Added support for pulling only a subset of orphan objects

[tomasherman]

I'm not sure if you are interested in this functionality, but I needed to be able to specify only a subset of files to be downloaded.

Usage is

git fat pull --match 'python_regex_here'

and it should only download and restore files matching the regex

[jedbrown]

Thanks for this; the functionality is certainly important. Specifying files using a Python regex is unconventional for a Git user. How about passing a git ls-files pattern or, for more precise users, a list of files on stdin? Also, a basic test would be appreciated. (I'm planning to convert the few tests to use sharness. The current script is cruddy, but at least it shows something.)

[tomasherman]

Yeah, using git ls-files makes more sense. I will use that instead and check out the sharness too.

Maybe the api could be: git fat pull 'pattern1' 'pattern2' etc? What do you think?

[bilderbuchi]

I'm planning to convert the few tests to use sharness

git-fat being primarily written in python, is there any particular reason for not choosing a python testing framework like py.test or nose?

re: API: just my 2 cents here, I think it makes very much sense to match the git API as closely as possible to minimize the potential for surprises and to minimize friction losses when learning git-fat.

[tomasherman]

I rewrote the thing using ls-files. I'm not really sure how to test the behavior. I can add some unit tests, though, if you decide to use python testing.

[jedbrown]

@bilderbuchi We had this discussion on gitifyhg, which is migrating from py.test and sh to sharness. Since the primary interface is command line, python testing ended up being "obfuscated shell scripting", with more effort involved to translate between discovering a problem and creating a test to demonstrate it. Also, since sharness is git.git's testing system, it would make eventual upstreaming of git-fat easier.

[tomasherman]

so basically we are going to assume there will be:

git fat pull followed by - and -- prefixed values followed by patterns

is that what you want? Yeah i will squish the commits when we have something you like :)

[jedbrown]

I wouldn't try to interpret anything before the stand-alone -- since that makes the intent unambiguous. Git has many commands that require -- in general, though some of them will try without -- in simple cases.

[bilderbuchi]

See e.g. man git add (cause it also can select files based upon a pattern) as to how the argument structure looks like. Also note that -- is optional and not always needed, as jed points out.

re sharness/py.test: thanks, that makes sense.

[tomasherman]

Oh ok, i think i understand now (sorry i'm quite a noob ;))

[tomasherman]

3rd time's the charm, maybe? :)

[tomasherman]

I squashed all the commits together too.

[jedbrown]

I merged as c23422388b975f13867457c86c78361dfdf8036e (cosmetic clean-up: default argument, eliminate trailing whitespace, pass pattern as list instead of concatenating). I also added a demonstration to test.sh (separate commit).

I wish github pull requests had a "superseded" feature, but I can't update the PR because it comes from your branch.

[tomasherman]

Cool, thanks