refactor: VVMマニフェストで#[serde(flatten)]を活用

crates/voicevox_core/Cargo.toml

@@ -29,7 +29,7 @@ derive_more.workspace = true
 duplicate.workspace = true
 easy-ext.workspace = true
 educe.workspace = true
-enum-map.workspace = true
+enum-map = { workspace = true, features = ["serde"] }
 fs-err.workspace = true
 futures-io.workspace = true
 futures-lite.workspace = true

crates/voicevox_core/src/manifest.rs

@@ -1,13 +1,14 @@
 use std::{
     collections::BTreeMap,
     fmt::{self, Display},
+    ops::Index,
     sync::Arc,
 };
 
 use derive_getters::Getters;
 use derive_more::Deref;
 use derive_new::new;
-use macros::IndexForFields;
+use enum_map::{Enum, EnumMap};
 use serde::{de, Deserialize, Deserializer, Serialize};
 use serde_with::{serde_as, DisplayFromStr};
 
@@ -81,24 +82,43 @@ pub struct Manifest {
 
 pub(crate) type ManifestDomains = inference_domain_map_values!(for<D> Option<D::Manifest>);
 
-#[derive(Deserialize, IndexForFields)]
+#[derive(Deserialize)]
 #[cfg_attr(test, derive(Default))]
-#[index_for_fields(TalkOperation)]
 pub(crate) struct TalkManifest {
-    #[index_for_fields(TalkOperation::PredictDuration)]
-    pub(crate) predict_duration_filename: Arc<str>,
+    #[serde(flatten)]
+    filenames: EnumMap<TalkOperationFilenameKey, Arc<str>>,
 
-    #[index_for_fields(TalkOperation::PredictIntonation)]
-    pub(crate) predict_intonation_filename: Arc<str>,
+    #[serde(default)]
+    pub(crate) style_id_to_inner_voice_id: StyleIdToInnerVoiceId,
+}
 
-    #[index_for_fields(TalkOperation::GenerateFullIntermediate)]
-    pub(crate) generate_full_intermediate_filename: Arc<str>,
+// TODO: #825 では`TalkOperation`と統合する。`Index`の実装もderive_moreで委譲する
+#[derive(Enum, Deserialize)]
+pub(crate) enum TalkOperationFilenameKey {
+    #[serde(rename = "predict_duration_filename")]
+    PredictDuration,
+    #[serde(rename = "predict_intonation_filename")]
+    PredictIntonation,
+    #[serde(rename = "generate_full_intermediate_filename")]
+    GenerateFullIntermediate,
+    #[serde(rename = "render_audio_segment_filename")]
+    RenderAudioSegment,
+}
 
-    #[index_for_fields(TalkOperation::RenderAudioSegment)]
-    pub(crate) render_audio_segment_filename: Arc<str>,
+impl Index<TalkOperation> for TalkManifest {
+    type Output = Arc<str>;
 
-    #[serde(default)]
-    pub(crate) style_id_to_inner_voice_id: StyleIdToInnerVoiceId,
+    fn index(&self, index: TalkOperation) -> &Self::Output {
+        let key = match index {
+            TalkOperation::PredictDuration => TalkOperationFilenameKey::PredictDuration,
+            TalkOperation::PredictIntonation => TalkOperationFilenameKey::PredictIntonation,
+            TalkOperation::GenerateFullIntermediate => {
+                TalkOperationFilenameKey::GenerateFullIntermediate
+            }
+            TalkOperation::RenderAudioSegment => TalkOperationFilenameKey::RenderAudioSegment,
+        };
+        &self.filenames[key]
+    }
 }
 
