writeout: add %time{}

[bagder]

Output the current UTC time using strftime format. %f is an extra curl specific flag to output the microsecond fraction of the current second.

TODO

• decide if we want this at all
• is the strftime good enough?
• add test case to verify functionality
• %z works in a portable fashion
• %Z works in a portable fashion
• check curl -h -w with this PR as it shows a formatting error not seen in the man page => managen: reset text mode at end of table marker #18139

[dfandrich]

FYIW, Python uses %f as the microseconds specifier, so lots of people will be used to that.

[bagder]

Some of our compiler warnings sometimes are quite counter productive:

../../src/tool_writeout.c:575:23: error: conversion from 'long long int' to 'long int' may change value [-Werror=conversion]
  575 |         cnow.tv_sec = (time_t)val;
      |                       ^

On the line above 575, we make sure that the value stored in val cannot be larger than what fits in a time_t...

[bagder]

Is there a better solution than to switch off those warnings (for this code)?

[icing]

Is there a better solution than to switch off those warnings (for this code)?

Disable debug builds? 😬

have you tried (val < TIME_T_MAX) ? (time_t)val : TIME_T_MAX;?

[bagder]

Grr, mingw:

struct timeval
{
	long tv_sec;
	long tv_usec;
};

[icing]

So, if I understand the approach correctly here, %time{fmt} will output the current time at the moment the writeout happens. Which usually is the end of the transfer. Usually, except when there is more than one transfer and other do some time consuming stuff (like a FTP blocking data connect).

Do I read this right?

Before looking at the details here, I was expecting to have a %start-time or so, that gives the timestamp of a transfer related event (like the start, could also be the end, whatever). And that timestamp to be in a fixed format, like rfc3339.

Can you explain what the advantages of the current approach and the repetition of strftime()? It's very flexible, but is this a good thing, I wonder?

[bagder]

So, if I understand the approach correctly here, %time{fmt} will output the current time at the moment the writeout happens. Which usually is the end of the transfer. Usually, except when there is more than one transfer and other do some time consuming stuff (like a FTP blocking data connect).

It is still after the transfer was just completed. yes.

Before looking at the details here, I was expecting to have a %start-time or so, that gives the timestamp of a transfer related event (like the start, could also be the end, whatever). And that timestamp to be in a fixed format, like rfc3339.

I think it is fair to consider a start-time as well, sure. But as my primary use-case here is for logging the transfer, I believe the end time is more useful .

Can you explain what the advantages of the current approach and the repetition of strftime()? It's very flexible, but is this a good thing, I wonder?

My primary use case is when this is used for logging (I know a real user wanting this). Basically doing curl $URL -w $FORMAT >> logfile - and since there are many different logfile formats and people want different things for their timestamps, I figured it was easy enough to offer that flexibility rather than picking a single format that is bound to end up the wrong one for some users.