Compiler Bug using D interface file for its intended purpose

[Richard (Rikki) Andrew Cattermole]

On 09/09/2026 5:14 AM, Rambutan wrote:
> On Tuesday, 8 September 2026 at 07:34:09 UTC, Richard (Rikki) Andrew 
> Cattermole wrote:
>> On 08/09/2026 6:31 PM, Rambutan wrote:
>>> This is apparently a bug, as D interface files containing only simple 
>>> type definitions to replace C header files in D source like di.di are 
>>> supposed to behave like header files.
>>
>> No they are not.
>>
>> D generates metadata that must be compiled in.
>>
>> .di files can only exist as derivatives of D files that have been 
>> linked in.
> 
> Ah, I see! Does this mean that officially .di files are not intended as 
> a mechanism to provide C header file information to D, and so such a use 
> is a technical exercise pushing the envelope?

Yes, basically.

Hence why they are generated by the compiler with the help of a compiler 
switch. Your not meant to make them yourself, although you can, after 
all they are regular d files with a different file extension.

> This use of .di files with C is the only one I have encountered in 
> practice, and they have been generated automatically from C header files 
> and edited to smooth out any wrinkles. Until this encounter with doubles 
> in structs this has all worked fine with no compilation of the .di file.

https://github.com/dlang/dmd/blob/master/druntime/src/core/stdc/stdio.d

Things work, right up until it doesn't work.