 #[serde_as]

crates/voicevox_core/src/voice_model.rs

@@ -11,7 +11,7 @@ use std::{
 use anyhow::{anyhow, Context as _};
 use derive_more::From;
 use easy_ext::ext;
-use enum_map::{enum_map, EnumMap};
+use enum_map::EnumMap;
 use futures_io::{AsyncBufRead, AsyncRead, AsyncSeek};
 use futures_util::future::{OptionFuture, TryFutureExt as _};
 use itertools::Itertools as _;
@@ -23,7 +23,7 @@ use crate::{
     asyncs::{Async, Mutex as _},
     error::{LoadModelError, LoadModelErrorKind, LoadModelResult},
     infer::{
-        domains::{inference_domain_map_values, InferenceDomainMap, TalkDomain, TalkOperation},
+        domains::{inference_domain_map_values, InferenceDomainMap, TalkDomain},
         InferenceDomain,
     },
     manifest::{Manifest, ManifestDomains, StyleIdToInnerVoiceId},
@@ -128,7 +128,7 @@ impl<A: Async> Inner<A> {
 
         let header = VoiceModelHeader::new(manifest, metas, path)?.into();
 
-        InnerTryBuilder {
+        return InnerTryBuilder {
             header,
             inference_model_entries_builder: |header| {
                 let VoiceModelHeader { manifest, .. } = &**header;
@@ -139,21 +139,10 @@ impl<A: Async> Inner<A> {
                         talk: |talk| {
                             talk.as_ref()
                                 .map(|manifest| {
-                                    let indices = enum_map! {
-                                        TalkOperation::PredictDuration => {
-                                            find_entry_index(&manifest.predict_duration_filename)?
-                                        }
-                                        TalkOperation::PredictIntonation => {
-                                            find_entry_index(&manifest.predict_intonation_filename)?
-                                        }
-                                        TalkOperation::GenerateFullIntermediate => {
-                                            find_entry_index(&manifest.generate_full_intermediate_filename)?
-                                        }
-                                        TalkOperation::RenderAudioSegment => {
-                                            find_entry_index(&manifest.render_audio_segment_filename)?
-                                        }
-                                    };
-
+                                    let indices = EnumMap::from_fn(|k| &manifest[k])
+                                        .into_array()
+                                        .try_map_(|s| find_entry_index(s))?;
+                                    let indices = EnumMap::from_array(indices);
                                     Ok(InferenceModelEntry { indices, manifest })
                                 })
                                 .transpose()
@@ -172,7 +161,20 @@ impl<A: Async> Inner<A> {
             },
             zip: zip.into_inner().into_inner().into(),
         }
-        .try_build()
+        .try_build();
+
+        #[ext]
+        impl<T, const N: usize> [T; N] {
+            fn try_map_<O, E>(self, f: impl FnMut(T) -> Result<O, E>) -> Result<[O; N], E> {
+                self.into_iter()
+                    .map(f)
+                    .collect::<Result<Vec<_>, _>>()
+                    .map(|vec| {
+                        vec.try_into()
+                            .unwrap_or_else(|_| unreachable!("should be same length"))
+                    })
+            }
+        }
     }
 
     fn id(&self) -> VoiceModelId {

crates/voicevox_core_macros/src/lib.rs

@@ -3,7 +3,6 @@
 mod extract;
 mod inference_domain;
 mod inference_domains;
-mod manifest;
 
 use syn::parse_macro_input;
 
@@ -103,38 +102,6 @@ pub fn derive_inference_output_signature(
     from_syn(inference_domain::derive_inference_output_signature(input))
 }
 
-/// 構造体のフィールドを取得できる`std::ops::Index`の実装を導出する。
-///
-/// # Example
-///
-/// ```
-/// use macros::IndexForFields;
-///
-/// #[derive(IndexForFields)]
-/// #[index_for_fields(TalkOperation)]
-/// pub(crate) struct TalkManifest {
-///     #[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::GenerateFullIntermediate)]
-///     pub(crate) generate_full_intermediate_filename: Arc<str>,
-///
-///     #[index_for_fields(TalkOperation::RenderAudioSegment)]
-///     pub(crate) render_audio_segment_filename: Arc<str>,
-///
-///     // …
-/// }
-/// ```
-#[cfg(not(doctest))]
-#[proc_macro_derive(IndexForFields, attributes(index_for_fields))]
-pub fn derive_index_for_fields(input: proc_macro::TokenStream) -> proc_macro::TokenStream {
-    let input = &parse_macro_input!(input);
-    from_syn(manifest::derive_index_for_fields(input))
-}
-
 /// # Example
 ///
 /// ```