[IoPoll] Take Time\Duration in Context::wait()

[nicolas-grekas]

Q	A
Bug fix?	no
New feature?	no
Deprecations?	no
Issues	-
License	MIT

Depends on #642, on top of which this branch sits — only the last commit belongs here.

php/php-src#23092 is merged and ships in 8.6 Beta 1, alongside Time\Duration. It replaces the $timeoutSeconds/$timeoutMicroseconds arguments of Io\Poll\Context::wait() by a single ?Time\Duration $timeout:

-public function wait(?int $timeoutSeconds = null, int $timeoutMicroseconds = 0, ?int $maxEvents = null): array
+public function wait(?\Time\Duration $timeout = null, ?int $maxEvents = null): array

The vendored phpt files get exactly the edits php-src made in that commit: wait(0) becomes wait(Time\Duration::fromSeconds(0)) and wait(0, 100000) becomes wait(Time\Duration::fromMicroseconds(100000)). Every file that was a byte-identical copy of ext/standard/tests/poll/ still is; the three that already differed (poll.inc, poll_stream_sock_rw_close.phpt, poll_stream_sock_rw_multi_level.phpt, all stale rather than adapted) keep exactly the differences they had.

symfony/polyfill-time is deliberately not a dependency: the polyfill never builds a Duration, it only reads ->negative, ->seconds and ->nanoseconds off the instance it is handed, and the type declaration is only resolved when a non-null value is passed. Calling wait() or wait(null) therefore works without it, and a caller that has a Duration to pass necessarily has the class already. The component README says where to get it.

One divergence worth a decision: native reports the negative-timeout error as argument #2,

if (timeout->duration.negative) {
    zend_argument_value_error(2, "must not be negative");

which looks like a leftover from the old signature, since $timeout is argument #1 and #2 is now $maxEvents. No phpt covers it. The polyfill reports #1; if upstream keeps #2 I will align.

[stof]

which looks like a leftover from the old signature, since $timeout is argument #1 and #2 is now $maxEvents. No phpt covers it. The polyfill reports #1; if upstream keeps #2 I will align.

I suggest reporting an issue upstream about that.

[stof]

the three that already differed (poll.inc, poll_stream_sock_rw_close.phpt, poll_stream_sock_rw_multi_level.phpt, all stale rather than adapted) keep exactly the differences they had.

do they differ intentionally or because php-src updated them after we vendored them ? In the later case, maybe we should sync them.

[TimWolla]

I suggest reporting an issue upstream about that.

php/php-src#23099

[nicolas-grekas]

php/php-src#23099 settles the argument number question: the negative timeout is reported as argument #1, which is what this polyfill already did, so there is nothing to align. Its new poll_ctx_wait.phpt is now vendored here byte-identical and passes as is, both messages included.