D and AI / dub registry and AI

[Laeeth Isharc]

On Wednesday, 11 February 2026 at 09:10:25 UTC, Lars Johansson 
wrote:
> I’m learning D, and as an experiment I asked Claude.ai to 
> convert a C connector to SAP ERP system to D.

Very cool experiment.

Something which may be obvious or not is that many of the larger 
coding models can be quite resourceful.  So - without being told 
- they will happily use gdb to debug segfaults when creating 
bindings.  Ask it if it missed anything out, and it might use nm 
to inspect the libX.so and see if any exported dynamic library 
functions were not wrapped.  I haven't had much need to write 
networking code, but presumably it takes away much of the pain 
there too.

I think that quite quickly the availability of libraries for a 
programming language start to become much less relevant to 
decisions made about choices from a first-principles rather than 
copying others perspective.

If you think about choice of language as  an economic problem, 
LLMs mean that the cost of porting over any library you don't 
have in that language falls, but especially the chance that you 
are in a mess that you can't solve collapses.  Since many people 
in corporations today are not part-owners of the business but 
hired hands and that often leads to a maximin type 
decision-making process then in time this might have some effect.

Probably languages like Ocaml and Rust benefit the most from this 
change in cost, but it should help D at least a bit.

I think another change - not just driven by AI but everything 
that's happened over the past decade (especially relating to 
security) - is that the cost of having dependencies - especially 
dependencies that you do not control or really understand - 
rises.  This happens at a time when the cost of rolling your own 
has collapsed (especially if the original library has a 
comprehensive test suite).  At the margin, companies probably 
will move more towards internalizing dependencies more than 
before.

Inspired by Robert Schadek and his talk where he mentioned 
'stealing' JS libraries, I ported an  SVG charting library from 
typescript to D.  Pretty automated and at least to get basics 
working not too many bugs to fix.  (And it could compare the 
results, look at the differences and fix the code itself).

When I started my career there were no libraries to speak of.  If 
you were doing a reconciliation then you had to write quicksort 
yourself (unless your firm had a library already written).  
People would use books like Numerical Recipes in C.  In a way 
that's what libraries become more than code that's taken 
wholesale.  Obviously, people _ought_ to respect licences, and I 
hope they will.

I think enterprise vendors - especially in in finance - are going 
to start to have interesting conversations with clients.  The 
default is that you eat what you get, and it's a very slow 
process to incorporate new features or bug fixes - months and 
maybe years.  The cost structure has changed so much, that eg 
with DLLs I think people will start to decompile them and patch 
them - initially to fix bugs, and later to do much more.  Vendors 
won't like it, but ultimately they probably won't have much of a 
choice

On a different note, it should be pretty easy now to do what had 
been discussed some time back - to give dub projects a quality 
score and to adopt the good ones that have been semi-abandoned up 
to scratch.