Files
zeekay d14b104790 feat(zap): v2 generic API, bufpool, transport allocator + UMA/RDMA cgo
Three orthogonal additions, one wire format.

v2 package — generic ad-hoc schema dispatch (~/work/lux/zap/v2):
- README + IMPOSSIBILITY + BENCH_RESULTS document the v1↔v2 split.
- builder/field/iter/list/pool/registry/schema/view + codegen package
  emit v1-equivalent fast paths from declarative .zap sources.
- _compile_fail_test/ + examples/ cover negative and positive cases.
- v2_test.go exercises the generic dispatch path.
- doc.go in v1 marks v1 deprecated for *new* schema authoring; existing
  v1 buffers stay parseable by both v1 and v2.

bufpool — pooled read slabs (bufpool.go + bufpool_test.go):
- Message gains optional *bufRef; Release() returns the slab to its
  pool when sourced from the pooled read path, GC-managed otherwise.
- ParseHeader returns (data, rootOffset) without allocating *Message,
  used by zapv2 to avoid two allocations per parse.
- bench_read_test.go + node_quic_stream_test.go cover both paths.

Transport allocator + UMA/RDMA cgo (transport/*):
- Transport.Caps()/Buffer interface: capability + slab-typed buffer
  contract. Bytes() vs DevicePtr() decoupled so UMA and GPUDirect
  can share the same call surface.
- DPDK PMD + GPUDirect RDMA + CUDA UMA real cgo backends gated behind
  build tags; stub paths remain default on darwin / non-Linux.
- transport_test.go covers Caps reporting + Buffer lifecycle on the
  default transport.

LLM.md + go.mod follow the additions.
2026-06-04 17:06:59 -07:00

88 lines
2.3 KiB
Go

// Copyright (C) 2025-2026, Lux Industries Inc. All rights reserved.
// See the file LICENSE for licensing terms.
package transport
import (
"context"
"errors"
"sync"
)
// defaultTransport is the always-available baseline. Uses an in-process
// channel pair as the wire — adequate for tests and single-host dev
// rigs. Production deployments wire this to the existing
// github.com/luxfi/zap NodeServer/NodeClient via a small adapter (not
// included here to keep the transport package dependency-free).
type defaultTransport struct {
mu sync.Mutex
outbox chan envelope
inbox chan envelope
closed bool
}
type envelope struct {
peer string
msg []byte
}
func newDefault() Transport {
return &defaultTransport{
outbox: make(chan envelope, 64),
inbox: make(chan envelope, 64),
}
}
func (t *defaultTransport) Name() string { return "default" }
func (t *defaultTransport) Caps() Capabilities {
return Capabilities{GPUResident: false, ZeroCopy: false, MinLatencyMicros: 50}
}
func (t *defaultTransport) Send(ctx context.Context, peer string, msg []byte) error {
t.mu.Lock()
if t.closed {
t.mu.Unlock()
return errors.New("zap/transport: closed")
}
t.mu.Unlock()
// Copy because the caller may reuse the buffer after Send returns
// (the Transport interface contract). Outbox is bounded; we respect
// ctx for back-pressure.
cp := append([]byte(nil), msg...)
select {
case t.outbox <- envelope{peer: peer, msg: cp}:
return nil
case <-ctx.Done():
return ctx.Err()
}
}
func (t *defaultTransport) Recv(ctx context.Context) (string, []byte, error) {
select {
case env := <-t.inbox:
return env.peer, env.msg, nil
case <-ctx.Done():
return "", nil, ctx.Err()
}
}
func (t *defaultTransport) Close() error {
t.mu.Lock()
defer t.mu.Unlock()
if t.closed {
return nil
}
t.closed = true
close(t.outbox)
close(t.inbox)
return nil
}
// Outbox / Inbox are package-private hooks used by the adapter that
// bridges this transport to github.com/luxfi/zap NodeServer/NodeClient.
// Kept un-exported to discourage callers from poking at the channels
// directly — the adapter lives next to the production node code.
func (t *defaultTransport) outboxChan() <-chan envelope { return t.outbox }
func (t *defaultTransport) inboxChan() chan<- envelope { return t.inbox }