druntime

[Jacob Carlborg]

On 2011-07-09 01:29, Iain Buclaw wrote:
> == Quote from Brad Roberts (braddr at slice-2.puremagic.com)'s article
>> On Fri, 8 Jul 2011, Iain Buclaw wrote:
>>> == Quote from Jonathan M Davis (jmdavisProg at gmx.com)'s article
>>>> On 2011-07-08 10:42, bioinfornatics wrote:
>>>>> @sean
>>>>> if you install ldc2 like:
>>>>> $ cmake . -DD_VERSION:STRING=2 -DCONF_INST_DIR:PATH=/etc
>>>>> $ make -j4 VERBOSE=2
>>>>> $ make -j4 install
>>>>>
>>>>> and try install druntime from
>>>>> https://github.com/D-Programming-Language/druntime.git I can't because
>>>>> make file is only for dmd. What i try to said, yes we need 1 druntime so
>>>>> for this reason druntime installer need support at least dmd, ldc, gdc.
>>>>> But is not case currently. And for this reason d2 can't go to Fedora 16.
>>>>> Because ldc2 use cmake for build 3 projects (ldc, druntime, phobos)
>>>>> I need 3 installer separately. And ldc2 use a druntime fork!
>>>>>
>>>>> thanks for any answer :-)
>>>> I believe that it's _expected_ that other compilers will use forks of
>>>> druntime. They may have to make changes to druntime to work, and Sean doesn't
>>>> want to have to maintain all of the differences for every compiler. Rather,
>>>> druntime is the reference implementation intended for dmd, and other compiler
>>>> maintainers do whatever they need to with their own version of it to get it
>>>> work with their compiler.
>>>> - Jonathan M Davis
>>>
>>> Exactly this, and the case is also vice versa with gdc. The druntime reference
>>> library also does many things that are unreasonable and incompatible with gdc (and
>>> I assume ditto ldc too).
>>>
>>> One future plan on my list is the restructuring of core/stdc to be more ports
>>> friendly (the source, not the installed files) - something to help push along ARM
>>> development for D2 with GDC, and hopefully for other archs to follow pursuit. The
>>> result being one elongated patch that won't be accepted upstream for sure. :~)
>>>
>>> Regards
>>> Iain
>> This is one area that I disagree with Sean on.  I think it's worth merging
>> in as much as we can to the druntime code base.  I'm not against having
>> separate trees vended by each compiler, but I hope/expect those to be the
>> lowest level details and not be detectably different from phobos or user
>> code.  Having core.stdc diverge, as just one example, is a recipe for
>> having code that only works on top of one specific runtime, which is NOT
>> what we want.
>
> Hardly - as core.stdc (forgot core.sys too :) is just a wrapper for the standard C
> library.
>
> Sweet story short, having version() else version() else version() for 6+ platforms
> and 12+ architectures is not workable in a single managed file - consider, for
> example, sys.stat with 15 variants of struct stat_t inside.
>
> Regards
> Iain
>

How about having a separate branch for each compiler in druntime? The 
compiler maintainers will maintain their own branch but if there's a 
change that doesn't actually affect any compiler specific code, i.e. 
adding a new function to core.stdc, then that change can be made to all 
branches.

-- 
/Jacob Carlborg