tracking issue for v2 migration testing

[alejandro-vaz]

this is the tracking issue for testing v2 migrations

goal: take the following crates, fork them, and migrate them to v2 to see if everything works fine, including testing and benchmarking them. no need to open a PR for them, just to see if migration would be unfold and if anything would break or become significantly slower

I've chosen crates which seem to be fairly up-to-date with smallvec, it is not a perfectly curated list.

1. gix-object
2. glib
3. cranelift-codegen
4. wasmtime
5. egui
6. spdx
7. spade
8. rpds
9. ort
10. libp2p-dns

the sole objective is go through migration and file issues for any conflicts or problems

[alejandro-vaz]

@fereidani @bolshoytoster I'd love help with this

I understand you may have commitments, right after September started this repository became a desert

it's perfectly fine if neither of you can

[bolshoytoster]

I can help. I'll start with gix-object if nobody else is doing it already.

right after September started this repository became a desert

I think it was mainly because we closed most of the existing issues.

[alejandro-vaz]

yeah I do agree that we have less issues, we peaked repo activity at around ~30 open issues

I'll open more soon, once v2.0.0-beta.1 gets released so we can start merging safely

though what is left is mostly code revision and edge case handling and fixing if something breaks

maybe go a bit more into performance optimization if necessary

[bolshoytoster]

Done gix-object.

The patch

I didn't bother converting the other crates in the repo, there are 12 that use smallvec.

diff --git a/gix-object/Cargo.toml b/gix-object/Cargo.toml
index 0b24abf..8e42448 100644
--- a/gix-object/Cargo.toml
+++ b/gix-object/Cargo.toml
@@ -60,7 +60,7 @@ bstr = { version = "1.12.0", default-features = false, features = [
     "std",
     "unicode",
 ] }
-smallvec = { version = "1.15.1", features = ["write"] }
+smallvec = { version = "2.0.0-alpha.13", features = ["std"] }
 serde = { version = "1.0.114", optional = true, default-features = false, features = [
     "derive",
 ] }
diff --git a/gix-object/src/encode.rs b/gix-object/src/encode.rs
index 842fbe8..c5b8f13 100644
--- a/gix-object/src/encode.rs
+++ b/gix-object/src/encode.rs
@@ -19,7 +19,7 @@ macro_rules! check {
     };
 }
 /// Generates a loose header buffer
-pub fn loose_header(kind: crate::Kind, size: u64) -> smallvec::SmallVec<[u8; 28]> {
+pub fn loose_header(kind: crate::Kind, size: u64) -> smallvec::SmallVec<u8, 28> {
     let mut v = smallvec::SmallVec::new();
     check!(v.write_all(kind.as_bytes()));
     check!(v.write_all(SPACE));
diff --git a/gix-object/src/lib.rs b/gix-object/src/lib.rs
index efd9cff..45b01ca 100644
--- a/gix-object/src/lib.rs
+++ b/gix-object/src/lib.rs
@@ -117,7 +117,7 @@ pub struct CommitRef<'a> {
     #[cfg_attr(feature = "serde", serde(borrow))]
     pub tree: &'a BStr,
     /// HEX hash of each parent commit. Empty for first commit in repository.
-    pub parents: SmallVec<[&'a BStr; 1]>,
+    pub parents: SmallVec<&'a BStr, 1>,
     /// The raw author header value as encountered during parsing.
     ///
     /// Use the [`author()`](CommitRef::author()) method to obtain a parsed version of it.
@@ -152,7 +152,7 @@ pub struct Commit {
     /// The hash of recorded working tree state.
     pub tree: gix_hash::ObjectId,
     /// Hash of each parent commit. Empty for the first commit in repository.
-    pub parents: SmallVec<[gix_hash::ObjectId; 1]>,
+    pub parents: SmallVec<gix_hash::ObjectId, 1>,
     /// Who wrote this commit.
     pub author: gix_actor::Signature,
     /// Who committed this commit.
diff --git a/gix-object/src/traits/mod.rs b/gix-object/src/traits/mod.rs
index 66aa0bc..cb48b8a 100644
--- a/gix-object/src/traits/mod.rs
+++ b/gix-object/src/traits/mod.rs
@@ -64,7 +64,7 @@ pub trait WriteTo {
     fn size(&self) -> u64;
 
     /// Returns a loose object header based on the object's data
-    fn loose_header(&self) -> smallvec::SmallVec<[u8; 28]> {
+    fn loose_header(&self) -> smallvec::SmallVec<u8, 28> {
         crate::encode::loose_header(self.kind(), self.size())
     }
 }

Changes in gix-object's benchmarks:

Benchmark results

     Running benches/decode_objects.rs (gitoxide/target/release/build/gix-object/ffbc5324dd554321/out/decode_objects-ffbc5324dd554321)
Gnuplot not found, using plotters backend
CommitRef(sig)          time:   [736.93 ns 738.43 ns 740.18 ns]
                        change: [−15.252% −14.110% −12.910%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 5 outliers among 100 measurements (5.00%)
  1 (1.00%) high mild
  4 (4.00%) high severe

