D Language Foundation July 2025 Quarterly Meeting Summary

[Mike Parker]

The D Language Foundation’s quarterly meeting for July 2025 took 
place on Friday the 4th at 15:00 UTC. It lasted a little under 30 
minutes.

Our quarterly meetings are where representatives from businesses 
big and small can bring us their most pressing D issues, status 
reports on their use of D, and so on.

## The Attendees

The following people attended the meeting:

* Walter Bright (DLF)
* Dennis Korpel (DLF/SARC)
* Mario Kröplin (Funkwerk)
* Razvan Nitu (DLF)
* Mike Parker (DLF)
* Carsten Rasmussen (Decard)
* Bastiaan Veelo (SARC)

## The Summary

### Carsten

Carsten said things were going well at Decard. It was very 
exciting on the business side because they were going into 
logistics and security. The problems they were having to solve 
had nothing to do with the language, just normal development 
stuff. They were comfortable with what they had in the system and 
were happy.

### Mario

Mario said he’d emailed Mathis Beer about the meeting three days 
ago, but suspected he'd forgotten about it. Mathis was 
programming in D, but Mario was currently the Java programmer, so 
he had nothing to bring to us this time.

### Walter

Walter said he’d previously complained about the lack of D 
articles on Hacker News, but one had shown up recently and turned 
out to be a significant success for us. He wanted more of that 
kind of visibility.

He then went into a little rant about people who said "I used D 
for years, but it had this one flaw, so I dropped it," and then 
presented that flaw as the reason D couldn't succeed. The example 
he’d seen lately was, "I can’t use D because the exception 
handler uses the garbage collector."

Carsten said they shouldn't use exceptions if that was a concern. 
Walter said using exceptions with the GC wasn’t going to break 
your program. He got frustrated with people turning a single 
technical detail into a deal-breaker.

I brought up a comment I’d seen on Dennis’s DConf talk from last 
year: "The single biggest mistake Bright made when designing D 
was the GC. That is the sole reason D hasn't replaced C++ by now. 
It's the biggest crutch to D adoption. Really, who wants to deal 
with this kind of nonsense?"

Carsten said, "No, no, no, no, no, no, no."

Walter said if it wasn't the GC, it would be something else. 
Throughout his career, he'd had people telling him, "If only your 
compiler had this feature, our company would switch." He'd add 
that feature and they'd say, "That's nice, but what we really 
need is this other feature." Then you ended up on a 
merry-go-round. They had no intention of ever using it.

Carsten added that the GC was one of the reasons he had switched 
from C++ to D. He'd gotten tired of hunting wild pointers and was 
too old to do that anymore. Walter said that the GC had ended up 
being a surprising asset for CTFE.

I asked Walter if that was all. He said he was done ranting for 
the moment, but was sure he'd think of something else to rant 
about.

### Bastiaan

Bastiaan had two work issues he wanted to bring up. He'd spoken a 
little bit about them with Dennis.

__alloca and 64-bit DMD__

Bastiaan reminded us he’d started using `alloca` as a solution to 
one of the thread contention problems he'd had. It worked fine 
with DMD in 32-bit and fine with LDC in 64-bit, but when 
compiling for 64-bit with DMD, he got an error saying `alloca` 
didn't work with exception handling. He was wondering why and 
whether it was really working on 32-bit and 64-bit LDC.

Walter recalled giving up on trying to get `alloca` working on 
Windows 64-bit. It was ugly to implement, though he couldn't 
remember what the actual problem had been. He’d never been able 
to figure out Win64 exception handling. DMD’s exception handling 
on Windows differed from MSVC’s. He suggested that Bastiaan's 
problem was probably related to that.

Dennis posted [a link to the relevant code in DMD's 
`eh.d`](https://github.com/dlang/dmd/blob/9f573c494acc38855027462bde162fabea9cf33f/compiler/src/dmd/backend/eh.d#L50) along with a comment that he thought Walter had probably written:

```
// BUG: alloca() changes the stack size, which is not reflected
// in the fixed eh tables.
```

Walter said we were kind of stuck with that. He suggested using a 
RAII malloc/free scheme, which would accomplish the same thing as 
`alloca`. Bastiaan pointed out that `malloc` required the global 
lock. One of the reasons they were using `alloca` was to avoid 
that.

As an alternative, Walter suggested that if they knew the maximum 
size of the allocation, they could use a struct of that size. 
That would then be the allocated storage. Bastiaan didn't think 
they could get rid of `alloca`. It wasn’t a deal-breaker because 
they could just use LDC, but it was a minor inconvenience that 
they couldn’t fall back on DMD.

Bastiaan recalled an initiative for nonallocating exceptions and 
wondered what the story was. Walter said there was a compiler 
switch that enabled them and it did work, but there were some 
issues with it he never fully understood. I said Razvan could 
probably explain it, and that I recalled it had to do with stack 
trace generation. Walter said that sounded like that was probably 
the issue.

__Link errors in LDC__

Bastiaan's second issue was an LDC linking problem after 
upgrading the front end. He didn’t expect a full answer on the 
call and would need to work on a reduction. He asked if Dennis 
had any thoughts on it.

