Use memchr for safe and performant read_bytestring (#703)

Author: JonasIsensee

CHANGELOG.md

@@ -1,6 +1,7 @@
 ## Unreleased
  - Support for directly reading from and writing to Vector{UInt8}
  - Fix for working with opened IOStream objects
+ - Proper boundscheck for internal `read_bytestring`
 
 ## 0.6.3
  - Fix edge case in datatype loading (#692)

src/io/mmapio.jl

@@ -201,14 +201,13 @@ end
 
 # Read a null-terminated string
 function read_bytestring(io::MmapIO)
-    # Find the null terminator position without allocating
     ptr = pconvert(Ptr{UInt8}, io.curptr)
-    len = 0
-    while unsafe_load(ptr, len + 1) != 0x00
-        len += 1
-    end
-    str = len == 0 ? "" : unsafe_string(ptr, len)
-    io.curptr += len + 1
+    maxbytes = io.endptr - io.curptr + 1
+    q = @ccall memchr(ptr::Ptr{UInt8}, 0::Int32, maxbytes::Csize_t)::Ptr{UInt8}
+    q == C_NULL && throw(EOFError())
+
+    str = unsafe_string(ptr, (q - ptr) % Int)
+    io.curptr += (q - ptr) % Int + 1
     str
 end
 
@@ -239,4 +238,4 @@ function end_checksum(io::MmapIO)
     @inbounds v = io.checksum_pos[io.nchecksum]
     io.nchecksum -= 1
     Lookup3.hash(Ptr{UInt8}(io.startptr + v), position(io) - v)
-end
+end