[Buildroot] toolchain-external: ld.so* vs ld.so.*

[Buildroot] toolchain-external: ld.so* vs ld.so.*

[Thomas De Schampheleire]

Hi Thomas,

I have a question on following commit:

commit 32bec8ee2fb00c6750fa842bbb0eb79b0c081fa2
Author: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Date:   Sun Jul 2 15:14:17 2017 +0200

    toolchain-external: copy ld*.so* for all C libraries

in which the base value of TOOLCHAIN_EXTERNAL_LIBS changed: (snippet)

+
+TOOLCHAIN_EXTERNAL_LIBS += ld*.so*
+
[..]
-ifeq ($(BR2_TOOLCHAIN_EXTERNAL_GLIBC)$(BR2_ARM_EABIHF),yy)
-TOOLCHAIN_EXTERNAL_LIBS += ld-linux-armhf.so.*
-else
-TOOLCHAIN_EXTERNAL_LIBS += ld*.so.*
-endif


The question is: did you intentionally remove the . before the final asterisk?
I.e. why is it not:

TOOLCHAIN_EXTERNAL_LIBS += ld*.so.*

as was the case before, even for the glibc+eabihf case?
I could not find a reference to why that specific change was made.

Background is that I now notice (after upgrading to 2018.02 coming from
2017.02.x) that an extra file is copied on my target system: the system used to
have just '/lib/ld.so.1' which is also what is encoded in the ELF files as
dynamic loader, but now there is also '/lib/ld-2.20.so' which is not actually
used and is non-stripped (due to an exception in target-finalize).
This adds about 150K on the root filesystem, which is quite a lot for an unused
file.

So I wonder what would be wrong with following patch:

diff --git a/toolchain/toolchain-external/pkg-toolchain-external.mk b/toolchain/toolchain-external/pkg-toolchain-external.mk
--- a/toolchain/toolchain-external/pkg-toolchain-external.mk
+++ b/toolchain/toolchain-external/pkg-toolchain-external.mk
@@ -108,7 +108,7 @@ endif
 # Definitions of the list of libraries that should be copied to the target.
 #
 
-TOOLCHAIN_EXTERNAL_LIBS += ld*.so* libgcc_s.so.* libatomic.so.*
+TOOLCHAIN_EXTERNAL_LIBS += ld*.so.* libgcc_s.so.* libatomic.so.*
 
 ifeq ($(BR2_TOOLCHAIN_EXTERNAL_GLIBC)$(BR2_TOOLCHAIN_EXTERNAL_UCLIBC),y)
 TOOLCHAIN_EXTERNAL_LIBS += libc.so.* libcrypt.so.* libdl.so.* libm.so.* libnsl.so.* libresolv.so.* librt.so.* libutil.so.*



Thanks,
Thomas

[Buildroot] toolchain-external: ld.so* vs ld.so.*

[Thomas Petazzoni]

Hello,

On Wed, 7 Mar 2018 13:26:47 +0100, Thomas De Schampheleire wrote:

> I have a question on following commit:

You like the difficult questions, pointing out a tiny detail (just a
dot!) in an old patch :-)

> The question is: did you intentionally remove the . before the final asterisk?
> I.e. why is it not:
> 
> TOOLCHAIN_EXTERNAL_LIBS += ld*.so.*
> 
> as was the case before, even for the glibc+eabihf case?
> I could not find a reference to why that specific change was made.
> 
> Background is that I now notice (after upgrading to 2018.02 coming from
> 2017.02.x) that an extra file is copied on my target system: the system used to
> have just '/lib/ld.so.1' which is also what is encoded in the ELF files as
> dynamic loader, but now there is also '/lib/ld-2.20.so' which is not actually
> used and is non-stripped (due to an exception in target-finalize).
> This adds about 150K on the root filesystem, which is quite a lot for an unused
> file.
> 
> So I wonder what would be wrong with following patch:
> 
> diff --git a/toolchain/toolchain-external/pkg-toolchain-external.mk b/toolchain/toolchain-external/pkg-toolchain-external.mk
> --- a/toolchain/toolchain-external/pkg-toolchain-external.mk
> +++ b/toolchain/toolchain-external/pkg-toolchain-external.mk
> @@ -108,7 +108,7 @@ endif
>  # Definitions of the list of libraries that should be copied to the target.
>  #
>  
> -TOOLCHAIN_EXTERNAL_LIBS += ld*.so* libgcc_s.so.* libatomic.so.*
> +TOOLCHAIN_EXTERNAL_LIBS += ld*.so.* libgcc_s.so.* libatomic.so.*
>  
>  ifeq ($(BR2_TOOLCHAIN_EXTERNAL_GLIBC)$(BR2_TOOLCHAIN_EXTERNAL_UCLIBC),y)
>  TOOLCHAIN_EXTERNAL_LIBS += libc.so.* libcrypt.so.* libdl.so.* libm.so.* libnsl.so.* libresolv.so.* librt.so.* libutil.so.*

