ludic/selfhost/backend/emit_addr.ludic
Orkuncakilkaya 0998cb5a18 feat(compiler): a slice index is checked against its length
A slice has carried { ptr, len, cap } since it existed and nothing ever read
the len: every index emitted a bare getelementptr. Running off the end of one
wrote into whatever the allocator had put next, and the program died somewhere
else entirely - a maroon-lake crash took a day to find because the stack named
a texture upload and the write was in a telemetry buffer five frames earlier.

Now each index loads the length and compares unsigned, which rejects a negative
index in the same instruction, and a failure aborts with the location the other
located errors use:

    oob.ludic:9: index out of range: 7, len 3

`words` and the other raw buffers are unchanged - they are a bare malloc with no
length to check, which is the argument for moving off them.

The SPIR-V variants are regenerated in the same commit: shaders --check was
failing against the edited GLSL.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-22 20:54:12 +03:00

87 lines
4.2 KiB
Text

# emit_addr.ludic — addresses of lvalues (struct fields and slice elements),
# shared by expression loads and assignment stores. Each returns the address
# register; the element/field type is written into g_addr_ty.
var g_addr_ty: pointer # out-param: the type at the computed address
# address of `base.field`
function emit_member_addr(e: Node) -> pointer {
let base = emit_expr(e.a)
let s = layout_node(base.ty)
if (s == null) { perr(`member access on non-aggregate {base.ty}`) }
let fidx = field_index(s, e.s)
if fidx < 0 { perr(`no such field {e.s}`) }
g_addr_ty = field_type(s, e.s)
let r = nreg()
emit(" "); emit(r); emit(" = getelementptr inbounds "); emit(layout_ty(base.ty))
emit(", ptr "); emit(base.code); emit(", i32 0, i32 "); emit(itoa(fidx)); emit("\n")
return r
}
# address of `base[index]` (slices only in this subset)
function emit_index_addr(e: Node) -> pointer {
let base = emit_expr(e.a)
if not is_slice_ty(base.ty) { # a raw pointer: address of element i
let bi = emit_expr(e.b) # (evaluate index first — it may set g_addr_ty)
if (base.ty == "words") { # a `words` buffer: 32-bit int elements
let rw = emit_bind(`getelementptr inbounds i32, ptr {base.code}, i32 {bi.code}`)
g_addr_ty = "int"
return rw
}
if (base.ty == "fixeds") { # a `fixeds` buffer: 32-bit fixed elements
let rf = emit_bind(`getelementptr inbounds i32, ptr {base.code}, i32 {bi.code}`)
g_addr_ty = "fixed"
return rf
}
if (base.ty == "floats") or (base.ty == "doubles") { # IEEE float / double elements
var et = "float"
if (base.ty == "doubles") { et = "double" }
let rd = emit_bind(`getelementptr inbounds {et}, ptr {base.code}, i32 {bi.code}`)
g_addr_ty = et
return rd
}
if (base.ty == "pointers") { # a `pointers` buffer: pointer elements
let rp = emit_bind(`getelementptr inbounds ptr, ptr {base.code}, i32 {bi.code}`)
g_addr_ty = "ptr"
return rp
}
let r = emit_bind(`getelementptr inbounds i8, ptr {base.code}, i32 {bi.code}`) # a ptr/str: bytes
g_addr_ty = "byte" # set last so the caller sees it
return r
}
let el = slice_elem(base.ty)
# load the data pointer from the slice header (field 0)
let dp = nreg()
emit(" "); emit(dp); emit(" = getelementptr inbounds %LSlice, ptr ")
emit(base.code); emit(", i32 0, i32 0\n")
let data = nreg()
emit(" "); emit(data); emit(" = load ptr, ptr "); emit(dp); emit("\n")
let ix = emit_expr(e.b) # (evaluate index BEFORE setting
# THE INDEX IS CHECKED AGAINST THE LENGTH. A slice has carried { ptr, len, cap } since it
# existed, and nothing ever read the len: every index was a bare getelementptr, so running
# off the end of one wrote into whatever the allocator had put next and the program died
# somewhere else entirely, minutes later, with a stack that named an innocent function.
# The comparison is UNSIGNED, which rejects a negative index in the same instruction.
if g_bounds {
g_uses_bounds = true
let lnp = nreg()
emit(" "); emit(lnp); emit(" = getelementptr inbounds %LSlice, ptr ")
emit(base.code); emit(", i32 0, i32 1\n")
let lnv = emit_bind(`load i32, ptr {lnp}`)
let okc = emit_bind(`icmp ult i32 {ix.code}, {lnv}`)
let lok = lbl("ixok")
let lbad = lbl("ixbad")
emit(` br i1 {okc}, label %{lok}, label %{lbad}\n`)
emit(`{lbad}:\n`)
let bmsg = emit_str_const(`{g_src_name}:{itoa(e.line)}: index out of range: `)
let bse = emit_bind(stdstream_rhs(2))
emit(` call i32 (ptr, ptr, ...) @fprintf(ptr {bse}, ptr @.fmt_bounds, ptr {bmsg}, i32 {ix.code}, i32 {lnv})\n`)
emit(" call void @exit(i32 1)\n unreachable\n")
emit(`{lok}:\n`)
}
let r = nreg() # g_addr_ty — a member-access index
emit(" "); emit(r); emit(" = getelementptr inbounds "); emit(llty(el)) # overwrites it)
emit(", ptr "); emit(data); emit(", i32 "); emit(ix.code); emit("\n")
g_addr_ty = el # set last so the caller sees the element type
return r
}