← Back to index

Day 211: GCC/Clang build report

Kim Jørgensen
#5029
In order to compile Handmade Hero on Linux and OS X there are some issues that needs to be fixed. I am looking for volunteers to bring these issues to Casey's attention during the stream.

GCC and Clang seems unhappy with casting a pointer directly to a 16 or 32 bit number. "cast from pointer to smaller type 'u32' (aka 'unsigned int') loses information".

This needs to be addressed in three places:
ParseMember in simple_preprocessor.cpp
1
printf("   {%s, MetaType_%.*s, \"%.*s\", (u32)&((%.*s *)0)->%.*s},\n",


GetOrCreateDebugViewFor in handmade_debug.cpp
1
u32 HashIndex = (((u32)ID.Value[0] >> 2) + ((u32)ID.Value[1] >> 2)) % ArrayCount(DebugState->ViewHash);


CollateDebugRecords in handmade_debug.cpp
1
Region->ColorIndex = (u16)OpeningEvent→BlockName;

This issue can be solved by casting to memory_index aka size_t before casting to u32/u16.

There is some leftover copy/pasta in handmade_platform.h. "extra tokens at end of #endif directive"
1
#endif    debug_table *


There is an empty method in handmade_debug.cpp. "control reaches end of non-void function"
1
2
3
4
inline rectangle2
PlaceRectangle(layout *Layout, v2 Dim)
{
}


_snprintf_s is not supported and needs to be replaced. This will be solved once the dependency to stdio is removed which Casey hopefully will do soon.

/Kim
people
#5030
Yeah, I haven't found a way to compile without "-fpermissive" warnings in gcc, but as long as it compiles and runs, warnings aren't too annoying :)
Nuno
#5036
Hi Kim

I was trying to compile this on a Raspberry Pi just to have a look at how it would behave and I came across the same problems :)

The other problems I came across were the thread_id which is not defined for this platform and of course the intrinsics.. But I think I'll open a new topic to discuss this once these compilation issues are solved.

Thanks
Nuno
Edited by Nuno on
Kim Jørgensen
#5179
All issues from my first post are still present so I'm still looking for somebody who can bring these to Casey's attention during the stream.

Any volunteers?

/Kim
Kim Jørgensen
#5255
Thank you for the help :)
Now we only need to get rid of _snprintf_s

/Kim
Mārtiņš Možeiko
#5257
Or make Casey add _snprintf_s functions for non-MSVC platform:
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
#if !COMPILER_MSVC
inline int _snprintf_s(char *buffer, size_t sizeOfBuffer, size_t count, const char *format, ...)
{
  assert(count <= sizeOfBuffer);
  va_list args;
  va_start(args, format);
  int result = vsnprintf(buffer, count, format, args);
  va_end(args);
  return result;
}

#ifdef __cplusplus
template <size_t size>
inline int _snprintf_s(char (&buffer)[size], size_t count, const char *format, ...)
{
  assert(count <= size);
  va_list args;
  va_start(args, format);
  int result = vsnprintf(buffer, count, format, args);
  va_end(args);
  return result;
}
#endif

#endif
Edited by Mārtiņš Možeiko on
Kim Jørgensen
#5258
Yeah, that would be great :)
skeeto
#5273
Casey fixed this issue in Day 220 while insisting the cast really was permitted. It's definitely permitted in C, but C++ I wasn't so sure about, so I looked it up. In C++11 §5.2.10¶4 (draft):

A pointer can be explicitly converted to any integral type large enough to hold it. The mapping function is implementation-defined.

I couldn't find anywhere in the spec that permitted a cast from a pointer to a smaller integer. Both GCC and Clang emit an error by default because the spec requires them to.

The following would be legal so long as the implementation provides uintptr_t:

1
(u32)(uintptr_t)pointer


Casting from pointer to uintptr_t is allowed, and casting from uintptr_t to u32 is always allowed. Since uintptr_t is optional, as far as I can tell it's not possible to cast a pointer to an integer in all C++ implementations.
Reply to Thread
← Back to index