manufactured by cvs2svn to create branch
'unlabeled-1.4.6'.
PROPS-END

Node-path: branches/unlabeled-1.4.6
Node-kind: dir
Node-action: add
Node-copyfrom-rev: 7697
Node-copyfrom-path: trunk

Node-path: branches/unlabeled-1.4.6/boost/boost/detail/compressed_pair.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/detail/ob_call_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/detail/ob_compressed_pair.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/detail/ob_type_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/detail/type_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/array.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/array_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/call_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/cast.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/compose.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/compressed_pair.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/config.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/cstdint.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/functional.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/integer.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/integer_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/iterator.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/min_rand.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/nondet_random.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/operators.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/progress.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/random.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/rational.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/smart_ptr.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/stdint.h
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/timer.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/type_traits.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/boost/utility.hpp
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/c++boost.gif
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/index.htm
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/libraries.htm
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/libs
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/more
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/people
Node-action: delete

Node-path: branches/unlabeled-1.4.6/boost/people.htm
Node-action: delete

Node-path: branches/unlabeled-1.4.6/CVSROOT
Node-action: delete

IMO in these cases the branch commit should be dropped from those repositories where it is deleted.

[dabrahams]

OK, that handled some of it. However, there are still lots of dangling branches. Please have a look and see what you can conclude!

[dabrahams]

One theory I had was that the source revision of the svn copy sometimes doesn't exist in a given Git repo, because no change in that revision fell into the repo in question. Maybe the code doesn't deal with that case?

[dabrahams]

My theory above seems to be wrong, even though the logic in that function is a bit questionable. Why does it return 0 instead of -1 when no commit is found here?

It looks like 4487ba0 doesn't quite deal with all the cases it was meant to address; see this commit for example. I imagine that the branch never registers as fully deleted in that case, but nonetheless it has no files in it and should be dropped.

[dabrahams]

Then there are cases like this one, where earlier commits in sandbox/boost/smart_ptr should really stay in the sandbox Git repo as they're part of an experimental policy-based smart pointer library. Instead they're showing up in the smart_ptr Git repo, which causes the commit that moves them to sandbox/boost/policy_ptr and sandbox/libs/policy_ptr to show up as a parent-less commit with no files in it.

[dabrahams]

With the exception of boostorg/smart_ptr@aa855dd48552, which looks like it might genuinely be a disconnected commit, all the commits in that repo now appear to be connected. Please verify and reopen if that's not so.