Re: C time_t 2038 problem s-os_lib.ads

[Keith Thompson <Keith.S.Thompson+u@gmail.com>]

Joakim Strandberg <joakimds@kth.se> writes:
> torsdag 23 september 2021 kl. 17:01:04 UTC+2 skrev m8il...@gmail.com:
>> On Thursday, September 23, 2021 at 2:26:11 PM UTC, Jeffrey R. Carter wrote: 
>> > On 9/23/21 12:42 PM, Kevin Chadwick wrote: 
>> > > I have noticed that C time_t appears to be Long_integer in Gnat s-os_lib.ads. 
>> > > 
>> > > Just wondering if it should be 64bit long long as OpenBSD has already moved to long long? 
>> > GNAT defines 
>> > 
>> > type Long_Integer is range -(2 **63) .. +(2 **63 - 1); 
>> > for Long_Integer'Size use 64; 

Does it do that on all operating systems?  C's type long is 32 bits on
64-bit Windows.  (Of course Ada's Long_Integer doesn't have to match C's
long.)

>> I see, thank you. 
>> 
>> GPS doesn't seem to jump to declaration for Long_Integer and grep hasn't turned it up, 
>> so I will just take your word for it.
>
> Well, yes Long_Integer is 64-bits, but long long in cpp is 128 bits
> which sounds like a discrepancy to me. On OpenBSD it indicates C
> time_t should be changed from Long_Integer to somethinge else that is
> 128-bits.  All packages in Ada has "with Standard; use Standard;"
> which brings Integer etc. into scope. Long_Integer should be defined
> in the Standard package. Under help in GPS it should be possible to
> find the Standard package.

If by "cpp" you mean C++, I've never seen an implementation where long
long is 128 bits (or anything other than exactly 64 bits).

In C and C++, int is required to be at least 16 bits (POSIX requires
32), long is at least 32 bits, and long long is at least 64 bits.  On
most 64-bit Linux-based systems, int is 32 bits, and long and long long
are both 64 bits.  On 64-bit MS Windows, int and long are both 32 bits,
and long long is 64 bits.  time_t is 64 bits on almost all 64-bit
systems.  I've never seen a 128-bit time_t; 64 bits with 1-second
resolution is good for several hundred billion years.

If an Ada implementation makes Integer, Long_Integer, and
Long_Long_Integer correspond to C and C++'s int, long, and long long,
then on a system (e.g.,. Windows) where long is 32 bits, defining time_t
as Long_Integer is going to cause problems in 2038 -- *and* it's likely
not going to match the system's C and C++ time_t definition.

I don't see a definition of "time_t" in s-os_lib.ads on my system.

If an Ada implementation is going to define a type that's intended to
match C's time_t, it should match the representation of that C type.
I presume GNAT gets this right.

-- 
Keith Thompson (The_Other_Keith) Keith.S.Thompson+u@gmail.com
Working, but not speaking, for Philips
void Void(void) { Void(); } /* The recursive call of the void */