[vlc-devel] [PATCH 04/15] vlc_common: add helper function for explicit nanosecond to/from mtime_t conversion
remi at remlab.net
Mon Jun 18 11:57:28 CEST 2018
There are no such thing as an "other way". If you hide the unit of the time type, then you have to expose the unit of the conversions to and from it.
Milli is not a unit. It's a fixed point representation. If you use tick as the unit, then milli means millitick, not millisecond. Talk about confusing.
On top of that, ms/us/ns are shorter than milli/micro/nano.
Le 18 juin 2018 09:58:14 GMT+03:00, Steve Lhomme <robux4 at ycbcr.xyz> a écrit :
>On 2018-06-16 10:53 AM, Rémi Denis-Courmont wrote:
>> Le lauantaina 16. kesäkuuta 2018, 11.32.50 EEST Steve Lhomme a écrit
>>> On 2018-06-16 10:23 AM, Rémi Denis-Courmont wrote:
>>>> Le lauantaina 16. kesäkuuta 2018, 10.34.18 EEST Steve Lhomme a
>>>>> I used milli instead of ms so it's not confused with msftime_t.
>>>>> micro and nano seems like logical siblings.
>>>> Then keep mtime and screw this series.
>>> I find mtime misleading. Especially because it assumes a certain
>>> and in many parts of the code 1000 or 1000000 is used instead of
>>> conversion because what could possibly go wrong ?
>> Well, these things regularly go wrong:
>> - failing to convert, typically the caching values between ms and µs,
>> - failing to expand before computation, typically from 32-bits to
>> mtime_t is not a particularly good name, but it's anyway assumed same
>> int64_t all over the place, so the name does not matter than much.
>>> With a tick (or something else) you know you have to take care of a
>>> conversion in the proper time format you're handling.
>> But that's exactly the point. If you use tick, jiffy or whatever
>> standard unit, then you have to define what the other unit is.
>I see it the other way. One should not know what a tick is nor the
>precision it can have. The API around it should hide all the
>intricacies. The fact the name implies it's in microseconds lead to a
>lot of dirty code we have right now.
>> That's why ns, us/µs, ms are far better names than nano, micro and
>I would like nsec, msec and µsec but it's not possible and using ms may
>lead to confusion. And us doesn't sound anything like µs it just looks
>>> mtime_t will stay for backward compatibility but IMO it should not
>>> used anymore.
>> I don't mind changing the name to something short and correctly
>> But it would be utterly vain in any case. The code assumes int64_t in
>> variety of subtle and not-so-subtle ways, no matter the typedef
>This is all fixed. See
>>>> As far as I am concerned, if you don't specify the time unit in a
>>>> program, you are using local jiffies, not seconds. Talk about
>>>> ISO C uses nsec, not nano. SI uses ms, µs and ns with us as the
>>>> standard transliteration of µs.
>>> We could use nsec instead of nano to align with the standard. But
>>> msec could be either milliseconds or microseconds.
>> Not really. VLC did confuse m for micro. But in the standards, it's
>> for micro and m for milli.
>I don't like it. But if everyone agrees on this fine. I can change that
>easily in my patches.
>vlc-devel mailing list
>To unsubscribe or modify your subscription options:
Envoyé de mon appareil Android avec Courriel K-9 Mail. Veuillez excuser ma brièveté.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the vlc-devel