Re: [PING] [PATCH v2] manual: Document missing feature test macros.

[Joseph Myers <joseph at codesourcery dot com>]

On Fri, 11 Aug 2017, Rical Jasan wrote:

> On 08/10/2017 03:26 PM, Joseph Myers wrote:
> > OK, but suggested followups:
> > 
> > * In the description of _DEFAULT_SOURCE, note that features from the C 
> > standard version used by the compiler are also enabled by default.  (With 
> > GCC 5 and later, which defaults to -std=gnu11, that means C11 features are 
> > enabled, but unlike C99 they aren't part of the 2008 edition of POSIX, so 
> > aren't already included in the description of default features.)
> 
> How would you like that said?  That whole paragraph refers to /either/
> feature test macros /or/ compiler options, and uses examples of the
> latter, so I already read it as implying if you use a specific compiler
> option such as -std=gnu11, you get what it entails, along with anything
> that is "default":

I read the existing text as saying that certain strict conformance options 
*disable* features that would otherwise be enabled by default; nothing 
about compiler options serving to *enable* visibility of features not 
otherwise listed as present by default.

Maybe more of a rework is needed in this area.  It's also the case that 
this section of the manual is preferentially referring to the old -ansi 
option, when I'd think -std=c11 would be a better example of a standards 
conformance option in most places (de-emphasising the use of options for 
old standards).  Information to include (some is already present) 
includes:

* Features from the compiler's language version are enabled, regardless of 
any feature test macros passed.

* If you use a strict conformance option, this disables features beyond 
those from the compiler's language version, and feature test macros need 
to be used to enable them.

* If you don't use a strict conformance option, but you use any C/POSIX 
feature test macros, this overrides the default enabling of POSIX.1-2008 
features.  (The logic for this in features.h actually checks for 
_ISOC99_SOURCE but not _ISOC11_SOURCE.  That inconsistency should be 
fixed, presumably by adding _ISOC99_SOURCE to the conditionals.)

* _ISOC99_SOURCE and _ISOC11_SOURCE are not generally needed; they are 
only relevant if you specifically want to add those newer library features 
on top of an older language version (e.g. -std=c90 -D_ISOC99_SOURCE).

> "@Theglibc{} has a complete implementation of the @w{ISO C99} standard,
> and to enable its features, the macro @code{_ISOC99_SOURCE} should be
> defined."

I think we only need to comment on the state of the glibc implementation 
when it's significantly incomplete (as for various __STDC_WANT_*).  It's 
more relevant to point out here that there is no need for this macro if 
your compiler (+ options) uses C99 or later.

-- 
Joseph S. Myers
joseph@codesourcery.com