On Fri, Sep 21, 2018 at 01:13:24AM +0200, Santiago Garcia Mantinan wrote:
> When including the patch I found that you are removing this part of the
> Makefile:
> -ifeq (,$(findstring nostrip,$(DEB_BUILD_OPTIONS)))
> - INSTALL_PROGRAM += -s
> -endif
>
> Which ends up with unstripped binaries, which make lintian unhappy, is there
> any reason to remove the strip flag there?
I'm sorry for not properly checking this. You are absolutely correct
that it is not that simple.
My changelog entry hints the intention: Defer stripping to dh_strip.
Unfortunately, I failed to notice that mbr does not call dh_strip, so in
effect no stripping is performed. Is there a reason for not calling
dh_strip?
Now you also asked why the -s flag needed to be removed. strip is an
architecture-dependent tool. If you run "strip" on a binary of a
different architecture, it loudly fails. Like the compiler, you must
call ${DEB_HOST_GNU_TYPE}-strip for this to work in a cross compilation
setting. The install program however just uses plain strip when given
the -s option. If you insist on not calling dh_strip (which would also
take care of generating a -dbgsym package), the way to fix this is:
INSTALL_PROGRAM += --strip-program=${DEB_HOST_GNU_TYPE}-strip
In my original patch, I considered this complexity not worth the effort
when dh_strip does the right thing.
Thank you for not blindly applying my patch but reviewing/testing it
instead.
Helmut