I looked at the commit and its commit message, and I can't remember why
ld*.so.* was changed to ld*.so*, so I'd say that your patch is probably
correct.

Is there a way to improve our runtime tests to catch problems like
this ?

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
http://bootlin.com

[Buildroot] toolchain-external: ld.so* vs ld.so.*

[Thomas De Schampheleire]

On Wed, Mar 07, 2018 at 01:58:53PM +0100, Thomas Petazzoni wrote:
> Hello,
> 
> On Wed, 7 Mar 2018 13:26:47 +0100, Thomas De Schampheleire wrote:
> 
> > I have a question on following commit:
> 
> You like the difficult questions, pointing out a tiny detail (just a
> dot!) in an old patch :-)

:-P

> 
> > The question is: did you intentionally remove the . before the final asterisk?
> > I.e. why is it not:
> > 
> > TOOLCHAIN_EXTERNAL_LIBS += ld*.so.*
> > 
> > as was the case before, even for the glibc+eabihf case?
> > I could not find a reference to why that specific change was made.
> > 
> > Background is that I now notice (after upgrading to 2018.02 coming from
> > 2017.02.x) that an extra file is copied on my target system: the system used to
> > have just '/lib/ld.so.1' which is also what is encoded in the ELF files as
> > dynamic loader, but now there is also '/lib/ld-2.20.so' which is not actually
> > used and is non-stripped (due to an exception in target-finalize).
> > This adds about 150K on the root filesystem, which is quite a lot for an unused
> > file.
> > 
> > So I wonder what would be wrong with following patch:
> > 
> > diff --git a/toolchain/toolchain-external/pkg-toolchain-external.mk b/toolchain/toolchain-external/pkg-toolchain-external.mk
> > --- a/toolchain/toolchain-external/pkg-toolchain-external.mk
> > +++ b/toolchain/toolchain-external/pkg-toolchain-external.mk
> > @@ -108,7 +108,7 @@ endif
> >  # Definitions of the list of libraries that should be copied to the target.
> >  #
> >  
> > -TOOLCHAIN_EXTERNAL_LIBS += ld*.so* libgcc_s.so.* libatomic.so.*
> > +TOOLCHAIN_EXTERNAL_LIBS += ld*.so.* libgcc_s.so.* libatomic.so.*
> >  
> >  ifeq ($(BR2_TOOLCHAIN_EXTERNAL_GLIBC)$(BR2_TOOLCHAIN_EXTERNAL_UCLIBC),y)
> >  TOOLCHAIN_EXTERNAL_LIBS += libc.so.* libcrypt.so.* libdl.so.* libm.so.* libnsl.so.* libresolv.so.* librt.so.* libutil.so.*
> 
> I looked at the commit and its commit message, and I can't remember why
> ld*.so.* was changed to ld*.so*, so I'd say that your patch is probably
> correct.

Thanks, I'll let it run for a while on our system to see if any issue pops up,
and then I will contribute a patch.

> 
> Is there a way to improve our runtime tests to catch problems like
> this ?

I'm not really sure: we can of course add a hardcoded check for different
toolchains about which files are present or should not be present, but it would
not be a generic check to catch this type of issue.

/Thomas