D Language Foundation October 2025 Quarterly Meeting Summary

[Richard (Rikki) Andrew Cattermole]

On 14/03/2026 1:52 AM, Richard (Rikki) Andrew Cattermole wrote:
> On 14/03/2026 1:43 AM, Mike Parker wrote:
>> Luís then brought up a second issue. Weka wanted to move away from 
>> statically linking everything and start using dynamic linkage instead, 
>> but they had run into a lot of problems there as well. He expected 
>> that once they got further into that transition, they would start 
>> filing bug reports. One example he had already seen was that export 
>> did not work properly together with |-fvisibility=hidden|. He thought 
>> this was probably LDC-specific, though he wasn't sure whether there 
>> was an equivalent DMD-side mechanism for the same thing. He had 
>> noticed cases where compiler-generated symbols, such as postblits, 
>> weren't being marked for export, which meant they weren't exported 
>> from the shared library.
>>
>> I asked whether he already had a bug report for that second issue. He 
>> said yes, there were two bug reports related to export not working 
>> properly, but he didn't have them handy. I told him to send them to me 
>> and I would forward them on to Dennis and Nick. He added that Ilya 
>> might also have some other issues to raise, as he was working on a new 
>> build system. Weka were integrating Bazel and had run into some issues 
>> there, though Luís was not sure of the details.
> 
> 
> https://github.com/dlang/DIPs/blob/master/DIPs/other/DIP1045.md
> 
> The only thing that was approved was -extI which is now implemented.
> 
> This was one of the things featured in the DIP.
 > The main issue for them was on the LDC side. Some of Weka's binaries, 
especially their unit test binaries, were so large that they exceeded 
four gigabytes. At that point, the linker started complaining about 
overflowing relocation address space. As he understood it, the 
relocation addresses were signed 32-bit values, so once the binary got 
past that size, the relocations overflowed.

For anyone who may be hitting this, try setting the code model to medium.

For ldc you can do this with: ``--code-model=medium``
That'll change it to 64bit relative addresses and may solve it.