Unable to parse some import values

[rgrinberg]

See, the following error message produced on 4.02.3/OSX:

compiling c program:
 | #include <stdio.h>
 | #include <ffi.h>
 |
 |
 | #define DUNE_ABS(x) ((x >= 0)? x: -(x))
 | #define DUNE_D0(x) ('0'+(DUNE_ABS(x)/1         )%10)
 | #define DUNE_D1(x) ('0'+(DUNE_ABS(x)/10        )%10), DUNE_D0(x)
 | #define DUNE_D2(x) ('0'+(DUNE_ABS(x)/100       )%10), DUNE_D1(x)
 | #define DUNE_D3(x) ('0'+(DUNE_ABS(x)/1000      )%10), DUNE_D2(x)
 | #define DUNE_D4(x) ('0'+(DUNE_ABS(x)/10000     )%10), DUNE_D3(x)
 | #define DUNE_D5(x) ('0'+(DUNE_ABS(x)/100000    )%10), DUNE_D4(x)
 | #define DUNE_D6(x) ('0'+(DUNE_ABS(x)/1000000   )%10), DUNE_D5(x)
 | #define DUNE_D7(x) ('0'+(DUNE_ABS(x)/10000000  )%10), DUNE_D6(x)
 | #define DUNE_D8(x) ('0'+(DUNE_ABS(x)/100000000 )%10), DUNE_D7(x)
 | #define DUNE_D9(x) ('0'+(DUNE_ABS(x)/1000000000)%10), DUNE_D8(x)
 | #define DUNE_SIGN(x) ((x >= 0)? '+': '-')
 |
 |
 | const char s0[] = {
 |   'B', 'E', 'G', 'I', 'N', '-', '0', '-',
 |   DUNE_SIGN((FFI_FASTCALL)),
 |   DUNE_D9((FFI_FASTCALL)),
 |   '-', 'E', 'N', 'D'
 | };
 |
run: cc -O  -Wall -D_FILE_OFFSET_BITS=64 -D_REENTRANT -O  '-I/usr/local/Cellar/libffi/3.2.1/lib/libffi-3.2.1/include' '-I' '/Users/rgrinberg/.opam/4.02.3/lib/ocaml' '-o' '/var/folders/tz/tvzjldvx3s34fnw1qs9lk1240000gn/T/ocaml-configuratordbaa52/c-test-55/test.o' '-c' '/var/folders/tz/tvzjldvx3s34fnw1qs9lk1240000gn/T/ocaml-configuratordbaa52/c-test-55/test.c'
-> process exited with code 0
-> stdout:
-> stderr:
Error: Unable to read variable "FFI_FASTCALL" of type int. Invalid value "+0000000004" in /var/folders/tz/tvzjldvx3s34fnw1qs9lk1240000gn/T/ocaml-configuratordbaa52/c-test-55/test.o found

@Chris00 can you take over this?

[Chris00]

On 4 January 2019 at 20:15 CET, Rudi Grinberg wrote: See, the following error message produced on 4.02.3/OSX: […]Error: Unable to read variable "FFI_FASTCALL" of type int. Invalid value "+0000000004" in /var/folders/tz/tvzjldvx3s34fnw1qs9lk1240000gn/T/ocaml-configuratordbaa52/c-test-55/test.o found @Chris00 can you take over this?

Sure. It is just `int_of_string` that does not like the `+` on 4.02.3. Any idea why that was not caught by the CI when I submitted the PR?

[rgrinberg]

Because we only run the tests on the latest version.

[Chris00]

On 4 January 2019 at 20:33 CET, Rudi Grinberg wrote: Because we only run the tests on the latest version.

Shouldn't they also be run on another version—say the first version of the compiler supported?

[rgrinberg]

Yeah, that certainly be preferable. I think some of our expect tests aren't entirely portable across different versions of the compiler. So it's not so easy to solve, but we should certainly address this.