Clean up conditionals for declaration of gets().

  gets() has the dubious honor of being the only C89 library feature
that has been completely removed from the current C and C++ standards.
glibc follows suit by not declaring it in _GNU_SOURCE mode either, but
it remains present in older compatibility modes.  Internally, two test
cases need to see stdio.h make the declaration, but all our internal
code is of course compiled under _GNU_SOURCE.  This is currently
kludged by duplicating the gets declaration, fortify wrapper and all,
in include/stdio.h.  Also, the conditional in the public headers for
deciding when to declare gets is complicated and repeated in two places.

This patch adds a new macro to features.h that encapsulates the
complicated rule for when to declare gets.  stdio.h and bits/stdio2.h
can simply test __GLIBC_USE (DEPRECATED_GETS).  Then, instead of
having a duplicate gets declaration in include/stdio.h,
debug/tst-chk1.c and stdio-common/tst-gets.c force that flag on.

Tested on x86_64-linux-gnu, OK?

(If you're wondering why this patch doesn't touch libio/iogets.c, it is
because that file defines _IO_gets, not gets.  Yes, this raises
troubling questions about inconsistencies between the internal
declarations in libioP.h and the public ones in stdio.h.)

zw

        * include/features.h (__GLIBC_USE_DEPRECATED_GETS): New macro.
        * libio/stdio.h, libio/bits/stdio2.h: Condition gets on
        __GLIBC_USE (DEPRECATED_GETS).  Update comments to indicate
        gets was removed from C++ in C++14.
        * include/stdio.h [!_ISOMAC]: Force gets to be declared by
        libio/stdio.h, instead of repeating its declaration.
        * debug/tst-chk1.c, stdio-common/tst-gets.c: Force gets to
        be declared, since we are testing it.
        * stdio-common/Makefile (tst-gets.c): Compile with
        -Wno-deprecated-declarations.
---
 debug/tst-chk1.c        |  5 +++++
 include/features.h      |  9 +++++++++
 include/stdio.h         | 18 ------------------
 libio/bits/stdio2.h     |  3 +--
 libio/stdio.h           | 13 +++++--------
 stdio-common/Makefile   |  3 +++
 stdio-common/tst-gets.c |  6 +++++-
 7 files changed, 28 insertions(+), 29 deletions(-)
  

On 21 Nov 2016 12:44, Zack Weinberg wrote:
> gets() has the dubious honor of being the only C89 library feature
> that has been completely removed from the current C and C++ standards.
> glibc follows suit by not declaring it in _GNU_SOURCE mode either, but
> it remains present in older compatibility modes.  Internally, two test
> cases need to see stdio.h make the declaration, but all our internal
> code is of course compiled under _GNU_SOURCE.  This is currently
> kludged by duplicating the gets declaration, fortify wrapper and all,
> in include/stdio.h.  Also, the conditional in the public headers for
> deciding when to declare gets is complicated and repeated in two places.
> 
> This patch adds a new macro to features.h that encapsulates the
> complicated rule for when to declare gets.  stdio.h and bits/stdio2.h
> can simply test __GLIBC_USE (DEPRECATED_GETS).  Then, instead of
> having a duplicate gets declaration in include/stdio.h,
> debug/tst-chk1.c and stdio-common/tst-gets.c force that flag on.

for the tests, what if we forced an older standard ?  that'd provide
double test coverage by also making sure things build in that mode.

your patch appears to have been mangled in transit.  might want to
check that.

> --- a/debug/tst-chk1.c
> +++ b/debug/tst-chk1.c
> @@ -16,6 +16,11 @@
>     License along with the GNU C Library; if not, see
>     <http://www.gnu.org/licenses/>.  */
>  +/* This file tests gets().  Force it to be declared.  */

needs a newline above the comment

> --- a/include/features.h
> +++ b/include/features.h
> @@ -370,6 +370,15 @@
>  # define __USE_FORTIFY_LEVEL 0
>  #endif
>  +/* The function 'gets' existed in C89, but is impossible to use safely.
> +   It has been removed from ISO C11, ISO C++14, and _GNU_SOURCE.  */

needs a newline above the comment

> --- a/stdio-common/tst-gets.c
> +++ b/stdio-common/tst-gets.c
> @@ -17,10 +17,14 @@
>     License along with the GNU C Library; if not, see
>     <http://www.gnu.org/licenses/>.  */
>  +/* This file tests gets().  Force it to be declared.  */

here too
-mike
  

On Sun, Nov 27, 2016 at 2:13 AM, Mike Frysinger <vapier@gentoo.org> wrote:
> On 21 Nov 2016 12:44, Zack Weinberg wrote:
>> gets() has the dubious honor of being the only C89 library feature
>> that has been completely removed from the current C and C++ standards.
>> glibc follows suit by not declaring it in _GNU_SOURCE mode either, but
>> it remains present in older compatibility modes.  Internally, two test
>> cases need to see stdio.h make the declaration, but all our internal
>> code is of course compiled under _GNU_SOURCE.  This is currently
>> kludged by duplicating the gets declaration, fortify wrapper and all,
>> in include/stdio.h.  Also, the conditional in the public headers for
>> deciding when to declare gets is complicated and repeated in two places.
>>
>> This patch adds a new macro to features.h that encapsulates the
>> complicated rule for when to declare gets.  stdio.h and bits/stdio2.h
>> can simply test __GLIBC_USE (DEPRECATED_GETS).  Then, instead of
>> having a duplicate gets declaration in include/stdio.h,
>> debug/tst-chk1.c and stdio-common/tst-gets.c force that flag on.
>
> for the tests, what if we forced an older standard ?  that'd provide
> double test coverage by also making sure things build in that mode.

That's a good idea but effectively impossible right now; there's no
way to turn off -include libc-symbols.h even for just one test.
Please don't ask me to add one.

> your patch appears to have been mangled in transit.  might want to
> check that.

Yeah, turns out opening "git format-patch" files in Thunderbird
doesn't do what I wanted it to.  I'll resend these patches properly in
a couple days.

Style errors have been corrected in my tree.

zw