improve: rework VoiceModel

Cargo.toml

@@ -8,6 +8,7 @@ anstream = { version = "0.5.0", default-features = false }
 anstyle-query = "1.0.0"
 anyhow = "1.0.65"
 assert_cmd = "2.0.8"
+async-fs = "2.1.2"
 async_zip = "=0.0.16"
 bindgen = "0.69.4"
 binstall-tar = "0.4.39"
@@ -33,10 +34,10 @@ enum-map = "3.0.0-beta.1"
 eyre = "0.6.8"
 flate2 = "1.0.25"
 fs-err = "2.11.0"
-futures = "0.3.26"
 futures-core = "0.3.25"
 futures-util = "0.3.25"
 futures-lite = "2.2.0"
+futures-io = "0.3.28"
 heck = "0.4.1"
 humansize = "2.1.2"
 indexmap = "2.0.0"

crates/voicevox_core/Cargo.toml

@@ -16,6 +16,7 @@ link-onnxruntime = []
 
 [dependencies]
 anyhow.workspace = true
+async-fs.workspace = true
 async_zip = { workspace = true, features = ["deflate"] }
 camino.workspace = true
 const_format.workspace = true
@@ -27,14 +28,15 @@ easy-ext.workspace = true
 educe.workspace = true
 enum-map.workspace = true
 fs-err = { workspace = true, features = ["tokio"] }
-futures.workspace = true
+futures-io.workspace = true
+futures-lite.workspace = true
+futures-util = { workspace = true, features = ["io"] }
 indexmap = { workspace = true, features = ["serde"] }
 itertools.workspace = true
 jlabel.workspace = true
 ndarray.workspace = true
 open_jtalk.workspace = true
 ouroboros.workspace = true
-rayon.workspace = true
 ref-cast.workspace = true
 regex.workspace = true
 serde = { workspace = true, features = ["derive", "rc"] }
@@ -49,7 +51,6 @@ tracing.workspace = true
 uuid = { workspace = true, features = ["v4", "serde"] }
 voicevox-ort = { workspace = true, features = ["download-binaries", "__init-for-voicevox"] }
 voicevox_core_macros = { path = "../voicevox_core_macros" }
-zip.workspace = true
 
 [dev-dependencies]
 heck.workspace = true

crates/voicevox_core/src/infer/domains.rs

@@ -1,14 +1,70 @@
 mod talk;
 
+use std::future::Future;
+
+use educe::Educe;
+use serde::{Deserialize, Deserializer};
+
 pub(crate) use self::talk::{
     DecodeInput, DecodeOutput, PredictDurationInput, PredictDurationOutput, PredictIntonationInput,
     PredictIntonationOutput, TalkDomain, TalkOperation,
 };
 
+#[derive(Educe)]
+// TODO: `bounds`に`V: ?Sized`も入れようとすると、よくわからない理由で弾かれる。最新版のeduce
+// でもそうなのか？また最新版でも駄目だとしたら、弾いている理由は何なのか？
+#[educe(Clone(bound = "V: InferenceDomainMapValues, V::Talk: Clone"))]
 pub(crate) struct InferenceDomainMap<V: InferenceDomainMapValues + ?Sized> {
     pub(crate) talk: V::Talk,
 }
 
+impl<T> InferenceDomainMap<(T,)> {
+    pub(crate) fn each_ref(&self) -> InferenceDomainMap<(&T,)> {
+        InferenceDomainMap { talk: &self.talk }
+    }
+
+    pub(crate) fn map<T2, Ft: FnOnce(T) -> T2>(
+        self,
+        fs: InferenceDomainMap<(Ft,)>,
+    ) -> InferenceDomainMap<(T2,)> {
+        InferenceDomainMap {
+            talk: (fs.talk)(self.talk),
+        }
+    }
+}
+
+impl<T, E> InferenceDomainMap<(Result<T, E>,)> {
+    pub(crate) fn collect(self) -> Result<InferenceDomainMap<(T,)>, E> {
+        let talk = self.talk?;
+        Ok(InferenceDomainMap { talk })
+    }
+}
+
+impl<T: Future> InferenceDomainMap<(T,)> {
+    pub(crate) async fn join(self) -> InferenceDomainMap<(T::Output,)> {
+        let talk = self.talk.await;
+        InferenceDomainMap { talk }
+    }
+}
+
+impl<'de, V: InferenceDomainMapValues + ?Sized> Deserialize<'de> for InferenceDomainMap<V>
+where
+    V::Talk: Deserialize<'de>,
+{
+    fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+    where
+        D: Deserializer<'de>,
+    {
+        let Repr { talk } = Repr::deserialize(deserializer)?;
+        return Ok(Self { talk });
+
+        #[derive(Deserialize)]
+        struct Repr<T> {
+            talk: T,
+        }
+    }
+}
+
 pub(crate) trait InferenceDomainMapValues {
     type Talk;
 }

crates/voicevox_core/src/manifest.rs

@@ -7,10 +7,14 @@ use std::{
 use derive_getters::Getters;
 use derive_more::Deref;
 use derive_new::new;
+use macros::IndexForFields;
 use serde::{de, Deserialize, Deserializer, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
 
-use crate::{StyleId, VoiceModelId};
+use crate::{
+    infer::domains::{InferenceDomainMap, TalkOperation},
+    StyleId, VoiceModelId,
+};
 
 #[derive(Clone)]
 struct FormatVersionV1;
@@ -65,26 +69,31 @@ impl Display for InnerVoiceId {
     }
 }
 
-#[derive(Deserialize, Getters, Clone)]
+#[derive(Deserialize, Getters)]
 pub struct Manifest {
     #[expect(dead_code, reason = "現状はバリデーションのためだけに存在")]
     vvm_format_version: FormatVersionV1,
     pub(crate) id: VoiceModelId,
     metas_filename: String,
     #[serde(flatten)]
-    domains: ManifestDomains,
+    domains: InferenceDomainMap<ManifestDomains>,
 }
 
-#[derive(Deserialize, Clone)]
-pub(crate) struct ManifestDomains {
-    pub(crate) talk: Option<TalkManifest>,
-}
+pub(crate) type ManifestDomains = (Option<TalkManifest>,);
 
-#[derive(Deserialize, Clone)]
+#[derive(Deserialize, IndexForFields)]
+#[cfg_attr(test, derive(Default))]
+#[index_for_fields(TalkOperation)]
 pub(crate) struct TalkManifest {
-    pub(crate) predict_duration_filename: String,
-    pub(crate) predict_intonation_filename: String,
-    pub(crate) decode_filename: String,
+    #[index_for_fields(TalkOperation::PredictDuration)]
+    pub(crate) predict_duration_filename: Arc<str>,
+
+    #[index_for_fields(TalkOperation::PredictIntonation)]
+    pub(crate) predict_intonation_filename: Arc<str>,
+
+    #[index_for_fields(TalkOperation::Decode)]
+    pub(crate) decode_filename: Arc<str>,
+
     #[serde(default)]
     pub(crate) style_id_to_inner_voice_id: StyleIdToInnerVoiceId,
 }