docs: document repeatable LFS upload path

2026-08-02 15:17:31 +08:00
parent 98e6cadcb6
commit f2bf297e84
3 changed files with 145 additions and 1 deletions
@@ -23,6 +23,24 @@ GIT_LFS_SKIP_SMUDGE=1 git clone git@192.168.10.248:cogito/calculet-npu-research-

 随后可按需执行 `git lfs pull --include='archives/<路径>'`。

+### 从当前网络追加 LFS 大文件
+
+Gitea 当前对 SSH LFS 上传返回的内部地址是 `http://gitea/...`。匿名克隆和 LFS 下载通过上面的 IP 地址已经验收；如果贡献者电脑不能解析主机名 `gitea`，SSH 推送 LFS 时可在另一个终端启动仅监听本机回环地址的临时转发：
+
+```bash
+python3 scripts/lfs-host-proxy.py --port 18765
+```
+
+然后在仓库终端执行：
+
+```bash
+http_proxy=http://127.0.0.1:18765 \
+HTTP_PROXY=http://127.0.0.1:18765 \
+no_proxy= NO_PROXY= git push
+```
+
+推送完成后用 `Ctrl-C` 关闭转发器。长期方案是在 Gitea 服务端把 `ROOT_URL`/LFS 外部地址改成团队实际可解析的 IP 或 DNS 名称。
+
 ## 目录

 | 路径 | 内容 | 用途 |
@@ -1,5 +1,5 @@
 174185a34cee598ff33cf4b2b75dbe1688d1000478f95f37dcc3f24ecb11cf46  ./.gitattributes
-3f01c5ebf226064b0f0992ea7506c038dfb9d453e35d2864db06f91966b5baaf  ./README.md
+0b94311217982321bcd106d3170256676b2313eda512af1519099b236a5290eb  ./README.md
 6dda8faa0e2a14362eac47dcdc192094e631fc47a37e978eb0f745e556b4e7be  ./checksums/main-archives.sha256
 4b1b7a92f9502c7d84a240888fac838d43af2bfbe7c821468d148a643dce5d6a  ./documentation/Calculet Runtime API C++ 开发文档-v78-20260710_074837.pdf
 4a7c3b83977fc853eebcd8e306dbd9a4f846d9cfb03c89214a088c7034ec632c  ./documentation/VPN接入及远程连接.pdf
@@ -269,6 +269,7 @@ c16023ac93e955138329561a82a0b5932ea168acb1346ec05197bee8f8b96af4  ./reports/Calc
 eb0692b1e09e8c0049ecddc4ecaf84860a4b31375535bbed3716b983f2579b30  ./reports/Calculet-NPU-逐算子优化实验手册-20260802.md
 a81d9dd8e5b8ae475f4eabdb2f187381ea4c45d77aa352c1d7ee30d2dea22502  ./reports/calculet_package_validate.rb
 29727a6e2734c9458ff1853a4f94e7fdf1e109d2fa6b51cd7062a5bd11924479  ./scripts/clone-bundle.sh
+363b3bf7e2c3d53df0c95843b770e1922b19b044e8be01e6acef088401821747  ./scripts/lfs-host-proxy.py
 fa1c6a60eab70ae3ee215f4850313e1abf1ed5765bfd9860e7b3042afa638e17  ./scripts/reassemble-archives.sh
 15e8c8711c017cb9a31a1499ad9f8479216eff47427a303cafbdc34959f797c2  ./scripts/verify-repository.sh
 46ce4c2a910cadf07a9f45d3f2ad0b5a151fbbc9c60a82bbcbb354e0bd71a00e  ./source/baseline-expanded/calrt-sdk-0.7.6/usr/local/include/calrt/calrt.h
@@ -0,0 +1,125 @@
+#!/usr/bin/env python3
+import argparse
+import http.client
+import shutil
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from urllib.parse import urlsplit, urlunsplit
+
+
+HOP_BY_HOP = {
+    "connection",
+    "keep-alive",
+    "proxy-authenticate",
+    "proxy-authorization",
+    "proxy-connection",
+    "te",
+    "trailers",
+    "transfer-encoding",
+    "upgrade",
+}
+
+
+class GiteaProxy(BaseHTTPRequestHandler):
+    protocol_version = "HTTP/1.1"
+    target_host = "192.168.10.248"
+
+    def do_GET(self):
+        self._forward()
+
+    def do_HEAD(self):
+        self._forward(send_body=False)
+
+    def do_POST(self):
+        self._forward()
+
+    def do_PUT(self):
+        self._forward()
+
+    def _forward(self, send_body=True):
+        parsed = urlsplit(self.path)
+        path = urlunsplit(("", "", parsed.path or "/", parsed.query, ""))
+        print(
+            f"{self.command} content-length={self.headers.get('Content-Length')} "
+            f"transfer-encoding={self.headers.get('Transfer-Encoding')}",
+            flush=True,
+        )
+        upstream = http.client.HTTPConnection(self.target_host, 80, timeout=3600)
+        upstream.putrequest(
+            self.command,
+            path,
+            skip_host=True,
+            skip_accept_encoding=True,
+        )
+
+        for name, value in self.headers.items():
+            lower_name = name.lower()
+            if lower_name in HOP_BY_HOP:
+                continue
+            upstream.putheader(name, "gitea" if lower_name == "host" else value)
+        transfer_encoding = self.headers.get("Transfer-Encoding", "").lower()
+        if transfer_encoding == "chunked":
+            upstream.putheader("Transfer-Encoding", "chunked")
+        upstream.endheaders()
+
+        if transfer_encoding == "chunked":
+            self._forward_chunked_body(upstream)
+        else:
+            content_length = int(self.headers.get("Content-Length", "0"))
+            remaining = content_length
+            while remaining:
+                chunk = self.rfile.read(min(1024 * 1024, remaining))
+                if not chunk:
+                    raise ConnectionError("request body ended early")
+                upstream.send(chunk)
+                remaining -= len(chunk)
+
+        response = upstream.getresponse()
+        self.send_response(response.status, response.reason)
+        for name, value in response.getheaders():
+            if name.lower() not in HOP_BY_HOP:
+                self.send_header(name, value)
+        self.send_header("Connection", "close")
+        self.end_headers()
+        self.close_connection = True
+
+        if send_body:
+            shutil.copyfileobj(response, self.wfile, 1024 * 1024)
+        upstream.close()
+
+    def _forward_chunked_body(self, upstream):
+        while True:
+            size_line = self.rfile.readline()
+            if not size_line:
+                raise ConnectionError("chunked request ended before size line")
+            upstream.send(size_line)
+            chunk_size = int(size_line.split(b";", 1)[0].strip(), 16)
+            if chunk_size == 0:
+                while True:
+                    trailer_line = self.rfile.readline()
+                    upstream.send(trailer_line)
+                    if trailer_line in (b"\r\n", b"\n", b""):
+                        return
+
+            remaining = chunk_size + 2
+            while remaining:
+                data = self.rfile.read(min(1024 * 1024, remaining))
+                if not data:
+                    raise ConnectionError("chunked request ended inside data")
+                upstream.send(data)
+                remaining -= len(data)
+
+    def log_message(self, fmt, *args):
+        print("%s - %s" % (self.address_string(), fmt % args), flush=True)
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--port", type=int, default=18765)
+    args = parser.parse_args()
+    server = ThreadingHTTPServer(("127.0.0.1", args.port), GiteaProxy)
+    print(f"Listening on 127.0.0.1:{args.port}", flush=True)
+    server.serve_forever()
+
+
+if __name__ == "__main__":
+    main()