From mboxrd@z Thu Jan  1 00:00:00 1970
From: Thomas Petazzoni <thomas.petazzoni@bootlin.com>
Date: Thu, 28 Mar 2019 12:14:41 +0100
Subject: [Buildroot] [PATCH v4] mtree: new package
In-Reply-To: <87bm1w9ck5.fsf@haabendal.dk>
References: <20190326142411.6943-1-esben.haabendal@gmail.com>
 <20190327201803.6b1a832d@windsurf> <87bm1w9ck5.fsf@haabendal.dk>
Message-ID: <20190328121441.02648d09@windsurf>
List-Id: <buildroot.busybox.net>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Transfer-Encoding: 7bit
To: buildroot@busybox.net

Hello Esben,

On Wed, 27 Mar 2019 23:28:58 +0100
Esben Haabendal <esben.haabendal@gmail.com> wrote:

> > If it has been resolved in glibc 2.23, should
> > we remove those work-arounds ?  
> 
> It depends.  If we do that, people using external toolchains with glibc
> older than 2.23 would see breakage.
> 
> > I just would like to make sure we don't work around this problem in
> > different ways in different packages.  
> 
> Sounds like a good plan.  Unfortunately, I don't see that any of the
> proposed workaround generally good.
> 
> For packages where large file support is not deemed an important
> feature, The -U_FILE_OFFSET_BITS workaround is good.  But for packages
> where large file support could be relevant, it seems like a bad idea to
> disable it just to stay compatible with old glibc versions that only a
> few people use together with new buildroot versions.
> 
> If you like, I will switch to the -U_FILE_OFFSET_BITS for mtree
> package.  I personally don't need large file support...

OK, I did a bit of testing, taking as example the libcgroup package. I
confirmed that without -U_FILE_OFFSET_BITS, libcgroup fails to build
with the Sourcery ARM toolchain (old glibc), but builds perfectly fine
with the Linaro toolchain (recent glibc).

So, from there, you have two options:

 (1) You don't have much time and you're not willing to not much
     additional work. In this case, use the -U_FILE_OFFSET_BITS trick in
     mtree. This way, we have the same quirk/workaround everywhere for this
     fts.h problem.

 (2) You have more time. In this case, add a hidden option in
     toolchain/Config.in called BR2_TOOLCHAIN_HAS_BROKEN_FTS. This hidden
     option should be selected by the broken ARM Codesourcery toolchain.
     Then all packages that do the -U_FILE_OFFSET_BITS trick can be changed
     to only do it when BR2_TOOLCHAIN_HAS_BROKEN_FTS is enabled
     (including mtree, of course).

     Obviously, this leaves old custom external toolchains potentially
     broken, as for these, the work around will not be applied. If this
     turns out to really be a problem in practice, we can always add an
     option BR2_TOOLCHAIN_EXTERNAL_CUSTOM_HAS_BROKEN_FTS, and let the
     user enable it + a check that if this option is not enabled, the
     external toolchain really does not have broken fts.h.

Let me know which option you chose :-)

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com