=========== Description =========== This is a RFC for a patch that allows to enable filters in the repositories created/managed by git-svn. ================== Example scenario ================== The usage scenario I’m facing is the migration of a large SVN repository to a git repository, with big files being transparently handled through git-lfs To recap, these migrations are usually handled through a transitioning repo created by git-svn, which is then pushed to a remote git server; this server acts as a read-only mirror for a while, until the main SVN server is shut down. For large repos, the process can take several weeks to adjust the migration scripts, branch mapping, authorship, etc., and it’s also common to update the mirror incrementally. Without this patch ================== Without this patch, the migration needs to be setup with a pipeline with: * incremental git svn fetch, * bfg-repo-cleaner with —convert-to-git-lfs option (which puts invalid .gitattributes files, see https://github.com/rtyley/bfg-repo-cleaner/issues/143) * git-filter-branch (to put in place the correct .gitattributes files) A minimal change in the pipeline results in unrelated branches after the rebases. Each deviation from the svn sources must be committed back or saved as patch and reapplied after the rebase. With this patch =============== Everything is much simpler: filters can be triggered during the fetch, so that git-lfs already runs while the repository is fetched. At the end of the process, the local git repository is already in its final form, and no further git-filter-branch/rebases is required. The transitioning git repo can diverge partially from the svn source thanks to git svn rebase. Details ======= In the current implementation git-svn uses the hash-object command with the --no-filters option. There is no gain in removing that option because the git-svn uses temporary file names. In this patch hash-object accepts a new format for the streaming input, files and paths, tab separated. git-svn can provide the real path to hash-object and thanks to that we can enable the filters. Status ====== The patch is in RFC, no new test where written (yet), just runned for regressions (https://travis-ci.org/naufraghi/git/builds/134170350). I'd like to gain some feedback before spending more time on the feature, mainly: * feature comments, * coding comments (neither Perl or C are my main languages), * user interface (option names, config placements) comments. Best, Matteo Matteo Bertini (4): hash-object.c: Allow distinct file/path in stdin mode too. Git.pm: Add $path and $enable_filters arguments to hash_and_insert_object. SVN/Fetcher.pm: Add svn-remote.<id>.enable-filters to enable the filters. git-svn.perl: Add git svn init --enable-filters option. builtin/hash-object.c | 29 +++++++++++++++++++++++++++-- git-svn.perl | 4 ++++ perl/Git.pm | 19 +++++++++++++------ perl/Git/SVN/Fetcher.pm | 16 ++++++++++++---- 4 files changed, 56 insertions(+), 12 deletions(-) -- 2.9.0.rc0.39.gb9f310b.dirty -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
