[Buildroot] [PATCH v3 2/7] pkg-autotools: optimistically set runstatedir

[Yann E. MORIN]

Norbert, All,

On 2022-02-22 11:22 +0100, Norbert Lange spake thusly:
> since Autotools 2.70 there is an option to set the
> runstatedir. To support configure scripts using older version,
> the variable is set directly instead of using the --runstatedir
> argument.
> 
> Signed-off-by: Norbert Lange <nolange79 at gmail.com>
> ---
>  package/pkg-autotools.mk | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/package/pkg-autotools.mk b/package/pkg-autotools.mk
> index 717ff81593..b6224b349d 100644
> --- a/package/pkg-autotools.mk
> +++ b/package/pkg-autotools.mk
> @@ -187,6 +187,7 @@ define $(2)_CONFIGURE_CMDS
>  		--exec-prefix=/usr \
>  		--sysconfdir=/etc \
>  		--localstatedir=/var \
> +		runstatedir=/run \

We have unfortunately had to revert this change as well as all the
various hacke we tried to fix it, for various reasons (list below
serving as a memo for the futur):

 1. --runstatedir only appeared in autotools 2.69b, and we still have
    packages that were autoconf-ed with previous versions, and thus
    do not support the option;

So we tried with passing it as a variable assignment, but...

 2. passing runstatedir=/path as a variable assignment on the command
    line actually breaks packages with very old autotools versions, like
    2.13, which get confused as they believe this is specifying the
    host;

So we got back to an option, but only if we could grep it in the
configure script; yet...

 3. packages that do recognise --runstatedir in their top-level
    configure script, may still break with configure scripts in
    sub-directories, written with older autoconf vrsions, so back to
    square one.

Given that we are nearing the release, given that FHS 3.15 does
acknowledge that /var/run can continue to exist for backward
compatibility, and that it suggests individual programs (not whole
systems!) to only use one or the other, we should just continue to
support /var/run (if at least for binary-only cruft that may still be
lying around and need /var/run anyway).

As a consequence, not having a generic solution to pass --runstatedir
(or an equivalent) is not a blocker, and is not a regression either.

Sure, having it would be better, but we can't make it work reasonably
well for now.

So, here is a brain-dmup of some thoughts about that (stil just thinking
about autotools here):

  - packages that actually want to access runstatedir will have a way to
    specify it, be it the option or the variable assignment, or even
    something else;

  - packages that have neither the option nor the variable, would most
    probably not need to access runstatedir at all.

As a consequence, what are the odds that a package needs runstatedir?
Are they legion, or are they few? If they are relatively few, then we
can just add the applicable solution (option or variable) on a
per-package basis.

And I think this is most probably the simplest and most efficient
solution, as we can't have a generic one.

Unless we can come with a much clever solution...

Speaking of cleverness: Peter suggested checking for --runstatedir
recursively in all configure scripts, and only pass it if they all
suport it.

However, I think this might be incorrect: the top-level configure script
could support it, and the package needs it, but sub-configures may not
support it (or there is a file named configure in a sub-dir, which is
not an actual configure script (or is not needed), in which case we
would miss the opportunity for the generic solution, and we'd be back on
a per-package _CONF_OPTS...

So, I'm still thinking the per-package _CONF_OPTS is way better
overall...

Untill we can get a better solution, or all upstream packages have been
fixed and bumped in Buildroot... :-]

Thanks for the initial patch! We tried, it did not work, we reverted.

Regards,
Yann E. MORIN.

>  		--program-prefix="" \
>  		--disable-gtk-doc \
>  		--disable-gtk-doc-html \
> -- 
> 2.34.1
> 
> _______________________________________________
> buildroot mailing list
> buildroot at buildroot.org
> https://lists.buildroot.org/mailman/listinfo/buildroot

-- 
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'