aquatic http protocol: request parsing: remove hashmap for performance

request-from-bytes:

time:   [720.22 ns 723.34 ns 726.73 ns]
change: [-53.265% -52.884% -52.477%] (p = 0.00 < 0.01)

Performance has improved.

aquatic_http_protocol/src/request.rs

@@ -1,5 +1,4 @@
 use anyhow::Context;
-use hashbrown::HashMap;
 use smartstring::{SmartString, LazyCompact};
 
 use super::common::*;
@@ -169,94 +168,16 @@ impl Request {
         let query_string = split_parts.next()
             .with_context(|| "no query string")?;
 
+        // -- Parse key-value pairs
+
         let mut info_hashes = Vec::new();
         let mut opt_peer_id = None;
-        let mut data = HashMap::new();
+        let mut opt_port = None;
+        let mut opt_bytes_left = None;
+        let mut event = AnnounceEvent::default();
+        let mut opt_numwant = None;
+        let mut opt_key = None;
  
-        Self::parse_key_value_pairs_memchr(
-            &mut info_hashes,
-            &mut opt_peer_id,
-            &mut data,
-            query_string
-        )?;
-
-        if location == "/announce" {
-            let numwant = if let Some(s) = data.remove("numwant"){
-                let numwant = s.parse::<usize>()
-                    .map_err(|err|
-                        anyhow::anyhow!("parse 'numwant': {}", err)
-                    )?;
-                
-                Some(numwant)
-            } else {
-                None
-            };
-            let key = if let Some(s) = data.remove("key"){
-                if s.len() > 100 {
-                    return Err(anyhow::anyhow!("'key' is too long"))
-                }
-
-                Some(s)
-            } else {
-                None
-            };
-            let port = if let Some(port) = data.remove("port"){
-                port.parse().with_context(|| "parse port")?
-            } else {
-                return Err(anyhow::anyhow!("no port"));
-            };
-            let bytes_left = if let Some(left) = data.remove("left"){
-                left.parse().with_context(|| "parse bytes left")?
-            } else {
-                return Err(anyhow::anyhow!("no left"));
-            };
-            let event = if let Some(event) = data.remove("event"){
-                if let Ok(event) = event.parse(){
-                    event
-                } else {
-                    return Err(anyhow::anyhow!("invalid event: {}", event));
-                }
-            } else {
-                AnnounceEvent::default()
-            };
-            let compact = if let Some(compact) = data.remove("compact"){
-                if compact.as_str() == "1" {
-                    true
-                } else {
-                    return Err(anyhow::anyhow!("compact set, but not to 1"));
-                }
-            } else {
-                true
-            };
-
-            let request = AnnounceRequest {
-                info_hash: info_hashes.pop().with_context(|| "no info_hash")?,
-                peer_id: opt_peer_id.with_context(|| "no peer_id")?,
-                port,
-                bytes_left,
-                event,
-                compact,
-                numwant,
-                key,
-            };
-
-            Ok(Request::Announce(request))
-        } else {
-            let request = ScrapeRequest {
-                info_hashes,
-            };
-
-            Ok(Request::Scrape(request))
-        }
-    }
-
-    /// Seems to be somewhat faster than non-memchr version
-    fn parse_key_value_pairs_memchr<'a>(
-        info_hashes: &mut Vec<InfoHash>,
-        opt_peer_id: &mut Option<PeerId>,
-        data: &mut HashMap<&'a str, SmartString<LazyCompact>>,
-        query_string: &'a str,
-    ) -> anyhow::Result<()> {
         let query_string_bytes = query_string.as_bytes();
 
         let mut ampersand_iter = ::memchr::memchr_iter(b'&', query_string_bytes);
@@ -271,7 +192,6 @@ impl Request {
             let value = query_string.get(equal_sign_index + 1..segment_end)
                 .with_context(|| format!("no value at {}..{}", equal_sign_index + 1, segment_end))?;
             
-            // whitelist keys to avoid having to use ddos-resistant hashmap
             match key {
                 "info_hash" => {
                     let value = Self::urldecode_20_bytes(value)?;
@@ -281,10 +201,32 @@ impl Request {
                 "peer_id" => {
                     let value = Self::urldecode_20_bytes(value)?;
 
