Octal integer literals

[colejohnson66]

Sometimes, when writing P/Invoke code, we encounter enum-like constants that are written in C using octal literals. For example, from <bits/poll.h> in glibc, POLLERR, POLLHUP, and POLLNVAL are written using C's octal literal notation (prefixed by a lone zero). These can be translated into C# by writing their equivalent decimal, hexadecimal, or binary value, but this is error-prone.

I propose that C# support octal literals, but with a caveat. In C, it is widely considered a mistake that any zero-prefixed number is octal. Instead, we should require a 0o or 0O (U+0030 DIGIT ZERO followed by either a U+004F LATIN CAPITAL LETTER O or U+006F LATIN SMALL LETTER O) prefix, similar to hexadecimal. There is some precedent for this from multiple languages:

• ECMAScript
• Haskell
• NASM
• PHP
• Python
• Rust
• ...and more

As with other integer literals, underscore-based separators would be allowed.

The language grammar would be altered as follows:

 // 6.4.5.3 Integer literals
 Integer_Literal
     : Decimal_Integer_Literal
     | Hexadecimal_Integer_Literal
     | Binary_Integer_Literal
+    | Octal_Integer_Literal
     ;

+fragment Octal_Integer_Literal
+    : ('0o' | '0O') Decorated_Octal_Digit* Integer_Type_Suffix?
+    ;

+fragment Decorated_Octal_Digit
+    : '_'* Octal_Digit
+    ;

+fragment Octal_Digit
+    : '0'..'7'
+    ;

Alternatively, a different letter can be chosen to avoid visual collision between U+0030 DIGIT ZERO and U+004F LATIN CAPITAL LETTER O. NASM resolves this potential issue by also allowing U+0072 LATIN SMALL LETTER Q as a separator, but this could cause confusion.

[HaloFour]

I remember this coming up during the discussion around binary literals but I don't see it mentioned in any of the meeting notes.

[colejohnson66]

Binary literals are definitely more useful than octal, so it makes sense for them to have been prioritized (assuming that's what happened).

[TahirAhmadov]

Why not 0C/0c?

[DavidArno]

That sounds a bit like bikeshedding. Surely whether it's 0c, 0o or Ook isn't as important as the desire to express numbers using octal?

[TahirAhmadov]

Not really, 1, I support the proposal and "liked" it right away, 2, 0O/0o looks weird and when adding a feature everything has to be done as close to perfectly as possible.

[huoyaoyuan]

Group separated binary literal can somehow solve it. For example, 0b_010_000.