What does % mean in an OCaml external declaration?

Many external declarations in the OCaml standard library have a % at the beginning of the function name, such as th开发者_开发知识库e definition of int_of_float:

external int_of_float : float -> int = "%intoffloat"

What does the '%' mean?

There's a lot of %foo special primitives hiding in the compiler. I think the best list is available in bytecomp/translcore.ml, in the ocaml compiler sources. Let's see how many I can list here:

• Comparisons: %equal, %notequal, %lessthan, %greaterthan, %lessequal, %greaterequal, %compare

These comparisons have specialized versions for int, float, string, nativeint, int32 and int64, and will auto-specialize if the types are known at compile-time.

• Other primitives: %identity, %ignore, %field0, %field1, %setfield0, %makeblock, %makemutable, %raise, %incr, %decr, %seqand, %seqor, %boolnot
• Int ops: %negint, %succint, %predint, %addint, %subint, %mulint, %divint, %modint, %andint, %orint, %xorint, %lslint, %lsrint, %asrint
• Int comparators (??): %eq, %noteq, %ltint, %leint, %gtint, %geint
• Float ops: %intoffloat, %floatofint, %negfloat, %absfloat, %addfloat, %subfloat, %mulfloat, %divfloat
• Float comparators: %eqfloat, %noteqfloat, %ltfloat, %lefloat, %gtfloat, %gefloat
• String ops: %string_length, %string_safe_get, %string_safe_set, %string_unsafe_get, %string_unsafe_set
• Array ops: %array_length, %array_safe_get, %array_safe_set, %array_unsafe_get, %array_unsafe_set
• Object manipulation: %obj_size, %obj_field, %obj_set_field, %obj_is_int
• Lazy: %lazy_force
• Nativeint,int32,int64 ops: %{nativeint,int32,int64}: _of_int, _to_int, _neg, _add, _sub, _mul, _div, _mod, _and, _or, _xor, _lsl, _lsr, _asr
• Int conversions: %nativeint_{of,to}_int32, int64_{of,to}_int32, int64_{of,to}_nativeint
• Bigarray operations: %caml_ba_ref_{1,2,3}, %caml_ba_set_{1,2,3}, %caml_ba_unsafe_ref_{1,2,3}, %caml_ba_unsafe_set_{1,2,3}
• Object Oriented: %send, %sendself, %sendcache

That's all I can find.

external with % are special external, that will be handled specially by the compiler. For example, with int_of_float, ocamlc will compile it into a call of some C function, but with ocamlopt, it will compile it into some special assembler opcode that transform double to integer.