Refactor Internal Time Retrieval Handling for Improved Consistency and Resolution #19202

marc-mabe · 2025-07-21T13:42:01Z

This pull request refactors how PHP internally handles current time retrieval and time measurement, introducing a more consistent and portable approach. The primary goals are to encapsulate platform-specific differences, improve time resolution, and prepare for long-term compatibility (e.g., Y2038 on WIN64).

Key Changes

Introduced zend_time.h
A new internal header that abstracts system time functions, replacing direct usage of <time.h> and related platform-specific APIs.
Added zend_realtime_spec
A unified wrapper around clock_gettime(), timespec_get(), gettimeofday(), and time() using the real/wall clock.
Returns a timespec structure with nanosecond precision (when available).
Defined zend_realtime_get macro
A lightweight macro wrapping time(NULL) for simple, low-resolution time retrieval.
Introduced zend_monotime_fallback
A macro that maps to zend_hrtime or falls back to zend_realtime_spec.
Useful for time measurements (e.g. timeout handling) where monotonic time is preferred, but wall time is an acceptable fallback.
Provided helper macros
Added utilities to simplify usage of timeval and timespec structures.
Replaced direct system time API usage
Internal calls to system time functions now use zend_realtime_* or zend_monotime_fallback.
Standardized time representation
Transitioned to timespec for current time where appropriate, while retaining timeval for files and stream-related functionality.

Benefits

Improved Portability and Readability
Platform-specific time logic is encapsulated, reducing conditionals and improving clarity.
Guaranteed Time API Availability
The new abstraction ensures time can always be retrieved via a fallback chain:
clock_gettime() → timespec_get() → gettimeofday() → time()
Y2038 Compatibility on WIN64
Addresses issues caused by long in timeval.tv_sec on 64-bit Windows systems.

Related issue: #17856 – time() and Y2038 problem on 64-bit Windows
Improved Resolution for Time Functions
microtime() and gettimeofday(true) now offer better time precision without breaking backward compatibility.
Removed Redundant Availability Checks
Previously unverified use of gettimeofday() has been replaced by a robust fallback mechanism.
Conditional checks for microtime(), gettimeofday(), and uniqid() have been removed.

ext/standard/uniqid.c

…andling

marc-mabe · 2025-10-19T03:15:53Z

@TimWolla I think this is ready to go now. Please could you have another look and/or point me to the right person to ping?

@bukka Please could you help me for how to fix the failing unit test (expecting gettimeofday but now using zend_monotime_fallback)

TimWolla

Had a rudimentary first look.

TimWolla · 2025-10-19T17:17:05Z

Zend/zend_time.h

Can you please make all the macros that could be (inline) functions a proper function? I think that should be all of them.

Please also properly “namespace” all the symbols by prefixing them with zend_time_*.

Can you please make all the macros that could be (inline) functions a proper function? I think that should be all of them.

👍

Please also properly “namespace” all the symbols by prefixing them with zend_time_*.

For the zend_realtime_spec, zend_realtime_get and zend_monotime_fallback functions I felt it sounds more natural and it's obvious these functions belong to time.
Would zend_time_real_get, zend_time_real_spec and zend_time_mono_fallback better match with your expected naming scheme?

Would zend_time_real_get, zend_time_real_spec and zend_time_mono_fallback better match with your expected naming scheme?

Yes. While it might sound a little awkward, I find it valuable to easily locate where the function is declared.

TimWolla · 2025-10-19T17:17:51Z

Zend/zend_time.c

+ZEND_API void zend_realtime_spec(struct timespec *ts) {
+#if defined(HAVE_CLOCK_GETTIME)
+
+	clock_gettime(CLOCK_REALTIME, ts);
+
+#elif defined(HAVE_TIMESPEC_GET)
+
+	timespec_get(ts, TIME_UTC);
+
+#elif defined(HAVE_GETTIMEOFDAY)
+
+	struct timeval tv;
+	gettimeofday(&tv, NULL);
+	zend_time_val2spec(tv, ts);
+
+#else
+
+	ts->tv_sec  = zend_realtime_get();
+	ts->tv_nsec = 0;
+
+#endif
+}


Are these all infallible at runtime? Complexity-wise this could also be an inline function.

