Build failure with GCC 15 (C23)

[kostadinsh]

Describe the bug
jq fails to build with upcoming GCC 15 that defaults to C23. This can be reproduced on older GCC and Clang versions by adding -std=gnu23 to CFLAGS.

Downstream Gentoo bug: https://bugs.gentoo.org/944014

Environment (please complete the following information):

• OS: Gentoo Linux
• jq version: happens with both master and 1.7.1

src/builtin.c:1794:4: error: initialization of 'jv (*)(void)' from incompatible pointer type 'jv (*)(jq_state *, jv)' [-Wincompatible-pointer-types]
 1794 |   {f_ ## name, #name, 1},
      |    ^~
src/libm.h:2:1: note: in expansion of macro 'LIBM_DD'
    2 | LIBM_DD(acos)
      | ^~~~~~~
src/builtin.c:1794:4: note: (near initialization for 'function_list[0].fptr')
 1794 |   {f_ ## name, #name, 1},
      |    ^~
src/builtin.c:1800:4: error: initialization of 'jv (*)(void)' from incompatible pointer type 'jv (*)(jq_state *, jv,  jv,  jv)' [-Wincompatible-pointer-types]
 1800 |   {f_ ## name, #name, 3},
      |    ^~
src/libm.h:27:1: note: in expansion of macro 'LIBM_DDD'
   27 | LIBM_DDD(atan2)
      | ^~~~~~~~
src/builtin.c:1800:4: note: (near initialization for 'function_list[5].fptr')
 1800 |   {f_ ## name, #name, 3},
      |    ^~
src/libm.h:27:1: note: in expansion of macro 'LIBM_DDD'
   27 | LIBM_DDD(atan2)
      | ^~~~~~~~
src/builtin.c:1809:4: error: initialization of 'jv (*)(void)' from incompatible pointer type 'jv (*)(jq_state *, jv)' [-Wincompatible-pointer-types]
 1809 |   {f_negate, "_negate", 1},
      |    ^~~~~~~~
src/builtin.c:1809:4: note: (near initialization for 'function_list[61].fptr')
src/builtin.c:1810:22: error: initialization of 'jv (*)(void)' from incompatible pointer type 'jv (*)(jq_state *, jv,  jv,  jv)' [-Wincompatible-pointer-types]
 1810 | #define BINOP(name) {f_ ## name, "_" #name, 3},
      |                      ^~
src/builtin.h:11:3: note: in expansion of macro 'BINOP'
   11 |   BINOP(plus) \
      |   ^~~~~
src/builtin.c:1811:1: note: in expansion of macro 'BINOPS'
 1811 | BINOPS
      | ^~~~~~
src/builtin.c:1810:22: note: (near initialization for 'function_list[62].fptr')
 1810 | #define BINOP(name) {f_ ## name, "_" #name, 3},
      |                      ^~
src/builtin.h:11:3: note: in expansion of macro 'BINOP'
   11 |   BINOP(plus) \
      |   ^~~~~
src/builtin.c:1811:1: note: in expansion of macro 'BINOPS'
 1811 | BINOPS
      | ^~~~~~

Above are just some of the errors the rest are in the build log that's attached
build.log

[emanuele6]

AC_PROG_CC_STDC should tell the compiler to use C99 (or C89 as a fallback if C99 is not supported).

jq/configure.ac

Line 14 in 8619f8a

m4_version_prereq([2.70], [], [AC_PROG_CC_STDC])

https://www.gnu.org/software/autoconf/manual/autoconf-2.60/html_node/C-Compiler.html#index-AC_005fPROG_005fCC_005fSTDC-743

Are you sure the build failure isn't just caused by you forcing -std=gnu23 in CFLAGS?

[kostadinsh]

Are you sure the build failure isn't just caused by you forcing -std=gnu23 in CFLAGS?

I don't have it in CFLAGS, it's the default for GCC 15.

See also: https://inbox.sourceware.org/gcc-cvs/20241115234634.9D1633858CDA@sourceware.org/T/#u

[emanuele6]

Downstream Gentoo bug: https://bugs.gentoo.org/944014

I missed that link, it doesn't look like you are adding it in CFLAGS

[emanuele6]

Are you sure the build failure isn't just caused by you forcing -std=gnu23 in CFLAGS?

I don't have it in CFLAGS, it's the default for GCC 15.

See also: https://inbox.sourceware.org/gcc-cvs/20241115234634.9D1633858CDA@sourceware.org/T/#u

@kostadinsh I know gcc changed the default, but that link doesn't say anything useful. Also adding -std=gnu23 to CFLAGS doesn't really help in reproucing issue since it is a problem related to autoconf, and not the code.
For example, gcc -std=c99 $(USERCFLAGSWITHFORCEDC23) would still cause problems which is what I was saying.
Here AC_PROG_CC_STDC should be dealing with passing something like -std=c99 to the compiler I think, anyway it looks like it isn't doing that in your setup.

It looks like AC_PROG_CC_STDC is now an obsolete undocumented macro, it probably needs to be replaced by something else.