Re: [Paparazzi-devel] Dependency problem

[Ben Laurie]

On 29 October 2013 20:00, Felix Ruess <address@hidden> wrote:
> Gotta go, but if I look at the generated .depend file in var/ac_name/ap the
> directory prefix is there and seems to be correct at first glance... some
> paths are absolute and some are relative though...

prefixes on the .h files are correct. The problem is with the .o (i.e.
to the left of the : ).

>
>
> On Tue, Oct 29, 2013 at 8:51 PM, Felix Ruess <address@hidden> wrote:
>>
>> Or something like the last solution mentioned here:
>> http://scottmcpeak.com/autodepend/autodepend.html
>>
>>
>> On Tue, Oct 29, 2013 at 8:47 PM, Felix Ruess <address@hidden>
>> wrote:
>>>
>>> Not entirely sure, but there is probably a way to use the -MT option
>>> together with .d dependency files:
>>>
>>> http://stackoverflow.com/questions/97338/gcc-dependency-generation-for-a-different-output-directory
>>>
>>>
>>> On Tue, Oct 29, 2013 at 8:30 PM, Ben Laurie <address@hidden> wrote:
>>>>
>>>> On 29 October 2013 17:51, Ben Laurie <address@hidden> wrote:
>>>> > On 29 October 2013 17:36, Felix Ruess <address@hidden> wrote:
>>>> >>
>>>> >>
>>>> >>
>>>> >> On Tue, Oct 29, 2013 at 6:14 PM, Ben Laurie <address@hidden> wrote:
>>>> >>>
>>>> >>> On 29 October 2013 14:09, Felix Ruess <address@hidden> wrote:
>>>> >>> > Why wouldn't it be legal?
>>>> >>> > The C preprocessor substitutes GPS_TYPE_H by the appropriate
>>>> >>> > header file
>>>> >>> > and
>>>> >>> > then it is included as usual.
>>>> >>>
>>>> >>> Sure, I get it, I just find it mildly surprising.
>>>> >>
>>>> >>
>>>> >> Well, we want to be able to simply include "subsystems/gps.h" in
>>>> >> other parts
>>>> >> of the code and not care about which GPS implementation is actually
>>>> >> used.
>>>> >> This makes that possible.
>>>> >>
>>>> >>> But I find it even more surprising that it doesn't appear to be
>>>> >>> picked
>>>> >>> up by dependency generation.
>>>> >>
>>>> >>
>>>> >> Would be great if you could look into this, I'm a bit short on time
>>>> >> right
>>>> >> now.
>>>> >
>>>> > Yeah, I'm going to.
>>>> >
>>>> >> Hint: maybe we should only use the -M flag of gcc instead of -MM
>>>>
>>>> OK. That's not the problem.
>>>>
>>>> The problem is that gcc -MM (or -M or anything else) strips the
>>>> directory off the front. You put it back as $(OBJDIR), using sed, but
>>>> that's not right - for example, gps_ubx.o needs to be in
>>>> $(OBJDIR)/subsystems/gps.
>>>>
>>>> AFAICS, there's no way to fix this easily. I think the only answer is
>>>> a script that generates dependencies a file at a time and puts the
>>>> subdirectories back in. I'd normally write such a thing in Perl or
>>>> Python, but I'm betting you don't want to add those as dependencies.
>>>> And I don't know how to write ML...
>>>>
>>>> Suggestions?
>>>>
>>>> >>
>>>> >>>
>>>> >>> > As mentioned in a previous mail:
>>>> >>> > Dependency generation is specified in the appropriate Makefile,
>>>> >>> > e.g. for
>>>> >>> > the
>>>> >>> > ARDrone in conf/Makefile.omap Line185
>>>> >>>
>>>> >>> Aha. Thanks.
>>>> >>>
>>>> >>> >
>>>> >>> >
>>>> >>> > On Tue, Oct 29, 2013 at 2:33 PM, Ben Laurie <address@hidden> wrote:
>>>> >>> >>
>>>> >>> >> On 29 October 2013 12:33, Felix Ruess <address@hidden>
>>>> >>> >> wrote:
>>>> >>> >> > I think that is because gps_ubx.h is only "transitively"
>>>> >>> >> > included via
>>>> >>> >> > gps.h
>>>> >>> >> > using #include GPS_TYPE_H
>>>> >>> >>
>>>> >>> >> Arg!
