ngoguey

Collection of IO and GC improvements

1

@metanivek and I brainstormed areas of improvement for GC and the new IO --- ### Benchmark impact of GC on main process performances > `benchmark` > status: ongoing, @Ngoguey42 Bench...

`Tree.mem` should use the backend's `mem` function

2

The mem functions are implemented this way: ```ml let mem t k = find t k >|= function None -> false | _ -> true let mem_tree t k =...

See #1524 We would need to re-implement it and benchmark this.

It is currently duplicated. See #1348

Users expect re-commiting to be a no-op on the disk

5

With irmin-pack under irmin 2.x, commiting a commit that was already contained in the pack file was a no-op on the disk (thanks to the `always` indexing strategy). With irmin-pack...

The `pack_store` typing disallows using `unsafe_add` (and such) if the store is not in `read_write` mode. This rule is not enforced (snapshot import exploits this), it should be fixed.

irmin-pack: Activity log

2

In order to simplify debugging we could add an append only ascii file that lists the timestamps of the creation, openings, gc starts and gc ends.

If the tree passed to `Commit.v` has a cached hash, it is cheap to compute the commit hash (using a portable commit). That commit hash could then be used to...

Mp example

4

Click here for an interpreted version of the example: https://github.com/Ngoguey42/irmin/blob/mp-example/examples/merkle_proofs.md Thanks to @art-w who helped me with mdx and reviewed the example

Double close isn't supported everywhere

5

irmin-pack mem's failure: ``` [failure] Pool.drop: double free Raised at Stdlib.failwith in file "stdlib.ml", line 29, characters 17-33 Called from Irmin_pack_mem__Content_addressable.Maker.Make.close in file "src/irmin-pack/mem/content_addressable.ml", line 95, characters 6-32 Called from...