[PATCH] Avoid default RTEMS application configuration

[Chris Johns]

On 22/10/2018 16:28, Sebastian Huber wrote:
> On 20/10/2018 02:04, Chris Johns wrote:
>> On 19/10/18 5:47 pm, Sebastian Huber wrote:
>>> ----- Am 18. Okt 2018 um 19:56 schrieb Chris Johns chrisj at rtems.org:
>>>
>>>> On 18/10/18 6:38 pm, Sebastian Huber wrote:
>>>>> Use a test body with a proper RTEMS application configuration to avoid a
>>>>> dependency on the default configuration.  Do not include
>>>>> <rtems/score/cpuopts.h> directly since this header file is an
>>>>> implementation detail.
>>>>>
>>>>> Update #3551.
>>>>> ---
>>>>>   rtems.py | 30 +++++++++++++++++-------------
>>>>>   1 file changed, 17 insertions(+), 13 deletions(-)
>>>>>
>>>>> diff --git a/rtems.py b/rtems.py
>>>>> index 1b0da60..c7a1966 100644
>>>>> --- a/rtems.py
>>>>> +++ b/rtems.py
>>>>> @@ -259,13 +259,18 @@ def configure(conf, bsp_configure = None):
>>>>>           #
>>>>>           # Checks for various RTEMS features.
>>>>>           #
>>>>> -        conf.multicheck({ 'header_name': 'rtems/score/cpuopts.h'},
>>>>> -                        msg = 'Checking for RTEMS CPU options header',
>>>>> -                        mandatory = True)
>>>>> -        load_cpuopts(conf, ab, rtems_path)
>>>> OK.
>>>>
>>>>> -        conf.multicheck({ 'header_name': 'rtems.h'},
>>>>> -                        msg = 'Checking for RTEMS header',
>>>>> -                        mandatory = True)
>>>> Why remove the test? I see the app test below checks for the header however the
>>>> test creates a nice specific error message.
>>> The test is not a simple compile test and dosen't only check that you can
>>> include <rtems.h>. In addition it checks that you can link a sample
>>> application successfully.
>> The test you have added is a good addition. I am asking, why not keep both
>> tests? It comes down to the error message and how user friendly it is. If
>> 'rtems.h' is present we can assume it is an installed RTEMS. The second test can
>> check the quality of the install. I would like to avoid users needing to dig
>> into a config log to find an error and then try to understand it to figure out
>> they have a bad path on the command line.
> 
> Waf doesn't simply check that you can include a header file. In addition it
> tries to link an executable:

Thanks, this now makes sense.

> 
> Using a simple main() requires the default configuration.
> 
>>
>> What is the error report with just the 'main' test you added and an invalid
>> RTEMS path?
> 
> When I remove the rtems.h from the BSP installation, then I get this:
> 
> Building a trivial RTEMS application          : no

Should this be:

 Checking for a valid RTEMS BSP installation    : no

?

> The configuration failed
> (complete log in /scratch/git-rtems-libbsd/build/config.log)
> 
> In config.log:
> 
> Building a trivial RTEMS application
> ==>
> #include <rtems.h>
> void Init(rtems_task_argument arg) { (void)arg; }
> #define CONFIGURE_APPLICATION_DOES_NOT_NEED_CLOCK_DRIVER 1
> #define CONFIGURE_MAXIMUM_TASKS 1
> #define CONFIGURE_RTEMS_INIT_TASKS_TABLE
> #define CONFIGURE_INIT
> #include <rtems/confdefs.h>
> 
> <==
> [1/2] Compiling build/.conf_check_506c35a0aa2e9f571a9e7ed38df523b2/test.c
> 
> ['/build/rtems/5/bin/powerpc-rtems5-gcc', '-qrtems',
> '-B/build/rtems/5/powerpc-rtems5/lib/',
> '-B/build/rtems/5/powerpc-rtems5/qoriq_e500/lib/', '--specs', 'bsp_specs',
> '-mcpu=8540', '-mcpu=8540', '-meabi', '-meabi', '-msdata=sysv', '-msdata=sysv',
> '-fno-common', '-fno-common', '-mstrict-align', '-mstrict-align', '-mspe',
> '-mspe', '-mabi=spe', '-mabi=spe', '-mfloat-gprs=double', '-mfloat-gprs=double',
> '-ffunction-sections', '-ffunction-sections', '-fdata-sections',
> '-fdata-sections', '../test.c', '-c', '-o',
> '/scratch/git-rtems-libbsd/build/.conf_check_506c35a0aa2e9f571a9e7ed38df523b2/testbuild/test.c.1.o']
> 
> err: ../test.c:1:10: fatal error: rtems.h: No such file or directory
>  #include <rtems.h>
>           ^~~~~~~~~
> compilation terminated.

This a pretty good. The gcc command line has the valid BSP details.

Chris