Release proposal: v1.39.0

[cjihrig]

Notable changes:

• uv_metrics_idle_time() and UV_METRICS_IDLE_TIME have been added for measuring the amount of time the event loop spends idle.
• uv_udp_using_recvmmsg() has been added to determine if a buffer is large enough for multiple datagrams should be allocated in the allocation callback of uv_udp_recvstart().
• On MinGW, the installation location has been updated to match Unix systems rather than Windows.
• uv_fs_copyfile() now tries to use copy_file_range() when possible.
• The test suite is now reported to pass on Darwin ARM64 (Apple Silicon).
• uv_{get,set}_process_title() now returns an error on platforms where uv_setup_args() is required, but has not yet been called.
• The _POSIX_PATH_MAX constant is no longer used, which could lead to buffer overflows in uv_fs_readlink() and uv_fs_realpath().

[vtjnash]

There's probably a handful of other PRs nearly ready to merge also, but waiting on either a quick correction or review!

[bnoordhuis]

Is #2725 really ready for release? It's a new API and those should ideally receive at least two sign-offs from maintainers.

Although it received three, it looks like two of them are from March, for an old incarnation of the PR.

[cjihrig]

@bnoordhuis I guess that's a good question. I'm fine with waiting on a release while that is determined.

[trevnorris]

@bnoordhuis I've uploaded the diff from the first accepted patch in March and the one that landed in v1.x to https://gist.github.com/trevnorris/8183ea638732f908ebb69f962e50b4e0.

I don't have any issue with waiting, but for posterity I want to point out that there was almost no change between the patch as it was accepted in March and as it landed today. The only major differences between the two are a mistake that accidentally broke ABI:

@@ include/uv.h: typedef struct uv_utsname_s uv_utsname_t;
  typedef enum {
 -  UV_LOOP_BLOCK_SIGNAL
-+  UV_LOOP_BLOCK_SIGNAL = 1,
-+  UV_METRICS_IDLE_TIME = 2
++  UV_LOOP_BLOCK_SIGNAL = 0,
++  UV_METRICS_IDLE_TIME
  } uv_loop_option;

and a bug fix for calculating idle time in Windows.

@@ src/win/core.c: static void uv__poll_wine(uv_loop_t* loop, DWORD timeout) {
 +      timeout = user_timeout;
 +      reset_timeout = 0;
 +    }
++
++    /* Placed here because on success the loop will break whether there is an
++     * empty package or not, or if GetQueuedCompletionStatus returned early then
++     * the timeout will be updated and the loop will run again. In either case
++     * the idle time will need to be updated.
++     */
++    uv__metrics_update_idle_time(loop);
 +
      if (overlapped) {
-+      uv__metrics_update_idle_time(loop);

@@ src/win/core.c: static void uv__poll(uv_loop_t* loop, DWORD timeout) {
 +    if (reset_timeout != 0) {
 +      timeout = user_timeout;
 +      reset_timeout = 0;
 +    }
++
++    /* Placed here because on success the loop will break whether there is an
++     * empty package or not, or if GetQueuedCompletionStatus returned early then
++     * the timeout will be updated and the loop will run again. In either case
++     * the idle time will need to be updated.
++     */
++    uv__metrics_update_idle_time(loop);
 +
      if (success) {
        for (i = 0; i < count; i++) {
          /* Package was dequeued, but see if it is not a empty package
           * meant only to wake us up.
           */
          if (overlappeds[i].lpOverlapped) {
-+          uv__metrics_update_idle_time(loop);

[bnoordhuis]

Okay, that seems harmless enough, although I can't really evaluate the windows changes.

[musm]

Any chance #2910 can make it?

[ulrikstrid]

If there was is any chance I would love it if #2951 got in as well. Let me know if there's anything I can do to make it happen.

[cjihrig]

libuv CI: https://ci.nodejs.org/view/libuv/job/libuv-test-commit/2001/ (✔️)
Node.js integration CI: https://ci.nodejs.org/view/libuv/job/libuv-in-node/146/

[cjihrig]

Small status update. The changes in this release are causing some problems with the Node.js test suite:

1. 07e4168 requires some additional calls to uv_setup_args() in the Node test suite. Addressed in test,doc: add missing uv_setup_args() calls nodejs/node#34751.
2. 12be29f (specifically, the change to uv_shutdown()) is causing Node's test/parallel/test-http2-buffersize.js test to become extremely flaky on macOS. Still investigating this one.

[cjihrig]

@bnoordhuis do you have any ideas on item 2 above?

[bnoordhuis]

12be29f (specifically, the change to uv_shutdown()) is causing Node's test/parallel/test-http2-buffersize.js test to become extremely flaky on macOS.

I'm not quite sure why but it's the clearing of the flags in the error path that causes the flakiness. With the patch below, the test starts passing again:

diff --git a/src/unix/stream.c b/src/unix/stream.c
index f86cf11e..6c0f5ad1 100644
--- a/src/unix/stream.c
+++ b/src/unix/stream.c
@@ -1189,7 +1189,6 @@ static void uv__read(uv_stream_t* stream) {
 #endif
       } else {
         /* Error. User should call uv_close(). */
-        stream->flags &= ~(UV_HANDLE_READABLE | UV_HANDLE_WRITABLE);
         stream->read_cb(stream, UV__ERR(errno), &buf);
         if (stream->flags & UV_HANDLE_READING) {
           stream->flags &= ~UV_HANDLE_READING;

Suggestion: revert the commit for now so we can move forward with the release? I'll open a pull request.

[cjihrig]

Partial revert in #2967

Revert in #2968