Re: [PULL 00/10] Gitlab-CI, qtest, moxie removal and misc patches

[Thomas Huth <thuth@redhat.com>]

On 20/05/2021 04.40, Philippe Mathieu-Daudé wrote:
> On 5/14/21 12:25 PM, Thomas Huth wrote:
>> On 14/05/2021 12.22, Philippe Mathieu-Daudé wrote:
>>> On 5/9/21 6:05 PM, Thomas Huth wrote:
>>>> On 07/05/2021 14.41, Paolo Bonzini wrote:
>>>>> On 07/05/21 11:45, Thomas Huth wrote:
>>>>>>
>>>>>>
>>>>>> diff --git a/Makefile b/Makefile
>>>>>> index bcbbec71a1..3088502329 100644
>>>>>> --- a/Makefile
>>>>>> +++ b/Makefile
>>>>>> @@ -85,7 +85,8 @@ x := $(shell rm -rf meson-private meson-info
>>>>>> meson-logs)
>>>>>>     endif
>>>>>>
>>>>>>     # 1. ensure config-host.mak is up-to-date
>>>>>> -config-host.mak: $(SRC_PATH)/configure $(SRC_PATH)/pc-bios
>>>>>> $(SRC_PATH)/VERSION
>>>>>> +config-host.mak: $(SRC_PATH)/configure $(SRC_PATH)/pc-bios
>>>>>> $(SRC_PATH)/VERSION \
>>>>>> +                $(SRC_PATH)/default-configs/targets
>>>>>>            @echo config-host.mak is out-of-date, running configure
>>>>>>            @if test -f meson-private/coredata.dat; then \
>>>>>>              ./config.status --skip-meson; \
>>>>>>
>>>>>> I.e. re-run configure if somethings in default-configs/targets
>>>>>> changed.
>>>>>> Does that look sane?
>>>>>
>>>>> I am not sure if using a directory is reliable (it's pre-existing for
>>>>> pc-bios).  However you probably can use
>>>>>
>>>>> # currently in tests/Makefile.include, move it to toplevel Makefile
>>>>> TARGETS=$(patsubst libqemu-%.fa, %, $(filter libqemu-%.fa,
>>>>> $(ninja-targets)))
>>>>> config-host.mak: $(SRC_PATH)/configure
>>>>> $(TARGETS:%=default-configs/targets/%)
>>>>>
>>>>> And then if a file goes missing it will trigger the rebuild of
>>>>> config-host.mak.
>>>>
>>>> Sounds like an idea, too ... but I'm unsure whether it's doable due to
>>>> the order of the statements there... TARGETS gets populated from
>>>> ninja-targets, but ninja-targets gets set *after* the config-host.mak
>>>> block ... would it be safe to move the config-host.mak block around?
>>>
>>> Not sure I understood Paolo's suggestion, I tried:
>>>
>>> -- >8 --
>>> diff --git a/Makefile b/Makefile
>>> index bcbbec71a1c..3ef3622228b 100644
>>> --- a/Makefile
>>> +++ b/Makefile
>>> @@ -152,6 +152,10 @@ ifneq ($(filter $(ninja-targets),
>>> $(ninja-cmd-goals)),)
>>>           +$(quiet-@)$(if $(MAKE.nq),@:, $(NINJA) -d keepdepfile \
>>>              $(NINJAFLAGS) $(sort $(filter $(ninja-targets),
>>> $(ninja-cmd-goals))) | cat)
>>>    endif
>>> +
>>> +TARGETS=$(patsubst libqemu-%.fa, %, $(filter libqemu-%.fa,
>>> $(ninja-targets)))
>>> +config-host.mak: $(SRC_PATH)/configure
>>> $(TARGETS:%=default-configs/targets/%)
>>> +
>>>    endif
>>>
>>>    # Force configure to re-run if the API symbols are updated
>>> ---
>>>
>>> But still get (after git pull this morning):
>>>
>>> $ ninja qemu-system-ppc
>>> [0/1] Regenerating build files.
>>> ...
>>> meson.build:1303:2: ERROR: Failed to load
>>> default-configs/targets/moxie-softmmu.mak: [Errno 2] No such file or
>>> directory: 'default-configs/targets/moxie-softmmu.mak'
>>> A full log can be found at meson-logs/meson-log.txt
>>> FAILED: build.ninja
>>>
>>> I'm keeping the directory unmodified in case you want me to try
>>> other fixes.
>>
>> If you run ninja directly, you certainly won't fix this issue by
>> patching the "Makefile" ;-)
> 
> OK... Not sure how to detect it from ninja then.

Every time you pull, there could be changes to the "configure" script which 
affect the build ... I guess you also won't notice those when you're only 
compiling with ninja? Thus at least after pulling from upstream, I think you 
have to type "make" once at least.

  Thomas