refactor: Replace bincode with postcard

[Muscraft]

I saw #98 and figured I would take a crack at moving away from bincode. I went with postcard as the replacement, as it is essentially a drop-in replacement for bincode given how phonenumber used it. postcard also uses varint encoding by default, which is a nice plus.

The only major difference (from what I can tell) is that the serialized format is different from bincode, but that shouldn't matter.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 67.25%. Comparing base (09f8974) to head (593759f).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #99      +/-   ##
==========================================
- Coverage   67.41%   67.25%   -0.16%     
==========================================
  Files          18       18              
  Lines        2145     2138       -7     
==========================================
- Hits         1446     1438       -8     
- Misses        699      700       +1     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[rubdos]

Hi! Thanks for filing your merge request. I appreciate you taking a crack at removing the advisory warning!

As you might have seen from my comment on the issue, I am more after a patch like this, without any runtime deserialization at all:

diff --git a/build.rs b/build.rs
index 44840cd..0512c95 100644
--- a/build.rs
+++ b/build.rs
@@ -1,5 +1,6 @@
 use std::env;
 use std::fs::File;
+use std::io::Write;
 use std::io::{BufReader, BufWriter};
 use std::path::Path;
 
@@ -17,15 +18,17 @@ fn main() {
         File::open(pnm_path).expect("could not open metadata file"),
     ))
     .expect("failed to load metadata");
+    let metadata = metadata.as_slice();
+
+    // TODO: remove
+    let metadata = &metadata[..4];
+
     println!("cargo:rerun-if-changed={pnm_path}");
 
     let mut out = BufWriter::new(
-        File::create(Path::new(&env::var("OUT_DIR").unwrap()).join("database.bin"))
+        File::create(Path::new(&env::var("OUT_DIR").unwrap()).join("database.rs"))
             .expect("could not create database file"),
     );
 
-    bincode::options()
-        .with_varint_encoding()
-        .serialize_into(&mut out, &metadata)
-        .expect("failed to serialize database");
+    write!(out, "pub const DATABASE: &[Metadata] = &{:?};", metadata).expect("write database");
 }
diff --git a/src/metadata/database.rs b/src/metadata/database.rs
index 46ae55e..24dca0c 100644
--- a/src/metadata/database.rs
+++ b/src/metadata/database.rs
@@ -15,8 +15,6 @@
 use crate::error;
 use crate::metadata::loader;
 use crate::Metadata;
-use bincode;
-use bincode::Options;
 use fnv::FnvHashMap;
 use once_cell::sync::Lazy;
 use regex_cache::{CachedRegex, CachedRegexBuilder, RegexCache};
@@ -27,18 +25,17 @@ use std::io::{BufReader, Cursor};
 use std::path::Path;
 use std::sync::{Arc, Mutex};
 
-const DATABASE: &[u8] = include_bytes!(concat!(env!("OUT_DIR"), "/database.bin"));
+mod database {
+    use crate::metadata::loader::*;
+    // `database.rs` contains
+    // pub const DATABASE: &[Metadata] = &[ .. ];
+    include!(concat!(env!("OUT_DIR"), "/database.rs"));
+}
+
+use database::DATABASE;
 
 /// The Google provided metadata database, used as default.
-pub static DEFAULT: Lazy<Database> = Lazy::new(|| {
-    Database::from(
-        bincode::options()
-            .with_varint_encoding()
-            .deserialize(DATABASE)
-            .unwrap(),
-    )
-    .unwrap()
-});
+pub static DEFAULT: Lazy<Database> = Lazy::new(|| Database::from(DATABASE.to_vec()).unwrap());
 
 /// Representation of a database of metadata for phone number.
 #[derive(Clone, Debug)]

This has quite some implications regarding borrowing (preferably we'd be on &[Metadata] and borrow everything all the way), and the resulting database.rs file is quite large (1.2MB). So this approach currently generates an awful lot of compile errors, but I think it could still work.

[rubdos]

My suggestion requires quite some refactoring though: the loader::Metadata is currently a heap-allocated-only structure, which is reused between the build.rs and the actual runtime. It'd need to be refactored to be either reusable through &'static references and heap allocated structures through some generics (T: AsRef, I suppose), or split out by some other manner.

[rubdos]

I spent a bit of time seeing how far it trickles down... Probably this really warrants a complete rewrite of the database loading system as a whole, splitting up the loader into a separate crate for use in the main crate's build.rs, and directly yielding a Database instead of the intermediary [Metadata]. That would also get rid of the "hacky" include! calls in build.rs. That's clearly out of scope for this PR!

Could you have a look at why the rustfmt fails? Then we can pull this in as is :-)

[Muscraft]

Would you like me to open a PR to update the version so this can be released?

[rubdos]

Nah, is fine, I'll do it in a few minutes :-)