Add builtin type Time by ngraybeal · Pull Request #1788 · sifiveinc/wake

Nathan Graybeal (ngraybeal) · 2026-04-13T21:46:46Z

Creates a builtin type Time for the main purpose of handling timestamps.

Time is stored as a unit of nanoseconds since linux epoch.

This uses the chrono library to help safely handle time. Since we are using c++17, I'm still using strftime to format the string. This does augment strftime with a %f formatter for specifying subsecond time.

Functions include:
getTime
formatTime in UTC
Time conversion to nanoseconds ,milliseconds and seconds (integer, double, double)
Println to specify format and timezone.

I did rename the database Time struct to DBTime to avoid conflicts.

This is a more narrowly scoped feature set for Time than original PR. The biggest difference is the avoidance of Timezone due to its complexities. The only way timezone can be specified is through the print statement, otherwise everything is handled in UTC.

Richard Xia (richardxia) · 2026-05-12T22:06:34Z

+# Example:
+#   printTimeLnLevel "%H:%M:%S" "UTC" (getTime Unit) logWarning "something happened"
+#   printTimeLnLevel "%H:%M:%S" "America/Los_Angeles" (getTime Unit) logReport "hello"
+export def printTimeLnLevel (format: String) (timezone: String) (time: Time) (logLevel: LogLevel) (message: String): Unit =


I'm wondering if it makes sense to expose the timezone as a parameter. Other than UTC, you wouldn't want any code to hardcode a time zone as a string, since it will be incorrect for any person not in that time zone.

Should there just be two versions of this function: one for always printing UTC, and the other for using the TZ environment variable? Or perhaps always use TZ and default to UTC if it's not set?

Your comment made sense, though i went with option 3 (not listed) and just followed glibc default behavior. Use TZ if set, otherwise default to system configuration.

This doesn't impact format time function in cpp or wake.

Abrar Quazi (AbrarQuazi) · 2026-05-13T14:47:46Z

+
+// Pre-process a format string to expand %f sub-second specifiers before passing to strftime.
+// Supported: %f (9 digits), %.f (trimmed), %.Nf (dot + N digits), %Nf (N digits) where N is 1-9
+static std::string expand_subsecond_formats(const char *fmt, int64_t nanoseconds) {


I know that theres no proper library in c++17 for subsecond support, but what's preventing us from just importing over https://github.com/howardhinnant/date instead of trying to roll out own implementation?

It doesn't support subsecond formatting. I feel the implementation is pretty straightforward and narrow in scope, so we can avoid the extra baggage.

Yeah, I recently gained a new appreciation for the "vendor or reimplement your (simple) dependencies" mentality. Especially since we'd only be exposing a minimal surface area anyway, may as well keep the dependency tree well-trimmed.

Abrar Quazi (AbrarQuazi) · 2026-05-13T15:05:02Z

+#
+# Example:
+#   def now = getTime Unit
+export def getTime (_: a): Time =


Should either mark this function as "unsafe" or add documentation that this breaks caching

similar comment for time_print

Probably wouldn't hurt to have this be either unsafe_getTime or getCurrentTime just to drive the point home, yeah, but the prefix is more a means of flagging the functions during reviews than it is any concrete policy. The main goal is just to make sure that anyone working with its values is aware of the potential unreproducibility, and getting the system time is already intuitively unreproducible.

Printing doesn't need the flag though -- it might have a user-observable side effect, but since it returns Unit nothing within wake can observe that it's any different than any other call to it.

Abrar Quazi (AbrarQuazi)

Some comments/questions

* stores current epoch time as nanoseconds * Add methods: getTime, formatTime in UTC get time in nanoseconds , milliseconds, and seconds Print time with any format and timezone

Sam May (ag-eitilt)

Not sure how many of these are better to fix here, vs fixing in a followup, but there's no big issues I had with any of it.

Sam May (ag-eitilt) · 2026-05-15T18:37:02Z

+# The Time type is a builtin of the wake language itself.
+#
+# A Time object represents a point in time with nanosecond
+# precision, stored as nanoseconds since the Unix epoch
+# (January 1, 1970 00:00:00 UTC).
+#
+from builtin export type Time


Can we export this in package Time to keep it all in one place?

Sam May (ag-eitilt) · 2026-05-15T18:38:52Z

+#
+# Example:
+#   def now = getTime Unit
+export def getTime (_: a): Time =


Probably wouldn't hurt to have this be either unsafe_getTime or getCurrentTime just to drive the point home, yeah, but the prefix is more a means of flagging the functions during reviews than it is any concrete policy. The main goal is just to make sure that anyone working with its values is aware of the potential unreproducibility, and getting the system time is already intuitively unreproducible.

Printing doesn't need the flag though -- it might have a user-observable side effect, but since it returns Unit nothing within wake can observe that it's any different than any other call to it.

Sam May (ag-eitilt) · 2026-05-15T18:45:41Z

+# Example:
+#   printTimeLn (getTime Unit) "build started"
+export def printTimeLn (time: Time) (message: String): Unit =
+    printTimeLnLevel '[%Y-%m-%dT%H:%M:%S%.3f%z]' time logReport message


Nit:

Suggested change

printTimeLnLevel '[%Y-%m-%dT%H:%M:%S%.3f%z]' time logReport message

printTimeLnLevel '[%Y-%m-%dT%H:%M:%S%.3f%z] ' time logReport message

I prefer not having the space to give more control to the caller.

Sam May (ag-eitilt) · 2026-05-15T18:49:15Z

+
+// Pre-process a format string to expand %f sub-second specifiers before passing to strftime.
+// Supported: %f (9 digits), %.f (trimmed), %.Nf (dot + N digits), %Nf (N digits) where N is 1-9
+static std::string expand_subsecond_formats(const char *fmt, int64_t nanoseconds) {


Yeah, I recently gained a new appreciation for the "vendor or reimplement your (simple) dependencies" mentality. Especially since we'd only be exposing a minimal surface area anyway, may as well keep the dependency tree well-trimmed.

Sam May (ag-eitilt) · 2026-05-15T18:53:14Z

+
+    // Look ahead from '%'
+    if (p[1] == '.' && p[2] >= '1' && p[2] <= '9' && p[3] == 'f') {
+      // %.Nf — dot + fixed N digits (N = 1-9)


Sneaky AI even sneaks its em-dashes into otherwise-ASCII code comments. 😂

Sam May (ag-eitilt) · 2026-05-15T19:06:14Z

+    if (timezone[0] == '\0') {
+      unsetenv("TZ");
+    } else {
+      setenv("TZ", timezone, 1);
+    }


Like I mentioned in the quick scan I did live, I would really like it if this syscall hadn't been implemented to depend on the user environment. Nothing we can do, though.

This is probably not multithread safe right? I know that since wakes runtime is single threaded - so its probably not a big deal right now - but in the future this might be an issue if want to make the runtime multihreaded some how. Would this be multiprocess/multiwake safe though?

I went ahead and added a mutex lock.

c++20 does have a localtime_rz to avoid all this

Sam May (ag-eitilt) · 2026-05-15T19:08:22Z

+    } else {
+      // Not a sub-second specifier — pass through for strftime
+      result += '%';
+    }


We need a special case to skip over both percents in %%foo.

Sam May (ag-eitilt) · 2026-05-15T19:11:13Z

+    } else if (p[1] >= '1' && p[1] <= '9' && p[2] == 'f') {
+      // %Nf — fixed N digits, no dot (N = 1-9)
+      int n = p[1] - '0';
+      result.append(nanos_str, n);
+      p += 2;


It would be nice to have rounding, not truncation, but if the standard sprintf truncates we definitely want to match that behaviour.

Interesting point. though does have the added complication of carrying over into seconds if it all rounds up.

Sam May (ag-eitilt) · 2026-05-15T19:12:19Z

+
+  std::string expanded = expand_subsecond_formats(fmt, nanoseconds);
+  char buffer[256];
+  strftime(buffer, sizeof(buffer), expanded.c_str(), &tm_info);


Is there any chance this could be abused to construct malicious formatting? So long as you make the %% fix, I'm having trouble constructing a format string which the expansion would collapse but the strftime would think was a length for some surrounding interpolation, and I'm also having trouble thinking of a case where even if that did happen it would be able to do anything significant. I'm not a black hat, though.

Add test Update comments

Abrar Quazi (AbrarQuazi)

LGTM

Nathan Graybeal (ngraybeal) requested review from Abrar Quazi (AbrarQuazi), Sam May (ag-eitilt) and Will Dietz (dtzSiFive) April 13, 2026 21:46

Nathan Graybeal (ngraybeal) force-pushed the nathang/time_builtin branch from 590805a to df5a066 Compare April 13, 2026 22:05

Will Dietz (dtzSiFive) reviewed May 8, 2026

View reviewed changes

Comment thread share/wake/lib/system/time.wake Outdated

Richard Xia (richardxia) reviewed May 12, 2026

View reviewed changes

Abrar Quazi (AbrarQuazi) reviewed May 13, 2026

View reviewed changes

Comment thread src/runtime/time.cpp

Abrar Quazi (AbrarQuazi) reviewed May 13, 2026

View reviewed changes

Comment thread src/runtime/time.cpp Outdated

Abrar Quazi (AbrarQuazi) reviewed May 13, 2026

View reviewed changes

Nathan Graybeal (ngraybeal) added 7 commits May 14, 2026 23:46

Add builtin type Time

417b39d

* stores current epoch time as nanoseconds * Add methods: getTime, formatTime in UTC get time in nanoseconds , milliseconds, and seconds Print time with any format and timezone

Comment update

252b71a

time_print to PRIM_IMPURE

80a40eb

Rework comment

9c5cf33

Drop timezone parameter in time_print

f0cf04b

Align function names

0e16137

formatting

d0358d3

Nathan Graybeal (ngraybeal) force-pushed the nathang/time_builtin branch from df5a066 to d0358d3 Compare May 15, 2026 07:52

Fix test

0cafa05

Sam May (ag-eitilt) approved these changes May 15, 2026

View reviewed changes

Bug fix for %%

9bcae03

Add test Update comments

Abrar Quazi (AbrarQuazi) approved these changes May 20, 2026

View reviewed changes

Add some thread safety

a71d6ff

Nathan Graybeal (ngraybeal) merged commit 5e15b6b into master May 22, 2026
11 checks passed

Nathan Graybeal (ngraybeal) deleted the nathang/time_builtin branch May 22, 2026 15:39

	printTimeLnLevel '[%Y-%m-%dT%H:%M:%S%.3f%z]' time logReport message
	printTimeLnLevel '[%Y-%m-%dT%H:%M:%S%.3f%z] ' time logReport message

Conversation

Nathan Graybeal (ngraybeal) commented Apr 13, 2026

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Nathan Graybeal (ngraybeal) May 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

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!

Abrar Quazi (AbrarQuazi) May 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Abrar Quazi (AbrarQuazi) left a comment

Choose a reason for hiding this comment

Uh oh!

Sam May (ag-eitilt) 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!

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!

Choose a reason for hiding this comment

Uh oh!

Abrar Quazi (AbrarQuazi) left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Nathan Graybeal (ngraybeal) May 15, 2026 •

edited

Loading

Abrar Quazi (AbrarQuazi) May 13, 2026 •

edited

Loading