[Buildroot] [PATCH v2] package/bash: bump to v5.3

[Joseph Kogut]

Hi Julien,

On Thu, Oct 23, 2025 at 1:31 PM Julien Olivain <ju.o at free.fr> wrote:
>
> Hi Joseph,
>
> Thanks for this v2!
>
> On 22/10/2025 22:01, Joseph Kogut wrote:
> > See the relevant announcement:
> > https://lists.gnu.org/archive/html/bug-bash/2025-07/msg00005.html
> >
> > Changes:
> > - Removed upstreamed multibyte compilation fix
> > - Rebased missing include patch that remains unmerged
> >   - Fixed Upstream: checkpackage warning
> > - Removed build failure workarounds enforcing older C standard
> >
> > Signed-off-by: Joseph Kogut <joseph.kogut at gmail.com>
> [...]
> > diff --git a/package/bash/bash.mk b/package/bash/bash.mk
> > index 7fb43c52c0..ed99bbadb2 100644
> > --- a/package/bash/bash.mk
> > +++ b/package/bash/bash.mk
> > @@ -4,7 +4,7 @@
> >  #
> >
> > ################################################################################
> >
> > -BASH_VERSION = 5.2.37
> > +BASH_VERSION = 5.3
>
> This new 5.3 version is introducing 3 build failures. See:
> https://gitlab.com/jolivain/buildroot/-/pipelines/2114876146
>
> For reference, the version 5.2.37 build correctly:
> https://gitlab.com/jolivain/buildroot/-/pipelines/2114892967
>
> >  BASH_SITE = $(BR2_GNU_MIRROR)/bash
> >  BASH_DEPENDENCIES = ncurses readline host-bison
> >  BASH_LICENSE = GPL-3.0+
> > @@ -26,15 +26,6 @@ BASH_CONF_ENV = \
> >       bash_cv_func_sigsetjmp=present \
> >       bash_cv_printf_a_format=yes
> >
> > -# Can be dropped when bash is bumped to 5.3 or newer.
> > -ifeq ($(BR2_TOOLCHAIN_GCC_AT_LEAST_15),y)
> > -BASH_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -std=gnu17"
> > -endif
> > -
> > -ifeq ($(BR2_HOST_GCC_AT_LEAST_15),y)
> > -BASH_CONF_ENV += CFLAGS_FOR_BUILD="$(HOST_CFLAGS) -std=gnu17"
> > -endif
>
> When host-gcc is >= 15 but target-gcc < 15, the build is failing
> because the buildconf.h (initially made for target) is reused for
> host. So we may need to keep this previous block setting
> CFLAGS_FOR_BUILD variable, after all.
>

It looks like we still need the HOST_GCC test, but not the
TOOLCHAIN_GCC test? With both removed, I get 28/33 failures with
test-pkg -a.

With just the toolchain check, I get 33/33 failures, and just the host
check I get 3/33 failures. It looks like those three failures are
still present even with both checks in place, so I'd say we still need
the host check, but not the toolchain check.

My apologies for not running test-pkg before sending v2.

> > -
> >  # The static build needs some trickery
> >  ifeq ($(BR2_STATIC_LIBS),y)
> >  BASH_CONF_OPTS += --enable-static-link
> >
> > ---
>
> Could you have a look, please?
>
> Best regards,
>
> Julien.