Improve the performance of the unique function by:
1. Pre-allocating map capacity with len(s) to avoid frequent map resizing
2. Pre-allocating result slice capacity with len(s) to reduce append overhead
3. Reducing the number of traversals performs well under the condition of a large number of elements
These changes maintain the original behavior (preserving element order)
while reducing memory allocation operations, especially effective for
large slices (100k+ elements) with benchmark showing ~25% speedup.
No breaking changes, the function signature and output order remain unchanged.
We should use `(*regexp.Regexp).MatchString` instead of
`(*regexp.Regexp).Match([]byte(...))` when matching string to avoid
unnecessary `[]byte` conversions and reduce allocations.
Example benchmark:
var allowedOrigin = regexp.MustCompile(".*.example.com")
func BenchmarkMatch(b *testing.B) {
for i := 0; i < b.N; i++ {
if match := allowedOrigin.Match([]byte("www.example.com")); !match {
b.Fail()
}
}
}
func BenchmarkMatchString(b *testing.B) {
for i := 0; i < b.N; i++ {
if match := allowedOrigin.MatchString("wwww.example.com"); !match {
b.Fail()
}
}
}
goos: linux
goarch: amd64
pkg: github.com/gotify/server/v2/api/stream
cpu: AMD Ryzen 7 PRO 4750U with Radeon Graphics
BenchmarkMatch-16 2076819 647.7 ns/op 16 B/op 1 allocs/op
BenchmarkMatchString-16 2536326 442.0 ns/op 0 B/op 0 allocs/op
PASS
ok github.com/gotify/server/v2/api/stream 3.552s
Signed-off-by: Eng Zer Jun <engzerjun@gmail.com>
GR = goroutine
[GR#1] http server gets closed
[GR#2] client.NotifyClose() will be executed
[GR#2] client.once.Do will be executed (lock's client.once.m)
[GR#1] stream.Close will be executed (lock's stream.lock)
[GR#1] client.Close will be executed (waits for client.once.m)
[GR#2] stream.remove will be executed (waits for stream.lock)
GR#1 holds lock stream.lock and waits for client.once.m
GR#2 holds lock client.once.m and waits for stream.lock
We prevent the deadlock with releasing the client.once.m lock earlier.
Colin
昨夜雨疏风骤