Re: [TCLCORE] TIP #423: Formatting Timestamps with Milliseconds

[Thomas P. <tom...@li...>]

> Specific concerns:
> 
> (1) Under normal circumstances you'll be taking an integer time from the 
> system clock (via [clock seconds]) or from a database, then converting 
> to floating point, then running it through [clock format].  The reverse 
> route is to scan a date/time string using [clock scan] to get a floating 
> point value, then converting that value to an int for storage or precise 
> manipulation.  I don't see any need to have the floating point step in 
> there.  It's the wrong approach.

? I don not get your "normal circumstances". My use-case 1): Get the system clock via [clock milliseconds] or from a database, and then make a graph/plot or display the values on the GUI.use-case 2): Get the system clock via [clock milliseconds] and store it to database, and then make a graph/plot or display the values on the GUI.
> 
> (2) As Keven points out, the format specifier suggests an equivalent 
> scan specifier:
> 
> On 2013/07/18 04:00 AM, Kevin Kenny wrote:
> > If we go this route, we should have the same format group recognized
> > by [clock scan], and cause the result to be expressed as a float.
> 
> So, based on the value of the -format parameter, [clock scan] may return 
> an int or it may return a float.  Bad!

Ok - understood.But consider  [expr 10/3.]: There will be a floating point result only if one of these number have a dot. Similar this could be handled in [clock] operations

> Imagine that I have an application that allows the user to specify a 
> date/time format string in a config file, and parses inputs using [clock 
> scan] and the configured format string.  Upgrade to a TIP#423-supporting 
> interp and you'll blow up my app by giving me a float when I expect an int.

You need to check the input from the user anyhow. Or what do you do when he configures unknown format specifiers? 
> 
> 
> Counter-proposal:
> 
> Add to [clock format], [clock scan] and [clock add] an additional 
> optional parameter that identifies the scale/magnitude/resolution of the 
> time (formerly 'seconds') argument, one of:
>    -millis true|false
>    -micros true|false
>    -scale 1|1000|1000000 (default 1)
>    -resolution seconds|millis|micros (default seconds)
> ... or something similar.
Do you mean that %S will output a float depending on such a switch?
ThxThomas