add root/checkedint.h and .c

[WalterBright]

This is a direct translation of druntime's core.checkedint to C++. The idea is to replace the several ad-hoc overflow checks in dmd with these. Even includes unit tests!

[ibuclaw]

Cool. How goes 2.066? Is it currently branched and taking cherry pickings from master?

I would initially suggest wait until after release to make sure this doesn't accidently get pulled in.

[yebblies]

Does this exactly match the core.checkedint api?

[WalterBright]

@ibuclaw this is for 2.067, 2.066 is branched

@yebblies yes

[yebblies]

1. These use a weird mix of int64_t etc and unsigned. I suggest you change it to always use the stdint types.
2. Why do you need the UINT64_MAX hackery?
3. Did your unrelated changes to posix.mak break Brad's recent pulls?

With the api matching, I will completely drop this file for ddmd.

[WalterBright]

1. I found too many problems with stdint types. I gave up on it and went with unsigned.
2. What do you mean?
3. ???

[yebblies]

What is going on here?

[WalterBright]

merge problem. Fixed.

[yebblies]

I found too many problems with stdint types. I gave up on it and went with unsigned.

Can you be more specific? Those types should be straightforward aliases.

[WalterBright]

The macro constants are typed incorrectly with dmc.

[yebblies]

The macro constants are typed incorrectly with dmc.

That's really worrying, but anyway - what does this have to do with using unsigned over uin32_t?

[WalterBright]

UINT32_MAX is the wrong type, so would have to be constantly cast, or the overloads in the unittests are incorrect.

[yebblies]

I have no idea what you mean. You've just redefined UINT32_MAX because it was the wrong type, and now you're saying you can't use uint32_t because UINT32_MAX is STILL the wrong type?

[WalterBright]

Also, dmd throughout uses int and unsigned, not int32_t and uint32_t, so this matches. unsigned long long, however, is a non-portable mess, so I use stdint for that.

[WalterBright]

This experience makes me long (!) for D's predictable integral types.

[yebblies]

Also, dmd throughout uses int and unsigned, not int32_t and uint32_t, so this matches. unsigned long long, however, is a non-portable mess, so I use stdint for that.

So is there a technical limitation that's stopping you from using int32_t or not?

This experience makes me long (!) for D's predictable integral types.

You could always put this off until after the switch to D.

[WalterBright]

int32_t is simply pointless. DMD uses int, it needs to overflow check on ints, not int32_t's.

We could put off everything until after the switch to D. But this addition shouldn't affect ddmd in any material way, you can just skip these files.

[andralex]

Auto-merge toggled on

[Orvid]

This failed to add the new files to the VC projects.

[yebblies]

This failed to add the new files to the VC projects.

VC projects are not officially supported. Feel free to make a pull request to update them.

[quickfur]

This pull introduced a compile error on Linux/x86_64 when building with make DEBUG=1:

g++ -m64 -c -Wno-deprecated -Wstrict-aliasing -fno-exceptions -fno-rtti -D__pascal= -DMARS=1 -DTARGET_LINUX=1 -DDM_TARGET_CPU_X86=1  -g -g3 -DDEBUG=1 -DUNITTEST -DDMDV2=1 -Iroot -Itk -Ibackend -MMD -MF e2ir.deps e2ir.c
In file included from root/checkedint.c:30:0:
root/checkedint.c: In function ‘void unittest2()’:
root/checkedint.c:98:35: error: call of overloaded ‘adds(long long int, long long int, bool&)’ is ambiguous
     assert(adds(2LL, 3LL, overflow) == 5);
                                   ^
root/checkedint.c:98:35: note: candidates are:
root/checkedint.c:56:5: note: int adds(int, int, bool&)
 int adds(int x, int y, bool& overflow)
     ^
root/checkedint.c:85:9: note: int64_t adds(int64_t, int64_t, bool&)
 int64_t adds(int64_t x, int64_t y, bool& overflow)
         ^
In file included from root/checkedint.c:30:0:
root/checkedint.c:102:48: error: call of overloaded ‘adds(long long int, long long int, bool&)’ is ambiguous
     assert(adds(INT64_MIN + 1LL, -1LL, overflow) == INT64_MIN);
                                                ^
root/checkedint.c:102:48: note: candidates are:
root/checkedint.c:56:5: note: int adds(int, int, bool&)
 int adds(int x, int y, bool& overflow)
     ^
root/checkedint.c:85:9: note: int64_t adds(int64_t, int64_t, bool&)
 int64_t adds(int64_t x, int64_t y, bool& overflow)
         ^
In file included from root/checkedint.c:30:0:
root/checkedint.c:109:35: error: call of overloaded ‘adds(long long int, long long int, bool&)’ is ambiguous
     assert(adds(0LL, 0LL, overflow) == 0);
                                   ^
root/checkedint.c:109:35: note: candidates are:
root/checkedint.c:56:5: note: int adds(int, int, bool&)
 int adds(int x, int y, bool& overflow)
     ^
root/checkedint.c:85:9: note: int64_t adds(int64_t, int64_t, bool&)
 int64_t adds(int64_t x, int64_t y, bool& overflow)
         ^

(there are many pages of similar errors that follow this).

[quickfur]

https://issues.dlang.org/show_bug.cgi?id=13460