From 45319ccf605ae037ac8bfdcf6591111d9020d08a Mon Sep 17 00:00:00 2001
From: Stefan Majewsky <majewsky@gmx.net>
Date: Sun, 18 Apr 2021 14:33:13 +0200
Subject: [PATCH] shave 4 bytes off the Gloss and LoanwordSource
 representations

This saves 5.29 MiB of payload when compiling with `--features full`.
The total payload in that configuration is now 66.72 MiB
(breaking down into 30.75 MiB raw text plus 35.97 MiB metadata).
---
 build.rs       | 19 +++++++++++--------
 src/lib.rs     |  4 ++--
 src/payload.rs | 20 ++++++++++----------
 3 files changed, 23 insertions(+), 20 deletions(-)

diff --git a/build.rs b/build.rs
index 476d53e..a931c36 100644
--- a/build.rs
+++ b/build.rs
@@ -277,7 +277,7 @@ impl ToPayload for jmdict_traverse::RawSense<'_> {
 
 impl ToPayload for jmdict_traverse::RawLSource<'_> {
     fn size() -> usize {
-        5
+        4
     }
 
     fn encode_one(&self, omni: &mut OmniBuffer, buf: &mut [u32]) {
@@ -287,26 +287,29 @@ impl ToPayload for jmdict_traverse::RawLSource<'_> {
         let r = omni.push_str(self.lang);
         buf[2] = r.start;
         buf[3] = r.end;
-        buf[4] = 0;
+        //`omni.text` is significantly shorter than 2^28 bytes, so we can shove those two booleans
+        //into the highest bits of one of the offset values
         if self.is_partial {
-            buf[4] |= 0x1;
+            buf[0] |= 0x10000000;
         }
         if self.is_wasei {
-            buf[4] |= 0x2;
+            buf[0] |= 0x20000000;
         }
     }
 }
 
 impl ToPayload for jmdict_traverse::RawGloss<'_> {
     fn size() -> usize {
-        3
+        2
     }
 
     fn encode_one(&self, omni: &mut OmniBuffer, buf: &mut [u32]) {
+        //`omni.text` is never larger than 30-40 MiB. That's slightly more than 2^24 bytes, but
+        //comfortably below 2^28 bytes. We can therefore use the upper 4 bits of `buf[0]` and
+        //`buf[1]`, respectively, to encode `self.lang` and `self.g_type`.
         let r = omni.push_str(self.text);
-        buf[0] = r.start;
-        buf[1] = r.end;
-        buf[2] = self.lang.to_u32() | (self.g_type.to_u32() << 16);
+        buf[0] = r.start | (self.lang.to_u32() << 28);
+        buf[1] = r.end | (self.g_type.to_u32() << 28);
     }
 }
 
diff --git a/src/lib.rs b/src/lib.rs
index 6cafe95..3998d51 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -341,9 +341,9 @@ wrap_iterator!(&'static str, 2, Strings);
 wrap_iterator!(PartOfSpeech, 1, PartsOfSpeech);
 wrap_iterator!(SenseTopic, 1, SenseTopics);
 wrap_iterator!(SenseInfo, 1, SenseInfos);
-wrap_iterator!(LoanwordSource, 5, LoanwordSources);
+wrap_iterator!(LoanwordSource, 4, LoanwordSources);
 wrap_iterator!(Dialect, 1, Dialects);
-wrap_iterator!(Gloss, 3, Glosses);
+wrap_iterator!(Gloss, 2, Glosses);
 
 ///An iterator providing fast access to objects in the database. Instances of this iterator
 ///can be copied cheaply.
diff --git a/src/payload.rs b/src/payload.rs
index e8a5d1a..f917df7 100644
--- a/src/payload.rs
+++ b/src/payload.rs
@@ -167,13 +167,13 @@ impl FromPayload<1> for SenseInfo {
     }
 }
 
-impl FromPayload<5> for LoanwordSource {
-    fn get(data: &[u32; 5]) -> Self {
+impl FromPayload<4> for LoanwordSource {
+    fn get(data: &[u32; 4]) -> Self {
         Self {
-            text: get_str(data[0], data[1]),
+            text: get_str(data[0] & 0x0FFFFFFF, data[1]),
             language: get_str(data[2], data[3]),
-            is_partial: (data[4] & 0x1) == 0x1,
-            is_wasei: (data[4] & 0x2) == 0x2,
+            is_partial: (data[0] & 0x10000000) == 0x10000000,
+            is_wasei: (data[0] & 0x20000000) == 0x20000000,
         }
     }
 }
@@ -184,12 +184,12 @@ impl FromPayload<1> for Dialect {
     }
 }
 
-impl FromPayload<3> for Gloss {
-    fn get(data: &[u32; 3]) -> Self {
-        let lang_code = data[2] & 0x0000FFFF;
-        let type_code = (data[2] & 0xFFFF0000) >> 16;
+impl FromPayload<2> for Gloss {
+    fn get(data: &[u32; 2]) -> Self {
+        let lang_code = (data[0] & 0xF0000000) >> 28;
+        let type_code = (data[1] & 0xF0000000) >> 28;
         Gloss {
-            text: get_str(data[0], data[1]),
+            text: get_str(data[0] & 0x0FFFFFFF, data[1] & 0x0FFFFFFF),
             language: jmdict_enums::EnumPayload::from_u32(lang_code),
             gloss_type: jmdict_enums::EnumPayload::from_u32(type_code),
         }