Most of these functions can return an error in case of using wrong/unsupported arguments.
On the same time here only the most basic functionality is used and in the previous implementation error returns where not checked.

in the previous implementation error returns where not checked.

Sure, but I don't think this is a reason that this should not be improved. Though it certainly raises the question if a broken clock is even recoverable.

Perhaps add a (void) in front to make it explicit that we don't want to handle the errors?

ext/random/random.c

TimWolla

Some more bits. I'm approximately halfway through the PR. Feel free to already make the requested changes with regard to the removal of the macros, this probably saves us both a review cycle.

TimWolla · 2025-10-20T18:47:22Z

ext/date/php_date.c

 	time->us = microsecond;
 }

 static void php_date_get_current_time_with_fraction(time_t *sec, suseconds_t *usec)


This one can probably be removed entirely now.

TimWolla · 2025-10-20T18:49:56Z

ext/date/php_date.h


-PHPAPI time_t php_time(void);
+/* DEPRECATED, use zend_realtime_get() instead */
+#define php_time() zend_realtime_get()


Can you please leave this as a (inline) function and add ZEND_ATTRIBUTE_DEPRECATED so that folks actually notice that this is deprecated?

marc-mabe · 2025-10-21T06:07:54Z

Some more bits. I'm approximately halfway through the PR. Feel free to already make the requested changes with regard to the removal of the macros, this probably saves us both a review cycle.

I'll be able to continue my work on it from tomorrow evening.

marc-mabe force-pushed the current_time_wrapper branch 14 times, most recently from c81c27f to 8c044c8 Compare July 23, 2025 13:19

marc-mabe requested review from SakiTakamachi, TimWolla, derickr, kamil-tekiela and zeriyoshi as code owners July 25, 2025 08:12

marc-mabe mentioned this pull request Jul 25, 2025

Handle broken hrtime in ftp #19219

Closed

marc-mabe force-pushed the current_time_wrapper branch 2 times, most recently from 7e3d98c to fdabd9b Compare July 26, 2025 08:46

TimWolla reviewed Jul 27, 2025

View reviewed changes

ext/standard/uniqid.c Outdated Show resolved Hide resolved

marc-mabe force-pushed the current_time_wrapper branch from 1058493 to 9429859 Compare July 28, 2025 07:02

marc-mabe force-pushed the current_time_wrapper branch from 9429859 to 3aa094c Compare September 21, 2025 05:50

marc-mabe requested a review from devnexen as a code owner September 23, 2025 06:12

github-actions bot added Extension: calendar Extension: pgsql Extension: phar labels Sep 23, 2025

marc-mabe force-pushed the current_time_wrapper branch from 315f346 to 956b7a0 Compare October 4, 2025 19:57

marc-mabe changed the title ~~Refactor basic time usages~~ Refactor Internal Time Retrieval Handling for Improved Consistency and Resolution Oct 5, 2025

marc-mabe force-pushed the current_time_wrapper branch from 956b7a0 to 88ec473 Compare October 5, 2025 05:41

marc-mabe and others added 6 commits October 19, 2025 04:34

Refactor basic time

cdebcf6

sapi/litespeed

5146bcb

ext/ftp: prefer zend_monotime_fallback over zend_hrtime for timeout h…

9bcfa57

…andling

Replace php_time() with zend_realtime_get()

d7ffbbf

Convert zend_realtime_get() to be a macro wrapper around time(NULL)

313143d

Replace time() in favor of zend_realtime_get()

abf0e46

marc-mabe force-pushed the current_time_wrapper branch from 88ec473 to abf0e46 Compare October 19, 2025 02:53

marc-mabe requested a review from TimWolla October 19, 2025 03:14

TimWolla reviewed Oct 19, 2025

View reviewed changes

TimWolla reviewed Oct 20, 2025

View reviewed changes

Refactor Internal Time Retrieval Handling for Improved Consistency and Resolution #19202

Are you sure you want to change the base?

Refactor Internal Time Retrieval Handling for Improved Consistency and Resolution #19202

Conversation

marc-mabe commented Jul 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

marc-mabe commented Oct 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

TimWolla left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

TimWolla left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marc-mabe commented Oct 21, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

marc-mabe commented Jul 21, 2025 •

edited

Loading

marc-mabe commented Oct 19, 2025 •

edited

Loading