Dennis thought it was a mix of compiler flags and preview 
switches causing a difference in symbol generation. He hadn't yet 
reproduced it.

Walter suggested it could also be a DRuntime/Phobos mismatch and 
stressed that DRuntime, Phobos, and user code all needed to be 
built with the same compiler version. Bastiaan believed they were 
doing that. He said the error mentioned "conflicting weak 
external definition". He'd only seen it a few times and had 
forgotten the context.

__Floating point issue__

Bastiaan said that aside from those issues, things were generally 
fine. The only other issue he could think of was that 64-bit 
floating point answers sometimes deviated from 32-bit. One reason 
was that he'd been relying on undefined behavior for some code 
using `bitmanip`. He was glad they'd disovered that. It was in 
order now.

The other was because 32-bit DMD used 80-bit floating point in 
intermediate results. This was code using floats, not doubles. 
LDC, even in 64-bit, was using 32-bit intermediates. He figured 
that was introducing a lot of rounding errors in the intermediate 
results.

Walter explained that if the target machine supported XMM 
floating point registers, the compiler would use them, otherwise 
it would use the 80-bit floating point coprocessor, which did 
everything in 80 bits. He said 32-bit LDC on the platform 
Bastiaan was working on wasn't using the XMM registers. He 
suggested looking at the LDC switches to see if there was one to 
enable it. That should clear up the problem. Bastiaan said that 
was a good point and he would look into it.

### Dennis and Razvan

After Bastiaan finished, I asked Dennis if he had anything 
work-related to add beyond what had come up in Bastiaan’s issues. 
Dennis did not. Neither he nor Razvan had anything DLF-related.

### Me

I said I would be be announcing the Symmetry Autumn of Code in 
the next few days. We had funding from Symmetry for three slots. 
I noted Emmanuel Nyarko had already told me he planned to apply. 
I asked all present to let me know if they had any open source 
libraries they were using that needed some work. That sort of 
thing could make potential project ideas.

I asked Carsten if he could send me an updated headcount of 
Decard's DConf attendees within the next week. We needed to send 
the caterers an approximation of our final numbers. Then I asked 
if anyone had anything else to add. Dennis asked if there was 
going to be a monthly meeting the following week. There was.

### AI interlude

As we were about to close, Walter told us he’d tried very hard to 
get an AI to draw a D-man. He'd failed miserably no matter how he 
crafted the prompt. He kept getting something that looked like 
Superman with a D on his chest.

Carsten offered to ask someone at his company who was "super good 
at that" to see if they could come up with a prompt that worked. 
I mentioned that some of the services let you upload a reference 
image. Walter said he’d tried that, too. He'd tried everything. 
All these people were saying AI was going to take over 
everything, and he hadn't been successful with getting it to do 
what he wanted.

I said it had been very useful for me with research and 
translation. Carsten agreed. He thought it was great for 
extracting information and translation. When it came to writing 
code you wanted to maintain, he thought it was just crap on crap, 
but maybe he was too old for that.

I noted that I'd gotten GPT to generate a Lua script as a plugin 
for Pandoc, because I knew nothing about Pandoc plugins. It had 
mistakes that I'd had to correct, but it got me further than I 
would have gotten on my own in a shorter amount of time. Carsten 
thought it was great for that sort of thing, scripts for specific 
tasks. But for something that you wanted to develop and maintain, 
he didn't think we'd be losing our jobs any time soon.

Walter said he’d been invited to give a short presentation about 
D that highlighted concurrent programming. He didn't know much of 
anything about concurrency, so he'd asked Grok for a sample 
program and it worked. Then he thought he should send the prompt 
to Bruce, who was an expert on concurrent programming. Bruce sent 
back a marvelously written piece of concurrent code that did the 
same thing. Walter scrapped the one from the AI and went with 
Bruce's. There was nothing like talking to an expert to get 
something we could be proud of. That had been fun.

Mario joked that it was good Mathis wasn’t there, otherwise the 
meeting would run for another hour. Mathis did nothing without 
AI. He put the code in a loop with the compiler and the AI was 
fixing his bugs. He was very deep into AI. Carsten said you could 
probably use it as a tool if you knew what you were doing. Mario 
agreed and mentioned a saying that it made good people better and 
bad people worse.

Walter said when he asked AI to write a piece of code, he knew 
how to specify what he wanted. It was really helpful in getting 
the results he wanted on that front. He was just annoyed that he 
couldn't get it to generate a D-man.

He said we had a shortage of D-man cartoons. He had a directory 
with a collection of all the ones he'd seen. For him to draw one 
was a long, complicated process because he had no talent for it. 
He would have to spend a lot of time with the image editor to 
make it look presentable. Being able to automate it would be 
wonderful. Something silly like D-man jumping rope or something. 
If anyone could figure out how to do it, he asked that they send 
him the prompt.

### Conclusion

Our July monthly meeting took place the following Friday, July 
11th. Our next quarterly happened on Friday, October 3rd, 2025.

If you are running or working for a business using D, large or 
small, and would like to join our quarterly meetings periodically 
or regularly to share your problems or experiences, please let me 
know.