>>>> >>> >>
>>>> >>> >> I didn't even know it was legal to do includes like this. But
>>>> >>> >> surprised they're not discovered. I couldn't track down where/how
>>>> >>> >> dependencies are handled - any hints?
>>>> >>> >>
>>>> >>> >> >
>>>> >>> >> >
>>>> >>> >> > On Tue, Oct 29, 2013 at 1:18 PM, Ben Laurie <address@hidden>
>>>> >>> >> > wrote:
>>>> >>> >> >>
>>>> >>> >> >> On 29 October 2013 12:03, Christophe De Wagter
>>>> >>> >> >> <address@hidden>
>>>> >>> >> >> wrote:
>>>> >>> >> >> > paparazzi (or actually make) tracks dependencies quite well,
>>>> >>> >> >> > however,
>>>> >>> >> >> > make
>>>> >>> >> >> > does not understand the xml-configuration well.
>>>> >>> >> >>
>>>> >>> >> >> Hmm. When I edit gps_ubx.h and do:
>>>> >>> >> >>
>>>> >>> >> >> % make AIRCRAFT=ardrone2_raw ap.compile
>>>> >>> >> >>
>>>> >>> >> >> it doesn't recompile anything. It should.
>>>> >>> >> >>
>>>> >>> >> >> >
>>>> >>> >> >> >  - So if for instance you change your xml and add a <define
>>>> >>> >> >> > ... >
>>>> >>> >> >> > somewhere,
>>>> >>> >> >> > make does not understand which control or driver files are
>>>> >>> >> >> > affected.
>>>> >>> >> >> > You
>>>> >>> >> >> > then MUST clean before recompile.
>>>> >>> >> >> >  - Even worse: if you change the board file to another arch
>>>> >>> >> >> > in the
>>>> >>> >> >> > middle of
>>>> >>> >> >> > some work, it will only rebuild the changed c files so you
>>>> >>> >> >> > can
>>>> >>> >> >> > have
>>>> >>> >> >> > object
>>>> >>> >> >> > files compiled with different architectures being linked
>>>> >>> >> >> > (which
>>>> >>> >> >> > does
>>>> >>> >> >> > fortunately not work)
>>>> >>> >> >> >  - And if you change a module.xml, the build system does not
>>>> >>> >> >> > necessarly
>>>> >>> >> >> > know
>>>> >>> >> >> > it should recompile (whjile for subsystems it does know as
>>>> >>> >> >> > it is
>>>> >>> >> >> > makefile)
>>>> >>> >> >> >
>>>> >>> >> >> > Our rules of thumb:
>>>> >>> >> >> >  - make clean anytime you need to be really sure
>>>> >>> >> >> >  - certainly clean after every xml change
>>>> >>> >> >> >
>>>> >>> >> >> > PS: thanks for the pull request.
>>>> >>> >> >> >
>>>> >>> >> >> > -Christophe
>>>> >>> >> >> >
>>>> >>> >> >> >
>>>> >>> >> >> > On Tue, Oct 29, 2013 at 12:01 PM, Ben Laurie <address@hidden>
>>>> >>> >> >> > wrote:
>>>> >>> >> >> >>
>>>> >>> >> >> >> On 28 October 2013 12:50, Felix Ruess
>>>> >>> >> >> >> <address@hidden>
>>>> >>> >> >> >> wrote:
>>>> >>> >> >> >> > Hi Ben,
>>>> >>> >> >> >> >
>>>> >>> >> >> >> > seems that this is because the gps_ubx_ucenter.c file is
>>>> >>> >> >> >> > INCLUDED
>>>> >>> >> >> >> > in
>>>> >>> >> >> >> > subsystems/gps/gps_ubx.c Line 271 rather than being added
>>>> >>> >> >> >> > as a
>>>> >>> >> >> >> > normal
>>>> >>> >> >> >> > source
>>>> >>> >> >> >> > file.
