Files
zeekay a3245a76f4 Rebrand: Hanzo S3 CLI -- remove all MinIO branding
- Replace copyright headers: "MinIO, Inc." -> "Hanzo AI, Inc." across 411+ Go files
- Replace "MinIO Object Storage stack" -> "Hanzo S3 stack" in all file headers
- Replace all user-facing "MinIO" references -> "Hanzo S3" in Usage, help text, examples
- Replace example alias "myminio" -> "mys3" throughout
- Replace example endpoints play.min.io -> s3.hanzo.ai, dl.min.io -> s3.hanzo.ai
- Replace doc URLs min.io -> hanzo.space
- Rewrite README.md/README_zh_CN.md: product is "Hanzo S3 CLI" (the s3 command)
- Update CONTRIBUTING.md, CONFLICT.md, NOTICE, code_of_conduct.md
- Update all Dockerfiles: labels, entrypoints, image refs -> ghcr.io/hanzos3/cli
- Update Makefile: build output, docker tags, install paths -> s3
- Update docker-buildx.sh: image tags -> ghcr.io/hanzos3/cli
- DO NOT change go.mod module path or import paths (github.com/minio/mc preserved)
- DO NOT change Go identifiers, SDK types, or wire protocol constants
- All unit tests pass
2026-02-21 14:19:07 -08:00

90 lines
3.1 KiB
Go

// Copyright (c) 2015-2022 Hanzo AI, Inc.
//
// This file is part of Hanzo S3 stack
//
// This program is free software: you can redistribute it and/or modify
// it under the terms of the GNU Affero General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// This program is distributed in the hope that it will be useful
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Affero General Public License for more details.
//
// You should have received a copy of the GNU Affero General Public License
// along with this program. If not, see <http://www.gnu.org/licenses/>.
package cmd
import "github.com/rjeczalik/notify"
// Dynamically sized logical channel: a pipe which never blocks even when
// it receives too many elements. Memory consumption is increased and decresed
// on the fly following the number of elements. Here is the benchmark which
// compares a regular channel which a large capacity (like 1 Million) in
// contrast to a PipeChan with an realtime increasing/decreasing capacity.
//
// BenchmarkRegular1M-4 200 172809399 ns/op
// BenchmarkPipeChan1M-4 100 316668338 ns/op
// BenchmarkRegular100K-4 2000 16540648 ns/op
// BenchmarkPipeChan100K-4 1000 31905966 ns/op
// BenchmarkRegular10K-4 20000 1637665 ns/op
// BenchmarkPipeChan10K-4 10000 3324329 ns/op
// BenchmarkRegular1K-4 200000 168512 ns/op
// BenchmarkPipeChan1K-4 100000 550623 ns/op
// PipeChan builds a new dynamically sized channel
func PipeChan(capacity int) (inputCh, outputCh chan notify.EventInfo) {
// A set of channels which store all elements received from input
channels := make(chan chan notify.EventInfo, 1000)
inputCh = make(chan notify.EventInfo, capacity)
// A goroutine which receives elements from inputCh and creates
// new channels when needed.
go func() {
// Create the first channel
currCh := make(chan notify.EventInfo, capacity)
channels <- currCh
for elem := range inputCh {
// Prepare next channel with a double capacity when
// half of the current channel is already filled.
if len(currCh) >= cap(currCh)/2 {
close(currCh)
currCh = make(chan notify.EventInfo, cap(currCh)*2)
channels <- currCh
}
// Prepare next channel with half capacity when
// current channel is 1/4 filled
if len(currCh) >= capacity && len(currCh) <= cap(currCh)/4 {
close(currCh)
currCh = make(chan notify.EventInfo, cap(currCh)/2)
channels <- currCh
}
// Send element to current channel
currCh <- elem
}
close(currCh)
close(channels)
}()
// Copy elements from infinite channel set to the output
outputCh = make(chan notify.EventInfo, capacity)
go func() {
for {
currCh, ok := <-channels
if !ok {
break
}
for v := range currCh {
outputCh <- v
}
}
close(outputCh)
}()
return inputCh, outputCh
}