Path: csiph.com!news.mixmin.net!eternal-september.org!reader01.eternal-september.org!.POSTED!not-for-mail
From: Tim Rentsch
Newsgroups: comp.lang.c
Subject: Re: Cake - C23 to C99 transpiler
Date: Wed, 14 Sep 2022 07:14:38 -0700
Organization: A noiseless patient Spider
Lines: 66
Message-ID: <86tu5am4a9.fsf@linuxsc.com>
References: <836c6515-87f5-42b2-98b0-8308173a9120n@googlegroups.com> <8735d39ou5.fsf@bsb.me.uk> <87leqv87t2.fsf@bsb.me.uk> <87fsh386ze.fsf@bsb.me.uk> <8735d380ta.fsf@bsb.me.uk> <87tu5ikivi.fsf@nosuchdomain.example.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Injection-Info: reader01.eternal-september.org; posting-host="e4a07c6eb40a7dcf36ed28254bbf399a"; logging-data="3147276"; mail-complaints-to="abuse@eternal-september.org"; posting-account="U2FsdGVkX18Yj3d0IIAEQL0r74d3Os/N2YwtVquQZjM="
User-Agent: Gnus/5.11 (Gnus v5.11) Emacs/22.4 (gnu/linux)
Cancel-Lock: sha1:2Pr+HWXcX2NK6inkfdPKeZ8IJ+M= sha1:KeXjZv6cNJ+OrN6vt9XiuVxUuaI=
Xref: csiph.com comp.lang.c:167712
Keith Thompson writes:
> Bart writes:
> [...]
>
>> Also, gcc gets away with allowing $ by default; tcc generally copies
>> gcc in terms of how a compiler is invoked, so this surprisingly goes
>> against that.
>
> gcc allowing $ in identifiers is a documented extension for most
> targets, so it doesn't warn about them even with "-pedantic". One could
> argue that this isn't necessarily an "extension", since "other
> implementation-defined characters" are explicitly permitted by C11
> 6.4.2.1.
>
> Interestingly, the C23 draft doesn't have that wording. Instead, it
> allows XID_Start and XID_Continue characters, so it expands the set of
> characters that *all* implementations must support,
I believe that conclusion is not correct. The C23 draft n3047 says
this:
An XID_Start character is an implementation-defined character
whose corresponding code point in ISO/IEC 10646 has the
XID_Start property. An XID_Continue character is an
implementation-defined character whose corresponding code point
in ISO/IEC 10646 has the XID_Continue property.
The presence of the modifying adjective "implementation-defined" in
both cases surely means, at the very least, that implementations
are allowed to subset the Unicode-specified XID_Start/XID_Continue
sets with regard to what characters are allowed in identifiers.
> but doesn't permit '$' (assuming '$' isn't in XID_Start or
> XID_Continue).
I'm not sure this statement is right either (and agreeing with the
assumption that '$' is not in XID_Start or XID_Continue). The
Unicode reference documentation is so bad that I can't make out
with any degree of certainty whether it allows various domains to
extend the XID_Start/XID_Continue sets for the application in
question. The presence of "implementation-defined" further muddies
the waters. To be clear, neither am I saying that I think the
statement is wrong; only that at present there is not enough
information to be confident of either conclusion.
> A conforming C23 implementation can still accept '$' in identifiers
> as an extension, but my understanding is that it must issue a
> warning.
If accepting '$' is only an extension, and not a consequence of
some implementation-defined behavior, then using '$' in an
identifier results in a syntax error, which requires a diagnostic.
Where is the uncertainty?
> In any case, as of C11 a conforming C compiler is not required to accept
> '$' in identifiers *at all*, even with a command-line option, and may
> reject any program that uses it. You can of course choose to rely on
> the behavior of specific compilers, but if you use '$' in identifiers
> then your code is not 100% portable. (REMINDER: "not 100% portable" is
> not necessarily a criticism.)
Yes.