Re: [Nasm-devel] git master branch disallows question mark in identifiers

[C. M. <pu...@38...>]

On at 2018-12-03 16:29 -0800, H. Peter Anvin wrote:
> Hi,
> 
> Thank you for reporting so quickly.

It pays off to browse the repo occasionally =)

> This I guess negated my assumption that ? was only used in TASM mode.

Yeah, and it was documented in the manual as well, at
https://www.nasm.us/xdoc/2.14/html/nasmdoc3.html#section-3.1 :

> Valid characters in labels are letters, numbers, _, $, #, @, ~, .,
> and ?. The only characters which may be used as the first character
> of an identifier are letters, . (with special meaning: see section
> 3.9), _ and ?.

> The ternary operator is *extremely* useful, in particular because of its
> error-suppressing properties (which means that a lot of workarounds like
> ((!!foo)*(bar))+(!foo)*(baz)) don't work the same.

Oh yeah, you're right. Yes, it'd definitely be useful to support both
the operator and question mark in identifiers.

(Side question: Would it be possible to "evaluate" expressions with
strings as result? Eg << db _FAT16 ? "FAT16" : "FAT12" >>. Currently I
use << db "FAT1",'2'+4*!!_FAT16 >>
https://bitbucket.org/ecm/ldosboot/src/b469e243526d4fbdfe79d6ef4ef677ff2372eb19/boot.asm#lines-490
, which only uses a string that fits as a numeric constant in a byte.)

> I think we can make this work at the cost of complicating the lexical analysis
> or require parentheses in certain ambiguous cases. The current NASM lexical
> analyzer doesn't backtrack, and works one token at a time. However, there are
> cases like:
> 
> foo?bar?baz:quux
> 
> ... where only at the end of the expression can you determine that either
> foo?bar or bar?baz has to be an identifier, but not which one.  And that still
> requires a lexical analysis of the entire expression before it is possible to
> make even that determination.
> 
> Interestingly, disallowing '?' itself as an identifier makes it possible to
> implement the MASM-like "db ?" syntax as equivalent to resb, and the "dup"
> syntax really isn't that hard to do either.
> 
> "%define ? 0" could still be made to work, since %define is followed by a
> token, not an expression. It would, of course, replace *ALL* ? (except in
> strings) with 0 (which would be a hack around your particular situation, but
> wouldn't work for anyone who cares about the symbols in the output format.

I don't think this actually works for my use case. If it would replace
*all* question marks with zero, one of my labels like "?frame_bp" would
look like "0frame_bp" to NASM, which is not a valid identifier. But I'm
not sure you mean that?

Regards,
ecm