Re: [Xen-devel] Hypervisor build can't work without `python` (Was: Re: Xen commit 9b0bc91b3 possibly removed too much info from README)

[Wei Liu <wei.liu2@xxxxxxxxxx>]

On Wed, Apr 17, 2019 at 02:44:21PM +0200, Juergen Gross wrote:
> On 17/04/2019 13:53, Andrew Cooper wrote:
> > On 17/04/2019 12:43, Juergen Gross wrote:
> >> On 17/04/2019 13:26, Wei Liu wrote:
> >>> On Wed, Apr 17, 2019 at 12:23:29PM +0100, Andrew Cooper wrote:
> >>>> On 17/04/2019 12:16, Wei Liu wrote:
> >>>>> On Wed, Apr 17, 2019 at 12:15:04PM +0100, Andrew Cooper wrote:
> >>>>>> On 17/04/2019 12:03, Wei Liu wrote:
> >>>>>>> On Wed, Apr 17, 2019 at 11:58:49AM +0100, Andrew Cooper wrote:
> >>>>>>>> On 17/04/2019 11:57, Wei Liu wrote:
> >>>>>>>>> On Wed, Apr 17, 2019 at 11:44:36AM +0100, Andrew Cooper wrote:
> >>>>>>>>>> On 17/04/2019 11:41, Wei Liu wrote:
> >>>>>>>>>>> On Wed, Apr 17, 2019 at 10:56:57AM +0100, Wei Liu wrote:
> >>>>>>>>>>>>> Here's what i did having pulled the master at commit cb70a26
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> tar xf /path/to/xen-cb70a26.tar.gz
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> cd xen-master/
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> PYTHON=/usr/bin/python3 ./configure --prefix=/usr   \
> >>>>>>>>>>>>>    --disable-seabios   \
> >>>>>>>>>>>>>    --disable-qemu-traditional  \
> >>>>>>>>>>>>>    --disable-rombios     \
> >>>>>>>>>>>>>    --disable-stubdom 2>&1 | tee ../config.log
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> make EFI_DIR=/usr/lib/efi world 2>&1 | tee ../make_world_3.out
> >>>>>>>>>>>>>
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> ...
> >>>>>>>>>>>>> gcc -Wp,-MD,tools/kconfig/.zconf.tab.o.d    -D_GNU_SOURCE
> >>>>>>>>>>>>> -D_DEFAULT_SOURCE -DCURSES_LOC="<ncurses.h>" 
> >>>>>>>>>>>>> -DNCURSES_WIDECHAR=1
> >>>>>>>>>>>>> -DLOCALE  -Itools/kconfig -c -o tools/kconfig/zconf.tab.o
> >>>>>>>>>>>>> tools/kconfig/zconf.tab.c
> >>>>>>>>>>>>> gcc  -o tools/kconfig/conf tools/kconfig/conf.o 
> >>>>>>>>>>>>> tools/kconfig/zconf.tab.o
> >>>>>>>>>>>>> tools/kconfig/conf -s --silentoldconfig Kconfig
> >>>>>>>>>>>>> make[3]: Leaving directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>> make -f Rules.mk _install
> >>>>>>>>>>>>> make[3]: Entering directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>> make -C tools
> >>>>>>>>>>>>> make[4]: Entering directory '/usr/src/xen/xen-master/xen/tools'
> >>>>>>>>>>>>> make symbols
> >>>>>>>>>>>>> make[5]: Entering directory '/usr/src/xen/xen-master/xen/tools'
> >>>>>>>>>>>>> gcc -Wall -Werror -Wstrict-prototypes -O2 -fomit-frame-pointer
> >>>>>>>>>>>>> -fno-strict-aliasing -Wdeclaration-after-statement -o symbols
> >>>>>>>>>>>>> symbols.c
> >>>>>>>>>>>>> make[5]: Leaving directory '/usr/src/xen/xen-master/xen/tools'
> >>>>>>>>>>>>> make[4]: Leaving directory '/usr/src/xen/xen-master/xen/tools'
> >>>>>>>>>>>>> make -f /usr/src/xen/xen-master/xen/Rules.mk 
> >>>>>>>>>>>>> include/xen/compile.h
> >>>>>>>>>>>>> make[4]: Entering directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>>  Xen 4.13-unstable
> >>>>>>>>>>>>> /bin/sh: python: command not found
> >>>>>>>>>>>>> make[4]: *** [Makefile:169: include/xen/compile.h] Error 127
> >>>>>>>>>>>> Here. The PYTHON invocation is from the hypervisor build, not 
> >>>>>>>>>>>> the tools
> >>>>>>>>>>>> build.
> >>>>>>>>>>>>
> >>>>>>>>>>>> Thanks for the log.  I will have a closer look.
> >>>>>>>>>>>>
> >>>>>>>>>>>>> make[4]: Leaving directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>> make[3]: *** [Makefile:137: /usr/src/xen/xen-master/xen/xen] 
> >>>>>>>>>>>>> Error 2
> >>>>>>>>>>>>> make[3]: Leaving directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>> make[2]: *** [Makefile:45: install] Error 2
> >>>>>>>>>>>>> make[2]: Leaving directory '/usr/src/xen/xen-master/xen'
> >>>>>>>>>>>>> make[1]: *** [Makefile:123: install-xen] Error 2
> >>>>>>>>>>>>> make[1]: Leaving directory '/usr/src/xen/xen-master'
> >>>>>>>>>>>>> make: *** [Makefile:165: world] Error 2
> >>>>>>>>>>>>>
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> Note the
