self-host P2 slice 4: strings, byte-identical to gen-c; slices refused #8014

Part of #5980 (phase P2: "strings/slices"). This is the fourth P2 slice; it follows #8000, #8007 and #8012.

Probed with t27c gen-c on master df0c185 (t27c lab):

• str and string become const char*;
• a string literal is written back with zig_escape;
• .len on a string parameter or local becomes strlen (with <string.h>);
• a string const becomes a #define.

Every lossy shape is listed in #8013. Today the core refuses every string with E_LEX and every slice type with E_TYPE.

This slice

1. specs/compiler/core/t27core.t27 accepts, byte-identical with gen-c:

   • string literals: printable ASCII, the escapes \n \t \\ \", and no ??;
   • the types str and string for parameters, returns, locals, module vars and consts;
   • a string value only where a string is declared: an initializer, an argument, a return value, an assignment, or a switch arm in one of those;
   • .len on a string parameter or local, with gen-c's <string.h>;
   • indexing a string.

2. New code E_STRING (15). The core refuses with it every lossy shape of gen-c: strings and slices lowered to bare pointers -- == compares addresses, .len on str and on slices is not C, escapes and trigraphs change value, a string where an integer goes compiles #8013:

   • a string operand of an operator;
   • a string where the declared type is not a string, or a non-string where it is;
   • .len on a str or on a module name;
   • an escape other than \n \t \\ \", and ??;
   • a repeated string.

   Slice types []T stay refused (E_TYPE, gen-c: strings and slices lowered to bare pointers -- == compares addresses, .len on str and on slices is not C, escapes and trigraphs change value, a string where an integer goes compiles #8013). So do struct fields and array elements of string type, because the core cannot follow the type of such an element without the P3 type checker.

3. A string-typed name is bound once in its function and module, so a name's string type is never ambiguous.

4. Fixtures: an accept file that runs, and one refusal per shape. The return "s" row of REFUSALS in core_selfhost.rs moves to a data fixture with its new code.

5. t27core_ddc.t27 is repinned.

Gate: cargo test -p t27c --test core_selfhost on the t27c lab, with the fixpoint, the core tests, every fixture and the corpus differential. The DDC also passes under Zig and t27b.

User Scenarios

• Given a valid string literal with printable ASCII and escapes \n \t \\ \", when compiled with t27c, then the core accepts it and emits byte-identical C to gen-c.
• Given a string operand used with an operator, when compiled with t27c, then the core refuses it with error code E_STRING (15).
• Given a .len access on a str type or module name, when compiled with t27c, then the core refuses it with E_STRING.

Requirements

• FR-001: The core MUST accept string literals containing printable ASCII and the escapes \n, \t, \\, \", and no ??.
• FR-002: The core MUST accept types str and string for parameters, returns, locals, module vars, and consts.
• FR-003: The core MUST accept .len on a string parameter or local and emit <string.h> with strlen.
• FR-004: The core MUST accept indexing a string.
• FR-005: The core MUST refuse every lossy shape from gen-c: strings and slices lowered to bare pointers -- == compares addresses, .len on str and on slices is not C, escapes and trigraphs change value, a string where an integer goes compiles #8013 with error code E_STRING (15).
• FR-006: Slice types []T MUST remain refused with E_TYPE.
• FR-007: A string-typed name MUST be bound once in its function and module so its type is never ambiguous.
• FR-008: Fixtures MUST include one accept file that runs and one refusal per lossy shape; the return "s" row from REFUSALS in core_selfhost.rs MUST move to a data fixture.
• FR-009: t27core_ddc.t27 MUST be repinned.

Success Criteria

• cargo test -p t27c --test core_selfhost exits 0 on the t27c lab (fixpoint, core tests, every fixture, corpus differential).
• The DDC passes under Zig and t27b (exit 0).
• specs/compiler/core/t27core.t27 compiles string literals, str/string types, .len, and indexing byte-identical to gen-c output.
• Each lossy shape from gen-c: strings and slices lowered to bare pointers -- == compares addresses, .len on str and on slices is not C, escapes and trigraphs change value, a string where an integer goes compiles #8013 produces E_STRING (15) in its refusal fixture.

Boundary

• specs/compiler/core/t27core.t27
• specs/tri/utils/string.t27
• bootstrap/tests/fixtures/trinity_matrix/strings.t27
• .trinity/seals/String.json