Hi,
If you trigger the new optional NetBSD CI task, the oauthvalidator
tests implode[1]. Apparently that OS's kevent() doesn't like zero
relative timeouts for EVFILT_TIMER[2]. I see that you worked around
the same problem for Linux timerfd already by rounding 0 up to 1, so
we could just do the same here, and it passes with the attached. A
cute alternative, not tested, might be to put NOTE_ABSTIME into fflag
if timeout == 0 (then it's an absolute time in the past, which should
fire immediately).
But I'm curious, how hard would it be to do this ↓ instead and not
have that problem on any OS?
* There might be an optimization opportunity here: if timeout == 0, we
* could signal drive_request to immediately call
* curl_multi_socket_action, rather than returning all the way up the
* stack only to come right back. But it's not clear that the additional
* code complexity is worth it.
[1] https://cirrus-ci.com/task/6354435774873600
[2]
https://github.com/NetBSD/src/blob/67c7c4658e77aa4534b6aac8c041d77097c5e722/sys/kern/kern_event.c#L1375
From 7deb153caf552c9bcb380f88eddbca94be33a0c2 Mon Sep 17 00:00:00 2001
From: Thomas Munro <thomas.mu...@gmail.com>
Date: Sat, 1 Mar 2025 00:27:01 +1300
Subject: [PATCH] Fix OAUTH on NetBSD.
NetBSD's EVFILT_TIMER doesn't like zero relative timeouts and fails with
EINVAL. Steal the workaround from the same problem on Linux from a few
lines up: round 0 up to 1.
This could be seen on the optional NetBSD CI task, but not on the NetBSD
build farm animals because they aren't finding curl and testing this
code.
---
src/interfaces/libpq/fe-auth-oauth-curl.c | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/src/interfaces/libpq/fe-auth-oauth-curl.c b/src/interfaces/libpq/fe-auth-oauth-curl.c
index ae339579f88..6e60a81574d 100644
--- a/src/interfaces/libpq/fe-auth-oauth-curl.c
+++ b/src/interfaces/libpq/fe-auth-oauth-curl.c
@@ -1363,6 +1363,16 @@ set_timer(struct async_ctx *actx, long timeout)
#ifdef HAVE_SYS_EVENT_H
struct kevent ev;
+#ifdef __NetBSD__
+
+ /*
+ * Work around NetBSD's rejection of zero timeouts (EINVAL), a bit like
+ * timerfd above.
+ */
+ if (timeout == 0)
+ timeout = 1;
+#endif
+
/* Enable/disable the timer itself. */
EV_SET(&ev, 1, EVFILT_TIMER, timeout < 0 ? EV_DELETE : (EV_ADD | EV_ONESHOT),
0, timeout, 0);
--
2.48.1
