On Thu, 23 Jun 2011 18:42:45 -0400, Jonathan M Davis <jmdavisp...@gmx.com>
wrote:
On 2011-06-23 15:10, Steven Schveighoffer wrote:
On Thu, 23 Jun 2011 18:02:02 -0400, Jonathan M Davis
<jmdavisp...@gmx.com>
wrote:
> On 2011-06-23 14:50, Andrej Mitrovic wrote:
>> On 6/23/11, Steven Schveighoffer <schvei...@yahoo.com> wrote:
>> > On Thu, 23 Jun 2011 07:27:34 -0400, Jimmy Cao <jcao...@gmail.com>
>>
>> wrote:
>> >> Thread.sleep( 70_000_000 ); // 7 sec
>> >
>> > Gah!
>> >
>> > Here, let me fix that for you:
>> >
>> > Thread.sleep(dur!"seconds"(7));
>> >
>> > :)
>> >
>> > -Steve
>>
>> I find it very odd that for seconds we use "seconds", but for
>> everything else we use "msec", "usec", "hnsecs" abbreviations.
>
> Because it makes sense to use the full names up to seconds. After
that,
> they're too long, so they get abbreviated.
You could argue that mseconds is not much longer than seconds. It's
still
an abbreviation.
You could also argue that seconds is too long, couldn't it be secs? One
thing that sucks about having inconsistent abbreviations, I have to look
them up, even when I know what I'm typing makes sense to a reader, it
just
may not be what the function expects. For instance, before I sent this
message, I had to look it up to make sure it was valid.
Unfortunately, there are so many variations on what they could be that
there's
no way to make them so that everyone is going to remember them
correctly. So,
no matter what I picked, someone would have been unhapy with it.
Right. I totally agree, if you have to have one, you just have to choose
and be done with it. The only reason we can have this conversation is
because it's really easy to just add another "overload". :)
But here's an idea -- milliseconds, msecs, and mseconds could all map to
the same template function.
hectonanoseconds, hm.. that's a long one :)
I thought about doing that (though it _never_ occurred to me to use
mseconds,
since that feels to me like you abbreviated half of it, which is just
weird to
me). The problem is that that feels a bit too much like the inconsistent
horror that is html where it accepts just about anything. I generally
prefer
things to be strict and consistent over trying to guess what was meant.
It
also makes the template constraints uglier. I'm not completely against
the
idea, but I'm not convinced that it's a good one either.
I'm not completely for it either, it was just a thought. Personally, I
think super-abbreviated and fully-spelled-out would be the two best
options to have. Since you are already sticking it in a namespace (dur),
it should be rather obvious what each is:
hns or hectonanoseconds
us or microseconds
ms or milliseconds
s or seconds
m or minutes
h or hours
d or days
w or weeks
mo or months (yeah, this one is sucky)
y or years
As it stands, all units seconds and greater are completely
unabbreviated, and
all of the sub-second units are abbreviated, which if you're going to
abbreviate some but not all feels very consistent to me (though again,
what
seems perfectly logical to one person may not seem that way to another).
It would be good to highlight this pattern in the documentation. It makes
it easier to remember which ones are abbreviated!
I
_don't_ want to change them to something else at this point. There's
enough in
core.time and std.datetime to get used to already, and that's a pretty
heavily-used feature in those modules. It _might_ be worth making it so
that
other variants of the units' names work, but I'm not convinced that
being lax
about it is necessarily a good idea.
To me, an optimal API is one you don't have to look up. It just makes
sense. There are various ways to achieve this, and it's probably
achievable with your scheme as long as you give an easy rule to remember
(i.e. subseconds are abbreviated, everything else is not).
One thing that would be good to have is to have a more defined error
message when you type the wrong term. i.e.:
Error: millis is not a valid unit, try hnsecs, usecs, msecs, ...
-Steve
