Re: technology discussion → does the world need a "new" C ?

On 07/07/2024 00:04, Keith Thompson wrote:

bart <bc@freeuk.com> writes:

On 06/07/2024 02:38, Lawrence D'Oliveiro wrote:

On Fri, 5 Jul 2024 14:31:44 +0100, bart wrote:
>

C also is the only language that is supposed to work on any kind of
processor ...

I don’t think there is anything innate in the design of C to ensure
that.
It was simply its popularity that meant it was usually the first language
implemented on a new processor.
For example, C assumes byte addressability.

>
C didn't define a 'byte' at all. It assumed 'char' addressability, but
allowed that 'char' to be any width at all. I think at some point a
minimum of 8 bits was applied.

 What???
 C defines a "byte" as an "addressable unit of data storage large enough
to hold any member of the basic character set of the execution
environment".  You know that.  C references going back to 1974 all talk
about bytes (the early ones are specific to the PDP-11).
 Perhaps you meant that there's no predefined type named "byte", but
nobody said there was.
 The requirement that a byte is at least 8 bits goes back at least to
C89.  K&R1 (1978) doesn't make this requirement explicit, but shows
examples of 8- and 9-bit bytes.
 [...]
 

Most of today's hardware evolved from such a model: 32- and 64-bit
words and addresses were an obvious natural progression. C however
still hasn't got the memo.

 Right, C makes it *so* difficult to support systems with 8-bit bytes and
32- or 64-bit word.
 

There's no 'byte' type. There's an odd selection of *5* char, short, int, long and long long types which cover the *4* 8/16/32/64 bit sizes.
They include some wonderful denotations such as 'long int unsigned long' (you choose the order you like).
None are defined by the language as any specific size, other than certain minimum widths, and that each successive type is no shorter than the last.
IF you include a special header, THEN it also provides some fixed-width sizes, quite likely defined internally on top of types above. However these are so ugly, that every other C program prefers to define its own aliases. Plus there are 100 special macros to define format codes, MIN/MAX values etc etc.
Meanwhile, format codes, and constant suffixes, are defined in terms of 0, 1 or 2 'longs'.
Most programs also use 'int' extensively, which although is very commonly 32 bits, C only guarantees it to be 16 bits. Lots also use 'long' which is variously 32 or 64 bits.
So it's not 'difficult', just incredibly messy. I haven't even mentioned least and fast types, whatever they do.
Pretty much every language touted as a C replacement which has fixed width integers, has a far tidier and smaller set of more concrete types (eg. 8-10, compared with what, 35?).
They also don't need CHAR_BIT to tell it that chars have 8 bits.
My point was that this evolution was apparent at least 40 years ago.