>>>> >>> >> >> >> > While this was probably easier to implement (to allow
>>>> >>> >> >> >> > access to
>>>> >>> >> >> >> > ubx
>>>> >>> >> >> >> > internals to the ucenter module) it's not very nice.
>>>> >>> >> >> >> > Christophe may have more details as he wrote the
>>>> >>> >> >> >> > gps_ubx_ucenter
>>>> >>> >> >> >> > module.
>>>> >>> >> >> >>
>>>> >>> >> >> >> I am sending a pull request that fixes this.
>>>> >>> >> >> >>
>>>> >>> >> >> >> In general, though, it seems paparazzi doesn't track
>>>> >>> >> >> >> dependencies.
>>>> >>> >> >> >> Am
>>>> >>> >> >> >> I right, or am I missing something?
>>>> >>> >> >> >>
>>>> >>> >> >> >> >
>>>> >>> >> >> >> > Cheers, Felix
>>>> >>> >> >> >> >
>>>> >>> >> >> >> >
>>>> >>> >> >> >> > On Sat, Oct 26, 2013 at 7:55 PM, Ben Laurie
>>>> >>> >> >> >> > <address@hidden>
>>>> >>> >> >> >> > wrote:
>>>> >>> >> >> >> >>
>>>> >>> >> >> >> >> If I modify gps_ubx_center.c, then rebuild ardrone2_raw,
>>>> >>> >> >> >> >> it
>>>> >>> >> >> >> >> does
>>>> >>> >> >> >> >> not
>>>> >>> >> >> >> >> recompile that file...
>>>> >>> >> >> >> >>
>>>> >>> >> >> >> >> _______________________________________________
>>>> >>> >> >> >> >> Paparazzi-devel mailing list
>>>> >>> >> >> >> >> address@hidden
>>>> >>> >> >> >> >>
>>>> >>> >> >> >> >> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >> >> >
>>>> >>> >> >> >> >
>>>> >>> >> >> >> >
>>>> >>> >> >> >> > _______________________________________________
>>>> >>> >> >> >> > Paparazzi-devel mailing list
>>>> >>> >> >> >> > address@hidden
>>>> >>> >> >> >> > https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >> >> >
>>>> >>> >> >> >>
>>>> >>> >> >> >> _______________________________________________
>>>> >>> >> >> >> Paparazzi-devel mailing list
>>>> >>> >> >> >> address@hidden
>>>> >>> >> >> >> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >> >
>>>> >>> >> >> >
>>>> >>> >> >> >
>>>> >>> >> >> > _______________________________________________
>>>> >>> >> >> > Paparazzi-devel mailing list
>>>> >>> >> >> > address@hidden
>>>> >>> >> >> > https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >> >
>>>> >>> >> >>
>>>> >>> >> >> _______________________________________________
>>>> >>> >> >> Paparazzi-devel mailing list
>>>> >>> >> >> address@hidden
>>>> >>> >> >> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >
>>>> >>> >> >
>>>> >>> >> >
>>>> >>> >> > _______________________________________________
>>>> >>> >> > Paparazzi-devel mailing list
>>>> >>> >> > address@hidden
>>>> >>> >> > https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >> >
>>>> >>> >>
>>>> >>> >> _______________________________________________
>>>> >>> >> Paparazzi-devel mailing list
>>>> >>> >> address@hidden
>>>> >>> >> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >
>>>> >>> >
>>>> >>> >
>>>> >>> > _______________________________________________
>>>> >>> > Paparazzi-devel mailing list
>>>> >>> > address@hidden
>>>> >>> > https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>> >
>>>> >>>
>>>> >>> _______________________________________________
>>>> >>> Paparazzi-devel mailing list
>>>> >>> address@hidden
>>>> >>> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>
>>>> >>
>>>> >>
>>>> >> _______________________________________________
>>>> >> Paparazzi-devel mailing list
>>>> >> address@hidden
>>>> >> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>> >>
>>>>
>>>> _______________________________________________
>>>> Paparazzi-devel mailing list
>>>> address@hidden
>>>> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>>>
>>>
>>
>
>
> _______________________________________________
> Paparazzi-devel mailing list
> address@hidden
> https://lists.nongnu.org/mailman/listinfo/paparazzi-devel
>