Issues with time_t in printf()

[mborgerson]

I was using printf() to show some intermediate results in a data collection program. I found that using printf() to show the time and two other variables didn't work as expected:


var1 = 1; var2 = 2;
Serial.printf("now: %lu var1: %u var2: %u\n", now(),var1,var2);
produced: "now: 1725965454 var1: 0 var2: 1"

Something about calling now() inside printf was messing up the following variable, so I put together the following program which illustrates the issue:

#include "TimeLib.h"

uint16_t var1, var2;
void setup() {
  // put your setup code here, to run once:
  Serial.begin(9600);
  setSyncProvider(get_TeensyTime);  // Use the time provided from loader
  delay(2000);
  Serial.println("\n\nTesting now() function in printf");
  var1 = 1;
  var2 = 2;
  Serial.printf("Starting with var1: %u   var2: %u\n", var1, var2);
}

void loop() {
  delay(2000);
  Serial.println("\n");
  Serial.printf("now: %lu\n", now());
  Serial.printf("var1: %u   var2: %u\n", var1, var2);
  Serial.printf("now: %lu  var1: %u   var2: %u\n", now(),var1,var2);
  uint32_t lnow = now();
  Serial.printf("lnow: %lu  var1: %u   var2: %u\n", lnow,var1,var2);
  time_t tnow = now();
  Serial.printf("tnow: %lu  var1: %u   var2: %u\n", tnow,var1,var2);
  Serial.printf("(uint32_t)now(): %lu  var1: %u   var2: %u\n", (uint32_t)now(),var1,var2);
  Serial.printf("(time_t)now():   %lu  var1: %u   var2: %u\n", (time_t)now(),var1,var2);
}

time_t get_TeensyTime(void) {
  return Teensy3Clock.get();
}

Running this code showed anomalous values in three cases, which seems to show that printf() doesn't know how to handle the time_t data type and something gets messed up in its internal parameter stack. I thought that Arduino defined time_t to be an unsigned long. What am I missing?

Testing now() function in printf
Starting with var1: 1   var2: 2

now: 1725965974
var1: 1   var2: 2
now: 1725965974  var1: 0   var2: 1     <--------------
lnow: 1725965974  var1: 1   var2: 2
tnow: 1725965974  var1: 0   var2: 1    <--------------
(uint32_t)now(): 1725965974  var1: 1   var2: 2
(time_t)now():   1725965974  var1: 0   var2: 1   <--------------

now: 1725965976
var1: 1   var2: 2
now: 1725965976  var1: 0   var2: 1    <--------------
lnow: 1725965976  var1: 1   var2: 2
tnow: 1725965976  var1: 0   var2: 1   <--------------
(uint32_t)now(): 1725965976  var1: 1   var2: 2
(time_t)now():   1725965976  var1: 0   var2: 1   <--------------

 

[MarkT]

Technically now() wasn't called inside printf - all the arguments are computed first, then the call is made. I suspect this is because time_t is a wider type than you expected, ie long long. Try %llu instead of %lu ??
Did you turn on compiler warnings and check them?

 

[mborgerson]

Technically now() wasn't called inside printf - all the arguments are computed first, then the call is made. I suspect this is because time_t is a wider type than you expected, ie long long. Try %llu instead of %lu ??
Did you turn on compiler warnings and check them?

I added a line to my test program to print the sizeof(time_t). It returned 8, so you are correct that the something in the current compiler or libraries is setting time_t to a long-long 64-bit variable type. All the versions of time.h, TimeLib.h, etc. I've looked at to show time_t as an unsigned long. or uint32_t. The only time I saw long-long for time_t was in reference to the Windows OS.

Could this be something that changed with the version 11 compiler?

 

[joepasquariello]

Could this be something that changed with the version 11 compiler?

Yes, though I'm not sure in which version the change occurred. If I understand correctly, it's a configuration choice in the tools, but perhaps the default configuration changed in 11.

 

[jmarsh]

Either newlib or the newer version of gcc (I forget which is responsible) uses 64-bits for time_t, because we're getting pretty close to the point in time where a signed integer count of the number of seconds passed since the beginning of 1970 will no longer fit in a 32-bit value.

 

[thebigg]

This is certainly not a recent change

https://forum.pjrc.com/index.php?th...a-1-updated-toolchain-trial.71048/post-312561

 

[MarkT]

Some time in 2038 I think for 2^31 seconds past 1970... With 64 bits you can have microsecond precision and last 250 thousand years or so...

 

[shawn]

The printf() functions should normally warn you when the parameter type doesn’t match, for example “%lu” and time_t, which should be “%llu”. However, the Teensyduino implementation of the Print interface (which contains a printf(), and is the one called by Serial.printf()) doesn’t enable this because the theory is there’s too many warnings. (I disagree with this decision.)

May I suggest using printf() instead of Serial.printf()? The standard library should have these warnings enabled. (Serial.printf() uses the definition from the Print class; the non-qualified printf() is the one from the standard library.)

Note: std::printf() (if you include stdio.h) will also work.

 

[shawn]

Note: std::printf() (if you include stdio.h) will also work.

I meant <cstdio>, not <stdio.h>.