> >>>>>>>>>>>>>
> >>>>>>>>>>>>> /bin/sh: python: command not found
> >>>>>>>>>>>>>
> >>>>>>>>>>> The issue here is on LFS there is no /usr/bin/python. So the 
> >>>>>>>>>>> hypervisor
> >>>>>>>>>>> build is broken because xen/include/compile.h invokes
> >>>>>>>>>>>
> >>>>>>>>>>>  @$(PYTHON) tools/fig-to-oct.py < .banner >> $@.new
> >>>>>>>>>>>
> >>>>>>>>>>> There are other appearances of PYTHON in hypervisor build, which 
> >>>>>>>>>>> are
> >>>>>>>>>>> equally broken.
> >>>>>>>>>>>
> >>>>>>>>>>> We had / have a policy to avoid making hypervisor build dependent 
> >>>>>>>>>>> on
> >>>>>>>>>>> configure. I have CC'ed more people on their opinions.
> >>>>>>>>>> Where has $(PYTHON) got lost?  It was specified correctly to begin 
> >>>>>>>>>> with.
> >>>>>>>>> Where is it specified?
> >>>>>>>> Oh - I see now.
> >>>>>>>>
> >>>>>>>> It is specified for ./configure, but not for make.
> >>>>>>> Right.
> >>>>>>>
> >>>>>>> It works for tools because tools build includes config/Tools.mk which
> >>>>>>> contains the correct python path. Tools.mk is generated by configure.
> >>>>>>>  
> >>>>>>> Obviously hypervisor build shouldn't include Tools.mk. Furthermore, 
> >>>>>>> the
> >>>>>>> policy of "hv shouldn't depend on configure" makes this route a no-go.
> >>>>>>>
> >>>>>>> Either we change the policy, or we document the deficiency, or we come
> >>>>>>> up with some clever trick to probe.
> >>>>>> Given the shebang in all of our scripts, would `env python` be
> >>>>>> sufficiently clever trickery, or is that a can of worms we'd prefer not
> >>>>>> to open?
> >>>>> I already tried env python -- it didn't work. I think it searches PATHS
> >>>>> just like a shell would.
> >>>> So this particular use of python is completely trivial.  It is an octal
> >>>> conversion on the banner string, to feed into a string for XEN_BANNER
> >>>>
> >>>> It exists because I removed a fork of figlet[1] from Xen's source tree
> >>>> with this script, but its probably just as easy to do with awk or
> >>>> something similar.
> >>>>
> >>>> ~Andrew
> >>>>
> >>>> [1] Yes - Xen did use to carry a fork of figlet which differed solely by
> >>>> being able to output an octal-escaped string.  No - I have no words
> >>>> which can possibly be used to defend this...
> >>>
> >>> But there are other less trivial ones. See xen/include/Makefile. XSM
> >>> also relies on it to generate policy. What do you want to do with those?
> >> Shouldn't _some_ settings of configure end up in the hypervisor build,
> >> too? Like CC, AWK, SHELL, HOSTCC?
> >>
> >> Maybe we need a rather small basic_tools.mk containing those settings
> >> (with a sane default, of course)?
> > 
> > In the past, there has been a vehement objection to requiring
> > ./configure before building Xen, and I still agree with this point of
> > view.  Noone should need to run ./configure to be able to build a
> > freestanding item with no userspace connection.
> > 
> > The problem Xen has is that we've got two vastly different projects in
> > the same source repo, and `make` runes which encompass both.
> > 
> > Could we go with some compromise where the results of configure are used
> > if they are available, but for people not wanting to run configure, if
> > they've got non-standard standard tools, they need to pass them to make
> > as well?  This is consistent with how cross compiling works.
> 
> That should be fairly easy.
> 
> Let Config.mk have:
> 
> -include config/basic_tools.mk
> 
> at the very beginning, with config/basic_tools.mk being created via
> configure and containing the basic tool variables:
> 
> CC := gcc
> AWK := awk
> PYTHON := python3
> ...
> 
> As Config.mk will set those variables with defualt values in case they
> are empty all is fine then.

Frankly I'm not a big fan of this approach -- you can potentially get
different toolchain depending on how you build stuff. I would rather
just document that for hypervisor build either python is available or
you should specify PYTHON= when invoking make.

Wei.

> 
> 
> Juergen

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel