• Nick Treleaven (13/13) Mar 25 I've written up a formal DIP which needs discussion in this forum
• Walter Bright (9/9) Apr 03 Thank you, Nick!
• Quirin Schroll (16/38) Apr 10 and
• Nick Treleaven (9/32) May 06 OK. I think Dennis allowed `[]` for a PR that didn't require an

Nick Treleaven <nick geany.org> writes:

I've written up a formal DIP which needs discussion in this forum 
first:
https://github.com/ntrel/DIPs/blob/6678d4ef9d233dd5780b80900debdf7066165655/DIPs/1NNN-NMT.md#missing-elements-in-static-array-initializers

Initially I also included a proposal for overriding `E.init` for 
missing elements (with syntax `[elements, ... = init]`) as I saw 
Walter propose syntax for that in the September meeting write-up:
https://forum.dlang.org/post/ucuhbblifjcjkfvikbqm forum.dlang.org

However I think that part is less compelling so I've removed it 
from the DIP above. Part of the justification for it is to make 
it easier to port C code which requires zero initialization. 
However, the above DIP suggests that the compiler warns the user 
about this when `E.init` is nonzero. The user would then need to 
set the missing elements to zero at runtime.

Walter Bright <newshound2 digitalmars.com> writes:

Thank you, Nick!

The rationale for accepting too few elements and default initializing the rest 
comes from:

```d
int array[1000] = [1,2,3];
```

where it is simply impractical to write out 997 initializers. This pattern 
occurs in C.

I like the proposed `[1,2,3,...]` syntax a lot better than the "Alternatives"
ideas.

Quirin Schroll <qs.il.paperinik gmail.com> writes:

On Wednesday, 25 March 2026 at 15:45:29 UTC, Nick Treleaven wrote:
 I've written up a formal DIP which needs discussion in this 
 forum first:
 https://github.com/ntrel/DIPs/blob/6678d4ef9d233dd5780b80900debdf7066165655/DIPs/1NNN-NMT.md#missing-elements-in-static-array-initializers

 Initially I also included a proposal for overriding `E.init` 
 for missing elements (with syntax `[elements, ... = init]`) as 
 I saw Walter propose syntax for that in the September meeting 
 write-up:
 https://forum.dlang.org/post/ucuhbblifjcjkfvikbqm forum.dlang.org

 However I think that part is less compelling so I've removed it 
 from the DIP above. Part of the justification for it is to make 
 it easier to port C code which requires zero initialization. 
 However, the above DIP suggests that the compiler warns the 
 user about this when `E.init` is nonzero. The user would then 
 need to set the missing elements to zero at runtime.

I disagree with two restrictions:

 Static initialization from an array initializer with missing 
 elements will be deprecated in the next edition, except when:
 * […]
 * […]
 * the array initializer is empty `[]`

and

 It is an error to use [elements, ...] initializer syntax when:
 * […]
 * there are no missing elements

The first one adds a special rule which I don’t think is 
justified. There’s no big drawback to writing `[...]`. Only for 
the current Edition, I’d not make this requirement because it’s 
needlessly breaking. Moving forward, `[]` should only initialize 
empty arrays, and `[...]` initializes arrays of any size, filling 
them with `init`.

The second one means `...` must produce one or more elements, 
when it’s rather obvious it can also produce none. It should be 
allowed to do so, otherwise meta-programming might require 
special handling to omit `...` when the number of elements and 
the desired size happen to line up. By far the easiest fix is to 
just allow `...` even if it inserts no elements because enough of 
them are in the initializer.

Nick Treleaven <nick geany.org> writes:

On Friday, 10 April 2026 at 14:56:15 UTC, Quirin Schroll wrote:
 I disagree with two restrictions:

 Static initialization from an array initializer with missing 
 elements will be deprecated in the next edition, except when:
 * […]
 * […]
 * the array initializer is empty `[]`

 and

 It is an error to use [elements, ...] initializer syntax when:
 * […]
 * there are no missing elements

 The first one adds a special rule which I don’t think is 
 justified. There’s no big drawback to writing `[...]`. Only for 
 the current Edition, I’d not make this requirement because it’s 
 needlessly breaking. Moving forward, `[]` should only 
 initialize empty arrays, and `[...]` initializes arrays of any 
 size, filling them with `init`.

OK. I think Dennis allowed `[]` for a PR that didn't require an 
edition, and the test suite uses that. We can start off strict, 
and see if we get complaints, or if buildkite projects break when 
the next edition becomes the default. I'm planning on making 
dscanner auto-fix code when its parser can tell there's a static 
array declaration with missing elements.

 The second one means `...` must produce one or more elements, 
 when it’s rather obvious it can also produce none. It should be 
 allowed to do so, otherwise meta-programming might require 
 special handling to omit `...` when the number of elements and 
 the desired size happen to line up. By far the easiest fix is 
 to just allow `...` even if it inserts no elements because 
 enough of them are in the initializer.

OK, accepted.

Thanks for your review, sorry for the delay!