use a boxed trait for the Repository

[danieleades]

depends on #50

this is what it would look like if the index enum was eliminated all-together (i've replaced it with a boxed trait).

The advantage of this is you can use any object capable of managing a git repo, not just the implementations defined in this crate.

i'd be interested to get your thoughts on this @Hirevo?

there's another alternative, which is to make the Index generic over the repository type, rather than boxing the trait. The calling code would then probably need to box the Index to avoid specifying a concrete type.

[danieleades]

this should be rebased once(/if) #50 is merged

[danieleades]

under this scheme, both the CLI and GIT2 strategies should be feature-gated (though can be included by default)

[codecov]

Codecov Report

Merging #52 into master will increase coverage by 0.25%.
The diff coverage is 7.04%.

@@            Coverage Diff            @@
##           master     #52      +/-   ##
=========================================
+ Coverage    0.00%   0.25%   +0.25%     
=========================================
  Files          50      47       -3     
  Lines        1988    1963      -25     
=========================================
+ Hits            0       5       +5     
+ Misses       1988    1958      -30     

Impacted Files	Coverage Δ
src/config/mod.rs	0.00% <0.00%> (ø)
src/index/repository/cli.rs	0.00% <0.00%> (ø)
src/index/repository/git2.rs	0.00% <0.00%> (ø)
src/index/mod.rs	13.51% <19.23%> (+13.51%)	⬆️
src/db/mod.rs	0.00% <0.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ff77d0c...740df21. Read the comment docs.

[Hirevo]

Initially, I would have said that there isn't much to be gained from using boxed dynamic dispatch vs an enum, but with the upcoming library-split of Alexandrie, I think dynamic dispatch makes more sense.
People could then implement their own index management crate and easily get Alexandrie to use it, if they want to.
Also, I guess that with what you pointed out in #51 (about mutexing the entire State), the type would be reduced to just Box<dyn Repository + 'static> (we would no longer require Send + Sync and allow for more diverse impls of Repository to be used in Index).
With all this, I would say that I am on board about doing this.

One thing, though, is that we may have the same kind of naming problem with Repo / Repository than with Index / Indexer.
If I remember correctly, an idea we had was to potentially rename Index / Indexer to LocalIndex / Index, so maybe we can do the same with LocalRepository / Repository ?

[danieleades]

i'll take another look at the naming conventions. I agree with you there's some work to do there.

I cannot for the life of me remember in which order these pull requests should be merged (if they're merged)