Re: GNU make 3.81 with GCC 14

[Luke Mewburn <luke%mewburn.net@localhost>]

On 24-09-08 15:24, Greg A. Woods wrote:
  | At Sun, 08 Sep 2024 17:48:41 -0400, Greg Troxel <gdt%lexort.com@localhost> wrote:
  | Subject: Re: GNU make 3.81 with GCC 14
  | >
  | > Lloyd Parkes <lloyd%must-have-coffee.gen.nz@localhost> writes:
  | >
  | > > Does the following look like a good idea to add to
  | > > tools/gmake/Makefile? I can create a PR for it easily enough.
  | > >
  | > > 	.if ${HAVE_GCC} > 13
  | > > 	HOST_CFLAGS+=		-std=gnu89
  | > > 	.endif
  | >
  | > My quick reaction is that in theory, each program is written in a
  | > specific language family member (LFM, that I'd call a "language" but
  | > others disagree), such as C89 or C89-with-gnu-stuff.
  | >
  | > Additionally, compilers support multiple LFMs, and unhelpfully default
  | > to newer ones, rater than enforcing that programs in newfangled LFMs
  | > have to pass --std.
  | >
  | > So if make that we have is documented to build with gnu89, we should
  | > just set --std=gnu89 always.
  | >
  | > Or is it c99-almost, and this is a kludge?
  | 
  | Well technically it might be true that each program is written in a
  | specific LFM, we have in fact been compiling the tools by default
  | (natively) with a compiler (GCC-9) that defaults to "-std=gnu11", so in
  | practice the tools are well ready to be compiled by any compiler that
  | implements "gnu11", which as I said earlier is, practically speaking,
  | all of them (all of the usable ones anyway).
  | 
  | On the other hand I expect all the tools code is only really using gnu99
  | features at most, so -std=gnu99 is maybe even safer (as in more widely
  | accepted by other vendor compilers) for now.
  | 
  | 
  | > What is gcc14 defaulting to
  | > for a c std?  Is it really anything newer than c99?
  | 
  | Well right from the fine GCC-14.2 manual it says:
  | 
  |     The default, if no C language dialect options are given, is -std=gnu17.
  | 
  | 
  | So I just started a build (of a slightly older NetBSD source tree) on a
  | FreeBSD-14 host with Clang-16, and I found I needed "-Wno-everything" as
  | well (for tools/mtree!), and I also found that GCC, even back as far as
  | 9.3, happily accepts the same option, so I would propose the following
  | for a good start on supporting modern compilers.
  | 
  | If or when someone encounters a compiler that doesn't support the
  | "-Wno-everything" option, and/or needs something different, then that
  | can be conditioned later, but for now covering GCC and Clang probably
  | covers 99% of the world of systems NetBSD can be built on.

[catching up on old threads]

I don't like your solution of unilaterally adding compiler options
to the host build. I've been bitten by this too much in various build
environments, including the tools build in the past.

If we think we need some compiler-specific options, then we should:

1. Augment tools/compat/configure.ac to use the NB_CHECK_CC_FLAG()
   macro (that I added to tools/compat on 2023-07-20) for the new options.
   Note that NB_CHECK_CC_FLAG() has specific workarounds for clang and
   gcc with different handling for unknown -Wfoo versus -Wno-foo.

2. Apply appropriate fixes to the tools wrapper Makefiles and if
   relevant to the in-tree Makefiles.
   This /may/ need tweaks in bsd.host.mk too, carefully cross-referenced
   with tools/compat/defs.mk.in.
   These fixes should be on a per-tool basis as required (minimal scope
   change). I.e., don't blanket add -std=gnu99 to all tools build, but
   only to those that need the (older) tools source to build with
   modern more sooky compilers.



  | Index: bsd.host.mk
  | ===================================================================
  | RCS file: /cvs/master/m-NetBSD/main/src/share/mk/bsd.host.mk,v
  | retrieving revision 1.5
  | diff -u -u -r1.5 bsd.host.mk
  | --- bsd.host.mk	9 Aug 2020 21:13:38 -0000	1.5
  | +++ bsd.host.mk	8 Sep 2024 21:10:45 -0000
  | @@ -16,14 +16,14 @@
  | 
  |  # Helpers for cross-compiling
  |  HOST_CC?=	cc
  | -HOST_CFLAGS?=	${HOST_DBG}
  | +HOST_CFLAGS?=	${HOST_DBG} -std=gnu99 -Wno-everything
  |  HOST_COMPILE.c?=${HOST_CC} ${HOST_CFLAGS} ${HOST_DTRACE_OPTS} ${HOST_CPPFLAGS} -c
  |  HOST_COMPILE.cc?=      ${HOST_CXX} ${HOST_CXXFLAGS} ${HOST_DTRACE_OPTS} ${HOST_CPPFLAGS} -c
  |  HOST_LINK.cc?=  ${HOST_CXX} ${HOST_CXXFLAGS} ${HOST_CPPFLAGS} ${HOST_LDFLAGS}
  |  .if defined(HOSTPROG_CXX)
  |  HOST_LINK.c?=   ${HOST_LINK.cc}
  |  .else
  | -HOST_LINK.c?=	${HOST_CC} ${HOST_CFLAGS} ${HOST_CPPFLAGS} ${HOST_LDFLAGS}
  | +HOST_LINK.c?=	${HOST_CC} ${HOST_CFLAGS} ${HOST_CPPFLAGS} ${HOST_LDSTATIC} ${HOST_LDFLAGS}
  |  .endif
  | 
  |  HOST_CXX?=	c++
  | 
  | 
  | oh, that last change is because tools should be static too, in my world.

Orthogonal change, and should be considered separately.


Luke.