Re: [Xen-devel] odd configure warnings on ARM

[Roger Pau Monnà <roger.pau@xxxxxxxxxxxxx>]

2012/3/16 Ian Campbell <Ian.Campbell@xxxxxxxxxx>:
> On Fri, 2012-03-16 at 09:22 +0000, Roger Pau Monnà wrote:
>> 2012/3/16 Ian Campbell <Ian.Campbell@xxxxxxxxxx>:
>> > On Fri, 2012-03-16 at 09:01 +0000, Roger Pau Monnà wrote:
>> >> 2012/3/15 Ian Campbell <Ian.Campbell@xxxxxxxxxx>:
>> >> > Hi Roger,
>> >> >
>> >> > I'm seeing:
>> >> > Â Â Â Âchecking for python version >= 2.3 ... yes
>> >> > Â Â Â Âconfigure: WARNING: Strange ldflag found in python-config 
>> >> > output: -Xlinker
>> >> > Â Â Â Âconfigure: WARNING: Strange ldflag found in python-config 
>> >> > output: -export-dynamic
>> >> > Â Â Â Âconfigure: WARNING: Strange ldflag found in python-config 
>> >> > output: -Wl,-O1
>> >> > Â Â Â Âconfigure: WARNING: Strange ldflag found in python-config 
>> >> > output: -Wl,-Bsymbolic-functions
>> >> >
>> >> > This seems to be tools/m4/python_devel.m4. Is there any reason to print
>> >> > hte unknown flags?
>> >>
>> >> Ian Jackson requested for this, as it was not originally part of my patch:
>> >>
>> >> http://lists.xen.org/archives/html/xen-devel/2012-03/msg00762.html
>> >>
>> >> But I don't really think it's necessary/useful for the end user.
>> >
>> > (adding Ian)
>> >
>> > I don't think that configure should be concerning itself with whatever
>> > random stuff the python developers feel it necessary to include in their
>> > LDFLAGS other than the one we need to look at.
>> >
>> > I'm not sure why we are filtering the result of python-config though.
>> > Shouldn't a test compile using Python.h be using the full set of
>> > whatever python-config says we should use?
>>
>> I'm filtering python-config output to detect which python library to
>> use, since python libraries have the version number attached
>> (-lpython2.6 or -lpython2.7 ...), so I need to know the exact name of
>> the library we are using to perform the correct AC_CHECK_LIB call,
>> which can only test a library at a time, so I can not use "-lpython2.6
>> -Xlinker -export-dynamic ..." as an argument to AC_CHECK_LIB.
>
> Will -lpython2.6 work without these other things though?

It seems to, since we are only using symbol names present in the
library to perform the linking test.

> What about setting LDFLAGS to include the pythohn-config output while
> testing -lpython2.6. Perhaps filtering only python2.6 from the flags?
>
> This doesn't seem like the sort of problem we would be the first to
> encounter. Google found
> http://www.gnu.org/software/autoconf-archive/ax_python.html

I've already found this one in the past, but it is outdated, and has
python versions hardcoded, which seems to be a headache to maintain.

> The second hit was
> http://git.savannah.gnu.org/gitweb/?p=autoconf-archive.git;a=blob_plain;f=m4/ax_python_devel.m4
>
> What about just using one of those?

The last one seems pretty good, it uses a different approach than
mine, using the python distutils module instead of the python-config
tool.

It also uses PYTHON_VERSION env variable to detect python version,
instead of PYTHON that we where using now. Is it suitable to switch
from PYTHON to PYTHON_VERSION to specify PYTHON version to use? Or
should I patch ax_python_devel.m4 to keep using PYTHON?

This also implies that the python parser has to be in the search path
for this macro to work (which should be the usual in most systems).

>> On the other hand, we could craft our own test, using all
>> python-config flags, but I would prefer to keep using provided
>> autoconf macros.
>
> Using autoconf macros or well known macros from elsewhere in preference
> to rolling our own seems like a good idea to me.
>
>>
>> >>
>> >> > I would guess that python-config could produce
>> >> > whatever is necessary for the platform and that the above are totally
>> >> > reasonable.
>> >> >
>> >> > At most this should be a message which ends up in config.log but not on
>> >> > stdout/err, is there a macro for that? If not I think the *) case should
>> >> > just be silent.
>> >> >
>> >> > What do you think?
>> >>
>> >> Autoconf does not provide a macro to write to config.log only, but
>> >> there's a little trick:
>> >>
>> >> http://ac-archive.sourceforge.net/doc/tricks.html
>> >
>> > What a strange way to provide that interface!
>> >
>> >> I'm going to change those messages so that they are only written to
>> >> config.log but not the screen.
>> >>
>> >> Is the configure script working correctly under ARM? I haven't been
>> >> able to get my hands on any ARM board, but I have a rpi reserved that
>> >> I hope doesn't take long (I could also use Qemu I know...).
>> >
>> > It works fine apart from that little niggle. Or at least I haven't
>> > noticed anything else.
>> >
>> > Ian.
>> >
>> >
>
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel