[D-runtime] How not to write OS specific code

[Brad Roberts]

Not just easier but also safer.  On demand means that the new entries are much more likely to be exercised and thus
known to be correct.

On 2/1/2011 8:03 PM, Sean Kelly wrote:
> If the default is not to define it then the user will get a compile error the first time he tries to use it. I'd love to
> proactively develop the Posix headers, but they're huge. It's easier to build them out as demand dictates. 
> 
> Sent from my iPhone
> 
> On Feb 1, 2011, at 6:38 PM, Walter Bright <walter at digitalmars.com <mailto:walter at digitalmars.com>> wrote:
> 
>> The assert is to force the person porting the header to a new system to check to see what is appropriate for that
>> system. I believe os specific code ought to be proactively written, not defaulted, even if the default is to leave it
>> undefined.
>>
>> Making it proactive means the maintainer looking at it knows that yes, it is undefined for system X, rather than being
>> overlooked.
>>
>> And I agree that having a default be a particular value is just a land mine waiting to blow the next person's foot off.
>>
>>
>> Sean Kelly wrote:
>>> I don't know that the static assert is appropriate, since that stuff is optional, and the convention in the Posix
>>> headers is not to have an assert for undefined blocks, but it's no big deal to me either way. The 'else' clause was
>>> clearly wrong however.
>>>
>>> Sent from my iPhone
>>>
>>> On Feb 1, 2011, at 10:55 AM, Walter Bright <<mailto:walter at digitalmars.com>walter at digitalmars.com
>>> <mailto:walter at digitalmars.com>> wrote:
>>>
>>>> Found this today in src/core/sys/posix/time.d:
>>>>
>>>> version( linux )
>>>> {
>>>>     enum CLOCK_MONOTONIC        = 1;
>>>>     enum CLOCK_MONOTONIC_RAW    = 4; // non-standard
>>>>     enum CLOCK_MONOTONIC_COARSE = 6; // non-standard
>>>> }
>>>> else
>>>> {
>>>>     enum CLOCK_MONOTONIC        = 4;
>>>> }
>>>>
>>>> This is *WRONG WRONG WRONG*.
>>>>
>>>> 1. it's wrong for OSX, which does not even define CLOCK_MONOTONIC
>>>> 2. there's no guarantee that on "other" operating systems CLOCK_MONOTONIC will be defined, or if it is, that it will
>>>> be 4.
>>>> 3. you only find out there's a problem when something mysterious happens.
>>>> 4. cut & pasting declarations from one operating system to another without checking is akin to dropping an anvil on
>>>> the head of the hapless developer who can't figure out why there are mysterious failures on that other system. The
>>>> developer then has to go back and hand-check every last one of those declarations.
>>>>
>>>> This is *not* acceptable practice. Declarations for an OS should never be inserted or enabled until they are checked.
>>>>
>>>> The corrected version is:
>>>>
>>>> version( linux )
>>>> {
>>>>     enum CLOCK_MONOTONIC        = 1;
>>>>     enum CLOCK_MONOTONIC_RAW    = 4; // non-standard
>>>>     enum CLOCK_MONOTONIC_COARSE = 6; // non-standard
>>>> }
>>>> else version (FreeBSD)
>>>> {   // time.h
>>>>     enum CLOCK_MONOTONIC         = 4;
>>>>     enum CLOCK_MONOTONIC_PRECISE = 11;
>>>>     enum CLOCK_MONOTONIC_FAST    = 12;
>>>> }
>>>> else version (OSX)
>>>> {
>>>>     // No CLOCK_MONOTONIC defined
>>>> }
>>>> else
>>>> {
>>>>     static assert(0);
>>>> }
>>>>
>>>> Note that the assert will trip when a new OS is tried, thus pointing the developer at places that need to be checked.
>>>> _______________________________________________