dtg3 / repostat Goto Github PK

Relates to #11.

Right now, we don't support data analysis of certain commits. Or atleast - we try to analyze them, and they return incorrect information that could skew our results. These appear to only be on merge commits.

For example - feeding Spoon-Knife to repostat gives us the following information for commit bdd3996:
FILES: 0
HUNKS: 0
LINES: 0
However, this commit changed 1 file, 3 hunks, and 26+2 lines.

More examples can be found by running repostat on faker, which has more merge commits, where 57e677f provides the output:
FILES: 0
HUNKS: 0
LINES: 0
Even though this commit changed 8 files, 7 hunks, and 830+1 lines.

Note that this issue is not on all merge commits. See commits e77b223 and b5dad5c. Repostat shows files, lines, and hunks > 0 for these merge commits. However, the files and hunks are off. Expecting output for files to be 2 and 3, respectively, yet repostat outputs 14 and 15.

Make each data field that will be stored a flag that can be set with OR operations to allow for greater configurability. For example, setFields( AUTHOR | COMMITTER | MSG ); would only read and store the author, committer, and commit message from the repository.

• Improve the CLI to be more robust in terms of what to collect and how to output the data
• Add threading to cut down on processing time
• Provide timing metrics to see if the tool scales
• D3 Reports

Output the information gathered from the repostat traversal to provide data drill down support in an interactive HTML report that render the charts (d3 is an option). It should be generic enough such that it could be customized to present different granularities of statistics as more information becomes available.

Some suggestions for useful charts:

• lines/files/hunks/commits (LFHC) over time
• LFHC within a branch
• correlations between LFHC within a commit

Does anyone have a preferred licence? I'm a fan of the MIT Licence since it's straight forward and easy to understand.

The \n character from a commit message is replaced with a space so that the row is preserved when viewed in JMP or excel. The same should be applied to a carriage return \r.

There isn't any freeing going on for the following things:

• Commits. Need to use git_commit_free.
• Trees. Need to use git_tree_free.
• Diffs. Need to use git_diff_free.
• Patchs. Need to use git_patch_free.

We don't want to be stuck in the same situation as before!

I'm stuck trying to use the git_diff struct that's returned when we call git_diff_tree_to_tree. Maybe someone better at C++ can figure it out?

I was testing by calling the git_diff_print function like so:

git_diff_print(diff, GIT_DIFF_FORMAT_PATCH, NULL, NULL);

Once this works we can convert it to a patch (which is already in there and seems to work ok even though printing segfaults).

Then we can get the following:

• Number of hunks per patch with git_patch_num_hunks.
• Number of additions and deletions with git_patch_line_stats.
• Number of files changed with git_diff_num_deltas (which already works without segfaults 😕).

Follow the merge commit info to identify named branches.