Module dwarf_line 

VCR-DBG-001 step 3 — compose the source-line table (the join half).

The DWARF Tier-1 bridge maps an ARM text offset back to a source file:line through three established facts:

1. each ARM instruction carries source_line = the wasm OP INDEX (ArmInstruction.source_line);
2. step 1 (FunctionOps.op_offsets) maps op-index → the wasm code BYTE OFFSET (module-relative);
3. step 2 parses the input wasm’s .debug_line → (code-section-relative address → file:line) rows.

This module is the join for the wasm half — op-index → source line — which step 4 (emit) composes with the ARM layout (ARM-text-offset → op-index is just source_line). It is pure plain-data (no gimli, no backend): the caller parses the rows and supplies them, so the module is Bazel-clean and unwired (frozen-safe) until the emitter consumes it.

The crux it encodes (validated on scripts/repro/dwarf_coherent.wasm, VCR-DBG-001 step-3 fixture): op_offsets are MODULE-relative while DWARF addresses are CODE-section-relative, and they differ by a single constant — the code section’s payload start. So normalization is one subtraction: dwarf_addr = op_offset - code_base.

Structs

DwarfTextReloc

A relocation a .debug_* section needs against the .text symbol so a host linker fixes up the embedded .text address when .text is placed. REL form: the in-place bytes already hold the addend (always 0 for our text-base references), so only the site (offset) and size travel here.

EmittedDwarfSection

One emitted .debug_* section: its ELF name, bytes, and the .text-symbol relocations it needs (empty for address-free sections like .debug_str).

InputDwarfLine

Result of reading the input wasm’s DWARF line table: the parsed rows plus the code-section payload start (code_base) the op-offset compose subtracts.

LineRow

One .debug_line row: a code-section-relative address and its source line. file is an opaque caller-supplied id (e.g. an index into the line program’s file table) so this stays gimli-free.

SourceLoc

A resolved source location for a wasm op.

Functions

emit_debug_sections

Emit an address-ordered (arm_addr, line) table as a FULL minimal DWARF unit (gimli::write) and return EVERY non-empty .debug_* section it produces — .debug_info, .debug_abbrev, .debug_str, .debug_line (and .debug_line_str/.debug_ranges etc. when non-empty). The caller composes the table (one address-sorted, de-duped sequence covering every function); this produces the section bytes for non-ALLOC ELF PROGBITS sections. Returns an empty Vec for an empty table (nothing to map ⇒ no sections ⇒ output stays byte-identical).

op_offsets_to_source

Map each wasm op (by its module-relative op_offsets byte offset) to a source location, by normalizing into the code-section-relative DWARF address space (op_offset - code_base) and taking the covering line-table row (the last row whose address is ≤ the op’s address — standard line-table lookup).

read_input_dwarf_line

Read the input wasm’s .debug_line into code-section-relative (addr → line) rows and report code_base. Returns an empty table (rows empty, the feature a no-op) when the input carries no .debug_* sections or no parseable line program — never an error for a DWARF-free module.