idk. by today most software already assumes 8 bit == byte in subtle ways all over the place to a point you kinda have to use a fully custom or at least fully self reviewed and patched stack of C libraries
so delegating such by now very very edge cases to non standard C seems fine, i.e. seems to IMHO not change much at all in practice
and C/C++ compilers are anyway full of non standard extensions and it's not that CHAR_BIT go away or you as a non-standard extension assume it might not be 8
> most software already assumes 8 bit == byte in subtle ways all over the place
Which is the real reason why 8-bits should be adopted as the standard byte size.
I didn't even realize that the byte was defined as anything other than 8-bits until recently. I have known, for decades, that there were non-8-bit character encodings (including ASCII) and word sizes were all over the map (including some where word size % 8 != 0). Enough thought about that last point should have helped me realize that there were machines where the byte was not 8-bits, yet the rarity of encountering such systems left me with the incorrect notion that a byte was defined as 8-bits.
Now if someone with enough background to figure it out doesn't figure it out, how can someone without that background figure it out? Someone who has only experienced systems with 8-bit bytes. Someone who has only read books that make the explicit assumption of 8-bit bytes (which virtually every book does). Anything they write has the potential of breaking on systems with a different byte size. The idea of writing portable code because the compiler itself is "standards compliant" breaks down. You probably should modify the standard to ensure the code remains portable by either forcing the compiler for non-8-bit systems to handle the exceptions, or simply admitting that compiler does not portable code for non-8-bit systems.
They still exist. You can still run OS 2200 on a Clearpath Dorado.[1] Although it's actually Intel Xeon processors doing an emulation.
Yes, indexing strings of 6-bit FIELDATA characters was a huge headache. UNIVAC had the unfortunate problem of having to settle on a character code in the early 1960s, before ASCII was standardized. At the time, a military 6-bit character set looked like the next big thing. It was better than IBM's code, which mapped to punch card holes and the letters weren't all in one block.
This would be a great setup for a time travelling science fiction where there is some legacy UNIVAC software that needs to be debugged, and John Titor, instead of looking for an IBM 5100, came back to the year 2024 to find a pre-P3477R0 compiler.
The UNIVAC 1108 (and descendants) mainframe architecture was not discontinued in 1986. The company that owned it (Sperry) merged with Burroughs in that year to form Unisys. The platform still exists, but now runs as a software emulator under x86-64. The OS is still maintained and had a new release just last year. Around the time of the merger the old school name “UNIVAC” was retired in a rebranding, but the platform survived.
Its OS, OS 2200, does have a C compiler. Not sure if there ever was a C++ compiler, if there once was it is no longer around. But that C compiler is not being kept up to date with the latest standards, it only officially supports C89/C90 - this is a deeply legacy system, most application software is written in COBOL and the OS itself itself is mainly written in assembler and a proprietary Pascal-like language called “PLUS”. They might add some features from newer standards if particularly valuable, but formal compliance with C99/C11/C17/C23/etc is not a goal.
The OS does contain components written in C++, most notably the HotSpot JVM. However, from what I understand, the JVM actually runs in x86-64 Linux processes on the host system, outside of the emulated mainframe environment, but the mainframe emulator is integrated with those Linux processes so they can access mainframe files/data/apps.
idk. by today most software already assumes 8 bit == byte in subtle ways all over the place to a point you kinda have to use a fully custom or at least fully self reviewed and patched stack of C libraries
so delegating such by now very very edge cases to non standard C seems fine, i.e. seems to IMHO not change much at all in practice
and C/C++ compilers are anyway full of non standard extensions and it's not that CHAR_BIT go away or you as a non-standard extension assume it might not be 8
> most software already assumes 8 bit == byte in subtle ways all over the place
Which is the real reason why 8-bits should be adopted as the standard byte size.
I didn't even realize that the byte was defined as anything other than 8-bits until recently. I have known, for decades, that there were non-8-bit character encodings (including ASCII) and word sizes were all over the map (including some where word size % 8 != 0). Enough thought about that last point should have helped me realize that there were machines where the byte was not 8-bits, yet the rarity of encountering such systems left me with the incorrect notion that a byte was defined as 8-bits.
Now if someone with enough background to figure it out doesn't figure it out, how can someone without that background figure it out? Someone who has only experienced systems with 8-bit bytes. Someone who has only read books that make the explicit assumption of 8-bit bytes (which virtually every book does). Anything they write has the potential of breaking on systems with a different byte size. The idea of writing portable code because the compiler itself is "standards compliant" breaks down. You probably should modify the standard to ensure the code remains portable by either forcing the compiler for non-8-bit systems to handle the exceptions, or simply admitting that compiler does not portable code for non-8-bit systems.
Hopefully we are; it's been a long time, but as I remember indexing in strings on them is a disaster.
They still exist. You can still run OS 2200 on a Clearpath Dorado.[1] Although it's actually Intel Xeon processors doing an emulation.
Yes, indexing strings of 6-bit FIELDATA characters was a huge headache. UNIVAC had the unfortunate problem of having to settle on a character code in the early 1960s, before ASCII was standardized. At the time, a military 6-bit character set looked like the next big thing. It was better than IBM's code, which mapped to punch card holes and the letters weren't all in one block.
[1] https://www.unisys.com/siteassets/collateral/info-sheets/inf...
This would be a great setup for a time travelling science fiction where there is some legacy UNIVAC software that needs to be debugged, and John Titor, instead of looking for an IBM 5100, came back to the year 2024 to find a pre-P3477R0 compiler.
Do UNIVACs care about modern C++ compilers? Do modern C++ compilers care about UNIVACs?
Given that Wikipedia says UNIVAC was discontinued in 1986 I’m pretty sure the answer is no and no!
The UNIVAC 1108 (and descendants) mainframe architecture was not discontinued in 1986. The company that owned it (Sperry) merged with Burroughs in that year to form Unisys. The platform still exists, but now runs as a software emulator under x86-64. The OS is still maintained and had a new release just last year. Around the time of the merger the old school name “UNIVAC” was retired in a rebranding, but the platform survived.
Its OS, OS 2200, does have a C compiler. Not sure if there ever was a C++ compiler, if there once was it is no longer around. But that C compiler is not being kept up to date with the latest standards, it only officially supports C89/C90 - this is a deeply legacy system, most application software is written in COBOL and the OS itself itself is mainly written in assembler and a proprietary Pascal-like language called “PLUS”. They might add some features from newer standards if particularly valuable, but formal compliance with C99/C11/C17/C23/etc is not a goal.
The OS does contain components written in C++, most notably the HotSpot JVM. However, from what I understand, the JVM actually runs in x86-64 Linux processes on the host system, outside of the emulated mainframe environment, but the mainframe emulator is integrated with those Linux processes so they can access mainframe files/data/apps.
That would be a resounding no then. Nice. So why are we talking about an obscure and irrelevant-to-the-discussion platform? Internet comments I swear.