CommitRefIter(sig)      time:   [1.0033 µs 1.0064 µs 1.0102 µs]
                        change: [+16.016% +17.313% +18.654%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 20 outliers among 100 measurements (20.00%)
  6 (6.00%) low severe
  4 (4.00%) low mild
  6 (6.00%) high mild
  4 (4.00%) high severe

TagRef(sig)             time:   [307.25 ns 308.22 ns 309.49 ns]
                        change: [−6.6257% −5.1096% −3.5831%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 7 outliers among 100 measurements (7.00%)
  4 (4.00%) high mild
  3 (3.00%) high severe

TagRefIter(sig)         time:   [383.39 ns 389.30 ns 396.48 ns]
                        change: [−5.3371% −4.0121% −2.7640%] (p = 0.00 < 0.05)
                        Performance has improved.

TreeRef()               time:   [163.19 ns 163.34 ns 163.49 ns]
                        change: [−7.5290% −5.2322% −3.3217%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  7 (7.00%) low severe
  1 (1.00%) low mild
  2 (2.00%) high severe

TreeRefIter()           time:   [62.189 ns 62.299 ns 62.431 ns]
                        change: [−16.839% −16.552% −16.269%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 9 outliers among 100 measurements (9.00%)
  2 (2.00%) low severe
  2 (2.00%) high mild
  5 (5.00%) high severe

     Running benches/edit_tree.rs (gitoxide/target/release/build/gix-object/76b863b37b9d74a5/out/edit_tree-76b863b37b9d74a5)
Gnuplot not found, using plotters backend
Benchmarking editor/small tree (empty -> full -> empty): Collecting 100 samples in estimated 5.0136 s (1editor/small tree (empty -> full -> empty)
                        time:   [4.5105 µs 4.7260 µs 4.9250 µs]
                        thrpt:  [2.0304 Melem/s 2.1160 Melem/s 2.2171 Melem/s]
                 change:
                        time:   [−13.357% −11.223% −8.9631%] (p = 0.00 < 0.05)
                        thrpt:  [+9.8456% +12.642% +15.416%]
                        Performance has improved.
Found 22 outliers among 100 measurements (22.00%)
  1 (1.00%) low mild
  5 (5.00%) high mild
  16 (16.00%) high severe
Benchmarking editor/deeply nested tree (empty -> full -> empty): Collecting 100 samples in estimated 5.0editor/deeply nested tree (empty -> full -> empty)
                        time:   [11.273 µs 11.289 µs 11.308 µs]
                        thrpt:  [4.0680 Melem/s 4.0746 Melem/s 4.0806 Melem/s]
                 change:
                        time:   [−7.6592% −7.3828% −7.1163%] (p = 0.00 < 0.05)
                        thrpt:  [+7.6616% +7.9713% +8.2945%]
                        Performance has improved.
Found 4 outliers among 100 measurements (4.00%)
  2 (2.00%) high mild
  2 (2.00%) high severe

Benchmarking cursor/small tree (empty -> full -> empty): Collecting 100 samples in estimated 5.0051 s (8cursor/small tree (empty -> full -> empty)
                        time:   [5.8261 µs 5.8341 µs 5.8438 µs]
                        thrpt:  [1.7112 Melem/s 1.7141 Melem/s 1.7164 Melem/s]
                 change:
                        time:   [+28.598% +28.951% +29.299%] (p = 0.00 < 0.05)
                        thrpt:  [−22.660% −22.451% −22.238%]
                        Performance has regressed.
Found 7 outliers among 100 measurements (7.00%)
  4 (4.00%) high mild
  3 (3.00%) high severe
Benchmarking cursor/deeply nested tree (empty -> full -> empty): Collecting 100 samples in estimated 5.0cursor/deeply nested tree (empty -> full -> empty)
                        time:   [5.2044 µs 5.4138 µs 5.6520 µs]
                        thrpt:  [8.1387 Melem/s 8.4969 Melem/s 8.8387 Melem/s]
                 change:
                        time:   [−2.1870% +1.3112% +5.0210%] (p = 0.48 > 0.05)
                        thrpt:  [−4.7809% −1.2943% +2.2359%]
                        No change in performance detected.
Found 32 outliers among 100 measurements (32.00%)
  19 (19.00%) low severe
  8 (8.00%) high mild
  5 (5.00%) high severe

Might be worth looking into the 28% regression in cursor/small tree (empty -> full -> empty), to make sure that's not just random variation.

I've also measured that the gix binary is 122,848 bytes (0.032%) larger on debug, and 59,496 bytes (0.16%) larger on release.
The log example is 788,576 bytes (0.486%) larger on debug, and 2,944 bytes (0.046%) smaller on release.

[alejandro-vaz]

that looks perfectly fine to me. thanks for digging into it

smallvec being 12x downstream is not nice but it'd be a waste of time to try touch them

the 28% percent regression might be something as you noted though, any idea what paths it touches??