Re: Use of DESTDIR in xeno-config

[Per Oberg <pero@wolfram.com>]

----- Den 16 nov 2018, på kl 14:29, Jan Kiszka <jan.kiszka@siemens.com> skrev: 

> On 16.11.18 09:14, Per Oberg via Xenomai wrote:
> > Hi everyone

>> I'm confused over the xeno-config use of DESTDIR and cross-compiling. In the
> > documentation (1) it is claimed that it follows the GNU coding standards.

> > Snippet:
>> "If DESTDIR was set when installing Xenomai - typically after cross-compiling -
>> DESTDIR must be set to the same value before calling xeno-config for accessing
> > the target-based directories and files from the build system."

>> The example that is given handles install, and the GNU coding standards (2) also
>> talks about installation. It even specifies: "DESTDIR should be supported only
>> in the install* and uninstall* targets, as those are the only targets where it
> > is useful."

> > However, xeno-config is hardly used during install.

> > My specific problem boils down to the following:
>> I have built a toolchain for Yocto/Poky that has a xeno-config in the sysroot
>> for the target machine. When using it together with autotools used in the
>> etherlab ethercat master i need to set DESTDIR at configure time so that when
>> xeno-config is called the paths to bootstrap.o and dynlib.ld. This feels wrong
> > because of the above statements.

> > Can anyone enlighten me of the correct way of doing this?


> From my reading of both the documentation and what you did, this seems
> consistent: You need to set DESTDIR when you access the toolchain - including
> xeno-config - from within the build system. Once you are on target system (or
> chroot into it), DESTDIR is out of the game.

> The problem when looking at Yocto-like toolchains (SDKs) is that those might be
> relocated (moved around, installed to different directory) after being
> generated. In that case, DESTDIR may actually have to be updated accordingly as
> well. I already received the suggestion internally to derive the necessary base
> path in that case for the location of the compiler, which is doable but feels
> not really correct to me as well.

I see the usefulness, but I fail to see how it's complicit with the GNU standards. What worries me (a little, not much) is that I may be getting other things wrong because of it. 

I'm thinking that this has to be a common problem for most stuff that is cross-compiled with distributed toolchains, and there is the "--sysroot=mysysroot" flag that will make -I=/usr/ and -L=/usr/ translate to -Isysroot/usr and -Lsysroot/usr. Are you saying that there is a hole in the way things are normally done because xenomai is special?

I will not persist, I know to little about it, but I am not convinced, yet.

I do like the idea with picking the location from the script, that would make sense i.m.h.o. 

Per Öberg 

> Jan

> > (1) https://xenomai.org/documentation/xenomai-3/html/man1/xeno-config/index.html
> > (2) https://www.gnu.org/prep/standards/standards.html

> > TTFN
> > Per Öberg



> --
> Siemens AG, Corporate Technology, CT RDA IOT SES-DE
> Corporate Competence Center Embedded Linux