Buildroot Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: Alexander Sverdlin <alexander.sverdlin@gmail.com>
To: Roy Kollen Svendsen <roykollensvendsen@gmail.com>,
	buildroot@buildroot.org
Subject: Re: [Buildroot] [PATCH] package/readline: link host shared library against ncurses
Date: Sun, 11 Oct 2026 10:28:11 +0200	[thread overview]
Message-ID: <8006a4ffe97e560506b5e7dfa8634f18dceaae6b.camel@gmail.com> (raw)
In-Reply-To: <20260602114732.1957638-1-roykollensvendsen@gmail.com>

Hi Roy,

thanks for looking into this!
I've just stubled upon the same issue, however...

On Tue, 2026-06-02 at 13:47 +0200, Roy Kollen Svendsen wrote:
> readline builds its shared library without linking the curses/termcap
> library, so termcap symbols (UP, BC, PC) are left undefined.
> 
> host-gdb pulls in host-readline, installing this libreadline.so into the
> host directory. buildroot runs its fakeroot finalization script with
> /bin/sh, and fakeroot puts $(HOST_DIR)/lib in LD_LIBRARY_PATH; so when
> /bin/sh is linked against readline (e.g. bash) it loads this host
> libreadline.so rather than the system one, and because UP is a data
> symbol the loader aborts before /bin/sh can start:
> 
>   /bin/sh: symbol lookup error: libreadline.so.8: undefined symbol: UP
> 
> This only happens when the host's /bin/sh links readline and host-gdb is
> enabled; otherwise the host libreadline.so is never loaded, which is why
> it is not seen on every build host.
> 
> Pass SHLIB_LIBS so readline's own build links the host shared library
> against the libncursesw that host-ncurses already provides (it carries
> the termcap symbols), making the host libreadline.so self-contained.
> Only the host library is changed: the target library is never loaded
> during the build, so it cannot break image generation.
> 
> Fixes: d98e15a3639d ("package/gdb: really fix build with GCC 15.x by using external readline")

One can argue that the problem was introduced even earlier by commit
baa81362ef ("package/readline: Use configure options to link with ncurses"),
but wasn't triggered until d98e15a3639d...

> Signed-off-by: Roy Kollen Svendsen <roykollensvendsen@gmail.com>
> ---
>  package/readline/readline.mk | 9 +++++++++
>  1 file changed, 9 insertions(+)
> 
> diff --git a/package/readline/readline.mk b/package/readline/readline.mk
> index f69cbe45e2..99d3711e7b 100644
> --- a/package/readline/readline.mk
> +++ b/package/readline/readline.mk
> @@ -30,5 +30,14 @@ define READLINE_INSTALL_INPUTRC
>  endef
>  READLINE_POST_INSTALL_TARGET_HOOKS += READLINE_INSTALL_INPUTRC
>  
> +# readline does not link its shared library against the curses/termcap
> +# library, leaving termcap symbols (UP, BC, PC) undefined. fakeroot runs
> +# its finalization script with $(HOST_DIR)/lib in LD_LIBRARY_PATH, so a
> +# readline-linked /bin/sh loads this host libreadline.so during root
> +# filesystem generation and aborts on the undefined symbols. Link it
> +# against the libncursesw that host-ncurses provides (it carries those
> +# symbols) via readline's SHLIB_LIBS.
> +HOST_READLINE_MAKE_OPTS += SHLIB_LIBS="-L$(HOST_DIR)/lib -lncursesw"

This assumes that host-ncurses always builds the wide variant... Should we
just mimic commit baa81362ef instead? Something like

HOST_READLINE_CONF_ENV = $(READLINE_CONF_ENV)
HOST_READLINE_CONF_OPTS = $(READLINE_CONF_OPTS)

?

> +
>  $(eval $(autotools-package))
>  $(eval $(host-autotools-package))

-- 
Alexander Sverdlin.
_______________________________________________
buildroot mailing list
buildroot@buildroot.org
https://lists.buildroot.org/mailman/listinfo/buildroot

  reply	other threads:[~2026-10-11  8:28 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-06-02 11:47 [Buildroot] [PATCH] package/readline: link host shared library against ncurses Roy Kollen Svendsen
2026-10-11  8:28 ` Alexander Sverdlin [this message]
2026-10-11  8:50   ` Thomas Petazzoni via buildroot
2026-10-11  9:15     ` Alexander Sverdlin
2026-10-11 21:24 ` Arnout Vandecappelle via buildroot

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=8006a4ffe97e560506b5e7dfa8634f18dceaae6b.camel@gmail.com \
    --to=alexander.sverdlin@gmail.com \
    --cc=buildroot@buildroot.org \
    --cc=roykollensvendsen@gmail.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox