fix some codegen bugs: NIM_BOOL, NIM_STATIC_ASSERT, --passc:-std=... (etc)

[timotheecour]

• --passc:-std=c++17 (or similar flags) are now honored instead of silently ignored, by putting --passc options after instead of before the default

• use correct definition for NIM_BOOL in C target. Previous definition was incorrect, as bool is not guaranteed to be compatible with unsigned char (could have different sizeof etc); furthermore, it would cause codegen error for C code that uses _Bool or (after include <stdbool.h>) bool
  note: it's >= C99 but I doubt nim codegen works with < C99; if needed an extra check (or conditional define) can be used

• NIM_NIL is incorrect for c++ < c++11, but fixing it would be hard and maybe not worth it, eg:

when true:
  {.emit:"""
  #include <stdio.h>
  struct Foo {
    int fun(void* a) { return 1;}
    int fun(int a){ return 2; }
  };
  """.}
  type Foo {.importcpp.} = object
  proc fun(foo: var Foo, a: pointer): cint {.importcpp.}
  var foo: Foo
  echo foo.fun(nil)

nim cpp -r --passc=-std=c++03 $timn_D/tests/nim/all/t10447.nim
2
nim cpp -r --passc=-std=c++17 $timn_D/tests/nim/all/t10447.nim
1

(requires rebuilding nim with this PR so that --passc=-std=c++03 is honored)

EDIT

• fixes Unable to override -std:gnu++14 #13933

links

• https://stackoverflow.com/questions/1608318/is-bool-a-native-c-type/1608350
• D20200328T203812

[Araq]

stdbool.h is useless though as it doesn't say anything about sizeof(bool) and Nim requires it to be 1 byte.

[timotheecour]

PTAL

stdbool.h is useless though as it doesn't say anything about sizeof(bool) and Nim requires it to be 1 byte.

• removed stdbool.h and used _Bool directly instead to defined NIM_BOOL, which still fixes issue in top post

• I'm defining a portable NIM_STATIC_ASSERT and checking NIM_STATIC_ASSERT(sizeof(bool)==1); if a weird platform shows up where sizeof(_Bool)!=1, we'll have a CT error instead of potentially weird bugs and we'll adjust nimbase.h accordingly (eg via #ifdef NIM_DEFINE_BOOL_AS_UCHAR)

• NIM_STATIC_ASSERT is better (and easier to reuse) than the old hack as it shows line number and failing condition; I added a test for checking its behavior

• NIM_STATIC_ASSERT is a key ingredient to support CT sizeof for importc types, see importc types: CT sizeof, $, {.sizeof.}, const CVAR {.importc.}: int RFCs#205 for details

• couldn't figure out how to use the testament discard spec for a test like tests/ccgbugs/mstatic_assert.nim so I'm instead compiling it in tests/trunner.nim (after refactoring that file from tests/vm/tevalffi.nim); the few other tests that also need an explicit compilation step (or non-trivial checks on the output) can also be simplified by merging their logic into trunner (better than making testament spec more complex with more magic flags IMO)

• related to that, I've fixed a bug so that CTFFI now tests the c++ (instead of C) target depending on NIM_COMPILE_TO_CPP

[Araq]

it's >= C99 but I doubt nim codegen works with < C99; if needed an extra check (or conditional define) can be used

We support C89.

[timotheecour]

We support C89.

done, that was a simple fix, PTAL

[timotheecour]

friendly ping @Araq