'Value too large for defined data type' errors with 32-bit OCaml

[jonahbeckford]

The smallest reproducible example is available at https://github.com/diskuv/diskuvbox/blob/e1d1819d4718a061cfa661b8d8ff21d1f27e04d5/support/ .

Assuming test files:

$ ls -lh _build/test_20m _build/test_5gb
-rw-r--r-- 1 root root  20M Aug  7 03:43 _build/test_20m
-rw-r--r-- 1 root root 5.0G Aug  7 03:43 _build/test_5gb

and an 32-bit OCaml interpreter or compiler:

#use "topfind";;
#require "bos";;
#require "fmt";;
#require "fpath";;

open Bos;;

let print_result =
  Fmt.pr "Result: %a@\n"
    Rresult.R.(
      pp
        ~ok:Fmt.(const string "Ok ( Fpath.v {|" ++ Fpath.pp ++ const string "|} )")
        ~error:Fmt.(const string "Error ( `Msg {|" ++ pp_msg ++ const string "|} )"))
    ;;

print_endline "\n\n========= 20 MB ========\n\n";;
print_result @@ OS.File.must_exist (Fpath.v "_build/test_20m") ;;

print_endline "\n\n========= 5 GB ========\n\n";;
print_result @@ OS.File.must_exist (Fpath.v "_build/test_5gb") ;;

we get:

========= 20 MB ========


- : unit = ()
# Result: Ok ( Fpath.v {|_build/test_20m|} )
- : unit = ()
#

========= 5 GB ========


- : unit = ()
# Result: Error ( `Msg {|file _build/test_5gb must exist: Value too large for defined data type|} )
- : unit = ()
#

I can reproduce the problem in both Linux x86 and Windows x86.

You can run it yourself by cloning https://github.com/diskuv/diskuvbox and running sh support/test-32bit.sh on a Linux machine that has Docker.

FYI: I have no problem using Bos to copy a 5GB file using OS.File.with_output. The problem seems to be with stat calls in other Bos functions.

[dbuenzli]

Unix.stat is used I guess Unix.LargeFile.stat should be used.