sentinel-terminated pointers

[andrewrk]

This implements #265.

• implement null terminated pointers
• make string literals null terminated pointers to arrays instead of array values
• delete C string literals and multiline C string literals from the language
• go ahead and do make null terminated pointer syntax support any sentinel value, not just null #3731
• get behavior tests passing
• get std lib tests passing
• self-hosted / zig fmt support for the new syntax
• slicing sentinel terminated array or pointer with .. as end index should give a sentinel-terminated slice slicing sentinel-terminated array or pointer with .. as end index should give a sentinel-terminated slice #3766
• replace syntax for [N]null T, []null T, and [*]null T, with the syntax from make null terminated pointer syntax support any sentinel value, not just null #3731
• add implicit casting tests
  • [:x]T to []T
  • [*:x]T to [*]T
  • [N:x]T to [N]T
  • *[N:x]T to *[N]T
  • [:x]T to [*:x]T
  • [N:x]T to [:x]const T
  • [N:x]T to [*:x]const T
• add additional tests for make null terminated pointer syntax support any sentinel value, not just null #3731
• Take advantage of null terminated pointers in the standard library (grep for #265). (follow-up issue: Finish taking advantage of sentinel-terminated pointers in the standard library #3767)
• Get the rest of the tests passing
• Add documentation to the language reference
• Look for regressions in zig community members' actively maintained projects there are already recent regressions such as error unions + optionals generating invalid llvm #3692 which are unrelated to this branch, so I'll be doing this in master branch in the upcoming week or two. Hold on to your butts.
• Remove type coercion from array values which do not have static lifetimes to references remove type coercion from array values to references #3768
• when comptime len index, reading gives comptime value. writing asserts the value is the sentinel, and is no-op make null terminated pointer syntax support any sentinel value, not just null #3731 (comment)
• compile error for sentinel mismatch when type coercing

[daurnimator]

* make string literals null terminated pointers to arrays instead of array values

Does this mean that you can't do "hello".len any more? does that end up invoking a comptime strlen?

What does this mean for concatenating two strings?

const a = "foo";
const b = "bar";
const c = a ++ b; // is this null terminated? do I end up with a null in the middle?

MakeMemNoAccess = valgrind.ToolBase("MC".*),

Are we going to end up with .* everywhere? I'm not entirely clear what the .* implies here (it's a pointer dereference

---

idea: instead of [*]null u8, [*]sentinel(0) u8. It's not obvious that null means 0: no where else in zig do we call 0 "null". We could open things up to allow arbitrary values in the future

[Rocknest]

This implementation does not seem quite right. It gives impression of a hack. Not sure.

Maybe there should be a type comptime_string that can coerce to [*]u8, [*]null u8 and slices.

Also i like @daurnimator's idea: [*]sentinel(0) u8 or [*]zero_sentinel u8.

[andrewrk]

Does this mean that you can't do "hello".len any more? does that end up invoking a comptime strlen?

This code works in master branch and in this branch: "hello".len == 5. No there is no comptime strlen. String literals have both a comptime known length and are also null terminated. They safely cast to both []const u8 and [*]null const u8.

What does this mean for concatenating two strings?

In the expression a ++ b, if either of a or b is a null terminated array, then the result is a null terminated array. It works with single-item pointers to arrays as well. If either is a single-item pointer to an array then the result is a single-item pointer to an array.

One of the main benefits of having null termination in the type system is that operations such as ++, **, and len property can work correctly.

Are we going to end up with .* everywhere? I'm not entirely clear what the .* implies here (it's a pointer dereference

typeOf("MC") == *const [2]null u8. So, dereferencing it gives you an array value of type [2]null u8. The question of whether to allow a type coercion from *[N]T to [N]T is a separate issue.

idea: instead of [*]null u8, [*]sentinel(0) u8. It's not obvious that null means 0: no where else in zig do we call 0 "null". We could open things up to allow arbitrary values in the future

Maybe you are onto something here, but please open a separate proposal. This is an implementation of the accepted-for-over-two-years proposal #265.

[daurnimator]

Maybe you are onto something here, but please open a separate proposal. This is an implementation of the accepted-for-over-two-years proposal #265.

Created as #3731

However now I'm curious: for non-pointer non-integer elements, what does null mean? e.g. [*]null enum{Foo, Bar}

[daurnimator]

Don't we need a extra bit of the test matrix here adding null pointers?

[andrewrk]

However now I'm curious: for non-pointer non-integer elements, what does null mean? e.g. [*]null enum{Foo, Bar}

Not allowed, with the current implementation. error: type "E" has no null value, cannot be used in null terminated pointer.

[daurnimator]

A good update to make would be for std.mem.toSlice and std.mem.toSliceConst to return null terminated slices.

[daurnimator]

A good update to make would be for std.mem.toSlice and std.mem.toSliceConst to return null terminated slices.

Attempting this I got:

broken LLVM module found: Call parameter type does not match function signature!
  %18 = getelementptr inbounds { %"[]u8", i16 }, { %"[]u8", i16 }* %0, i32 0, i32 0, !dbg !126586
 %"[:0]u8"*  call fastcc void @mem.toSlice(%"[]u8"* sret %18, i8* %16), !dbg !126586
Call parameter type does not match function signature!
  %name = alloca %"[:0]u8", align 8
 %"[]u8"*  %48 = call fastcc i1 @mem.eql(%"[:0]u8"* %name, %"[]u8"* @5875), !dbg !211774
Call parameter type does not match function signature!
  %name = alloca %"[:0]u8", align 8
 %"[]u8"*  %84 = call fastcc i1 @mem.eql(%"[:0]u8"* %name, %"[]u8"* @5874), !dbg !211810
Call parameter type does not match function signature!
  %hostname = alloca %"[:0]u8", align 8
 %"[]u8"*  call fastcc void @mem.copy(%"[]u8"* %9, %"[:0]u8"* %hostname), !dbg !251182
Call parameter type does not match function signature!
  %27 = getelementptr inbounds %"?[]const u8", %"?[]const u8"* %0, i32 0, i32 0, !dbg !284345
 %"[:0]u8"*  call fastcc void @mem.toSlice(%"[]u8"* sret %27, i8* %25), !dbg !284345