-                    *opt_peer_id = Some(PeerId(value));
+                    opt_peer_id = Some(PeerId(value));
                 },
-                "port" | "left" | "event" | "compact" | "numwant" | "key" => {
-                    data.insert(key, value.into());
+                "port" => {
+                    opt_port = Some(value.parse::<u16>().with_context(|| "parse port")?);
+                },
+                "left"  => {
+                    opt_bytes_left = Some(value.parse::<usize>().with_context(|| "parse left")?);
+                },
+                "event"  => {
+                    event = value.parse::<AnnounceEvent>().map_err(|err|
+                        anyhow::anyhow!("invalid event: {}", err)
+                    )?;
+                },
+                "compact"  => {
+                    if value != "1" {
+                        return Err(anyhow::anyhow!("compact set, but not to 1"));
+                    }
+                },
+                "numwant"  => {
+                    opt_numwant = Some(value.parse::<usize>().with_context(|| "parse numwant")?);
+                },
+                "key" => {
+                    if value.len() > 100 {
+                        return Err(anyhow::anyhow!("'key' is too long"))
+                    }
+                    opt_key = Some(value.into());
                 },
                 k => {
                     ::log::info!("ignored unrecognized key: {}", k)
@@ -298,7 +240,28 @@ impl Request {
             }
         }
 
-        Ok(())
+        // -- Put together request
+
+        if location == "/announce" {
+            let request = AnnounceRequest {
+                info_hash: info_hashes.pop().with_context(|| "no info_hash")?,
+                peer_id: opt_peer_id.with_context(|| "no peer_id")?,
+                port: opt_port.with_context(|| "no port")?,
+                bytes_left: opt_bytes_left.with_context(|| "no left")?,
+                event,
+                compact: true,
+                numwant: opt_numwant,
+                key: opt_key,
+            };
+
+            Ok(Request::Announce(request))
+        } else {
+            let request = ScrapeRequest {
+                info_hashes,
+            };
+
+            Ok(Request::Scrape(request))
+        }
     }
 
     /// The info hashes and peer id's that are received are url-encoded byte

aquatic_http_protocol/target/criterion/request-from-bytes/latest/estimates.json

@@ -1 +1 @@
-{"mean":{"confidence_interval":{"confidence_level":0.95,"lower_bound":1540.8390649054281,"upper_bound":1558.4887891210565},"point_estimate":1549.4245398703538,"standard_error":4.517692089474024},"median":{"confidence_interval":{"confidence_level":0.95,"lower_bound":1506.8523919746872,"upper_bound":1513.8541497552624},"point_estimate":1510.1168042545828,"standard_error":1.754123550033509},"median_abs_dev":{"confidence_interval":{"confidence_level":0.95,"lower_bound":45.152694401086514,"upper_bound":54.71728947836145},"point_estimate":49.43491905391041,"standard_error":2.470252770404891},"slope":{"confidence_interval":{"confidence_level":0.95,"lower_bound":1528.8205165274087,"upper_bound":1544.1141471464193},"point_estimate":1536.161119110567,"standard_error":3.9201641311296234},"std_dev":{"confidence_interval":{"confidence_level":0.95,"lower_bound":121.9529239151999,"upper_bound":162.2045693733035},"point_estimate":142.20072433991157,"standard_error":10.304640393049143}}
+{"mean":{"confidence_interval":{"confidence_level":0.95,"lower_bound":725.4551837276395,"upper_bound":735.0471255185286},"point_estimate":730.0304941357815,"standard_error":2.44315882364003},"median":{"confidence_interval":{"confidence_level":0.95,"lower_bound":710.2692059508443,"upper_bound":713.9114247675105},"point_estimate":711.6918749031568,"standard_error":0.939200067715216},"median_abs_dev":{"confidence_interval":{"confidence_level":0.95,"lower_bound":21.74599816051584,"upper_bound":25.76476135212916},"point_estimate":23.51684998144059,"standard_error":1.0048983533788292},"slope":{"confidence_interval":{"confidence_level":0.95,"lower_bound":720.2235427692906,"upper_bound":726.7290589621539},"point_estimate":723.3388090543597,"standard_error":1.655406933329102},"std_dev":{"confidence_interval":{"confidence_level":0.95,"lower_bound":60.65633946284278,"upper_bound":94.33752282501924},"point_estimate":77.10614491661927,"standard_error":8.653988561946822}}

aquatic_http_protocol/target/criterion/request-from-bytes/latest/tukey.json

@@ -1 +1 @@
-[1272.4120390677972,1377.6532072370742,1658.2963223551465,1763.5374905244237]
+[598.8933477988796,648.0355807205904,779.0815351784856,828.2237681001964]