Separate buffer handling for info and format access. (#281)

* use info-local buffer

* add values to constructor

* impl drop for Info

* lifetimes

* record reference does not need to be mut

* api tuning

* Move buffer along with values so that slices do not outlive the buffer.

* fmt

* also use separate buffers for format data.

* go back to previously pinned htslib commit

Cargo.toml

@@ -28,6 +28,8 @@ serde_bytes = { version = "0.11", optional = true }
 bio-types = ">=0.6"
 thiserror = "1"
 hts-sys = { version = "1.11.0", path = "hts-sys", default-features = false }
+derefable = "0.1"
+derive-new = "0.5"
 
 [features]
 default = ["bzip2", "lzma", "curl"]

src/bcf/mod.rs

@@ -781,6 +781,7 @@ fn bcf_open(target: &[u8], mode: &[u8]) -> Result<*mut htslib::htsFile> {
 
 #[cfg(test)]
 mod tests {
+    use super::record::Buffer;
     use super::*;
     use crate::bcf::header::Id;
     use crate::bcf::record::GenotypeAllele;
@@ -797,15 +798,16 @@ mod tests {
         assert_eq!(bcf.header.samples(), [b"NA12878.subsample-0.25-0"]);
 
         for (i, rec) in bcf.records().enumerate() {
-            let mut record = rec.expect("Error reading record.");
+            let record = rec.expect("Error reading record.");
             assert_eq!(record.sample_count(), 1);
 
             assert_eq!(record.rid().expect("Error reading rid."), 0);
             assert_eq!(record.pos(), 10021 + i as i64);
             assert!((record.qual() - 0f32).abs() < std::f32::EPSILON);
+            let mut buffer = Buffer::new();
             assert!(
                 (record
-                    .info(b"MQ0F")
+                    .info_shared_buffer(b"MQ0F", &mut buffer)
                     .float()
                     .expect("Error reading info.")
                     .expect("Missing tag")[0]
@@ -816,7 +818,7 @@ mod tests {
             if i == 59 {
                 assert!(
                     (record
-                        .info(b"SGB")
+                        .info_shared_buffer(b"SGB", &mut buffer)
                         .float()
                         .expect("Error reading info.")
                         .expect("Missing tag")[0]
@@ -918,12 +920,13 @@ mod tests {
             &b"evenlength"[..],
             &b"ss6"[..],
         ];
+        let mut buffer = Buffer::new();
         for (i, rec) in vcf.records().enumerate() {
             println!("record {}", i);
             let mut record = rec.expect("Error reading record.");
             assert_eq!(
                 record
-                    .info(b"S1")
+                    .info_shared_buffer(b"S1", &mut buffer)
                     .string()
                     .expect("Error reading string.")
                     .expect("Missing tag")[0],
@@ -961,11 +964,12 @@ mod tests {
             &[i32::missing()][..],
         ];
         let f1 = [false, true];
+        let mut buffer = Buffer::new();
         for (i, rec) in vcf.records().enumerate() {
             let mut record = rec.expect("Error reading record.");
             assert_eq!(
                 record
-                    .info(b"F1")
+                    .info_shared_buffer(b"F1", &mut buffer)
                     .float()
                     .expect("Error reading float.")
                     .expect("Missing tag")[0]
@@ -1339,7 +1343,10 @@ mod tests {
         let mut record = reader.empty_record();
         reader.read(&mut record).unwrap().unwrap();
 
-        assert_eq!(record.info(b"SVLEN").integer().unwrap(), Some(&[-127][..]));
+        assert_eq!(
+            *record.info(b"SVLEN").integer().unwrap().unwrap(),
+            &[-127][..]
+        );
     }
 
     #[test]
@@ -1360,7 +1367,14 @@ mod tests {
         let mut rec = reader.empty_record();
         let _ = reader.read(&mut rec);
 
-        assert_eq!(rec.info(b"ANN").string().unwrap().unwrap().len(), 14);
+        assert_eq!(
+            rec.info_shared_buffer(b"ANN", Buffer::new())
+                .string()
+                .unwrap()
+                .unwrap()
+                .len(),
+            14
+        );
     }
 
     #[test]
@@ -1369,7 +1383,14 @@ mod tests {
         let mut rec = reader.empty_record();
         let _ = reader.read(&mut rec);
 
-        assert_eq!(rec.info(b"X").string().unwrap().unwrap().len(), 2);
+        assert_eq!(
+            rec.info_shared_buffer(b"X", Buffer::new())
+                .string()
+                .unwrap()
+                .unwrap()
+                .len(),
+            2
+        );
     }
 
     #[test]
@@ -1417,4 +1438,43 @@ mod tests {
         assert_eq!(n_alt, [1, 2]);
         assert_eq!(n_missing, [1, 0]);
     }
+
+    #[test]
+    fn test_obs_cornercase() {
+        let mut reader = Reader::from_path("test/obs-cornercase.vcf").unwrap();
+        let first_record = reader
+            .records()
+            .next()
+            .unwrap()
+            .expect("Fail to read record");
+
+        assert_eq!(
+            *first_record.info(b"EVENT").string().unwrap().unwrap(),
+            [b"gridss33fb_1085"]
+        );
+        assert_eq!(
+            *first_record.info(b"MATEID").string().unwrap().unwrap(),
+            [b"gridss33fb_1085h"]
+        );
+    }
+
+    // #[test]
+    // fn test_buffer_lifetime() {
+    //     let mut reader = Reader::from_path("test/obs-cornercase.vcf").unwrap();
+    //     let first_record = reader
+    //         .records()
+    //         .next()
+    //         .unwrap()
+    //         .expect("Fail to read record");
+
+    //     fn get_value<'a, 'b>(record: &'a Record) -> &'b [u8] {
+    //         // FIXME: this should not be possible, because the slice outlives the buffer.
+    //         let buffer: BufferBacked<'b, _, _> = record.info(b"EVENT").string().unwrap().unwrap();
+    //         let value: &'b [u8] = buffer[0];
+    //         value
+    //     }
+
+    //     let buffered = first_record.info(b"EVENT").string().unwrap().unwrap();
+    //     assert_eq!(get_value(&first_record), buffered[0]);
+    // }
 }