Re: [Qemu-devel] [RFC PATCH qemu] git-submodule.sh: Do not try writing to source directory if not necessary

[Alexey Kardashevskiy]

On 26/10/17 18:02, Daniel P. Berrange wrote:
> On Thu, Oct 26, 2017 at 11:54:59AM +1100, Alexey Kardashevskiy wrote:
>> On 26/10/17 08:11, Daniel P. Berrange wrote:
>>> On Wed, Oct 25, 2017 at 07:10:40PM +1100, Alexey Kardashevskiy wrote:
>>>> On 25/10/17 17:57, Daniel P. Berrange wrote:
>>>>> On Wed, Oct 25, 2017 at 12:45:10PM +1100, Alexey Kardashevskiy wrote:
>>>>>> On 25/10/17 03:27, Daniel P. Berrange wrote:
>>>>>>> On Tue, Oct 24, 2017 at 07:58:53PM +1100, Alexey Kardashevskiy wrote:
>>>>>>>> The new git-submodule.sh script writes .git-submodule-status to
>>>>>>>> the source directory every time no matter what. This makes it 
>>>>>>>> conditional.
>>>>>>>>
>>>>>>>> Signed-off-by: Alexey Kardashevskiy <address@hidden>
>>>>>>>> ---
>>>>>>>>
>>>>>>>> I compile out of tree on a remote guest system where I mount the
>>>>>>>> source directory as "readonly" and build directory as "rw" and
>>>>>>>> scripts/git-submodule.sh tries writing to the source directory even 
>>>>>>>> when
>>>>>>>> I manually update modules on a host machine which is quite annoying.
>>>>>>>>
>>>>>>>> Is this something acceptable? Or I am missing something here?
>>>>>>>
>>>>>>> How did you update the modules - did you manually run  'git submodule 
>>>>>>> update...'
>>>>>>> or did you use the git-submodule.sh script on your host machine ?
>>>>>>
>>>>>>
>>>>>> I run scripts/git-submodule.sh. Which is not thrilling either as I rather
>>>>>> expect source tree not to be affected in any way when running "make".
>>>>>
>>>>> Oh, did you pass the list of sub-modules to it when running
>>>>>
>>>>> eg, ./scripts/git-submodule.sh update ui/keycodemapdb
>>>>>
>>>>> the list of submodules you need is printed in the configure output 
>>>>> summary.
>>>>
>>>> Sure, otherwise it does nothing.
>>>>
>>>>
>>>>>
>>>>>>> If you run git-submodule.sh on the host, then it should save the status
>>>>>>> file, and then when you run make on the guest system, it should notice
>>>>>>> that you're already updated and never even invoke 'git-submodule.sh 
>>>>>>> update'
>>>>>>
>>>>>>
>>>>>> scripts/git-submodule.sh also tries writing to the source directory (I
>>>>>> should probably have fixed that branch too) but this failure is not fatal
>>>>>> for "make" but makes it want to try "update" and then "make" fails.
>>>>>
>>>>> This shouldn't have happened in your case though, if you have already run
>>>>> 'git-submodule.sh update ...list of modules...' on the host machine, with
>>>>> the same list of modules that the guest 'configure' printed out.
>>>>
>>>> It does not matter if I run git-submodule.sh or not - "git-submodule.sh
>>>> status" will try writing to the read only folder anyway and it will fail
>>>> and Makefile's  git_module_status will be set to 1.
>>>
>>>
>>> Ahhhh, great, now I understand why you're hitting the problem !
>>>
>>>> If I do as below (and that's what I should have done as I said), then
>>>> "git-submodule.sh update" is not invoked and we are good. I am not
>>>> reposting it yet as 1) my shell skills are crap (need to delete the temp
>>>> file or rewrite the whole thing not to use temp file or rewrite it in
>>>> python - why do not people use python everywhere?!) 2) I still hope we stop
>>>> doing this from Makefile :)
>>>
>>> I agree using a tmpfile is the right fix here.
>>
>>
>> I still think not doing "git update" from Makefile is the right fix here,
>> is that a final decision? Why cannot "configure" do this (and ideally have
>> a way not to do this at all, like --no-git-submodules-update)?  Just
>> checking...
> 
> That ends up doing exactly the same thing. We would need to make sure that
> configure gets re-run when the submodules change.


Submodules change once/twice a year so we could live quite happily without
this automation imho. What if I do not want submodules to be updated in
order to reproduce some bug and confirm that the newer version of submodule
has fixed it?

btw I looked now to my tree with upstream QEMU and dtc in that tree is 8
month old (v1.4.4 and there is v1.4.5 available). Is not it supposed to
update to something more recent automatically? git-submodule.sh does not do
checkout or pull.


> So you just go from having
> 'make' run git-submodule.sh, to have make run config.status which runs
> configure, which runs git-submodule.sh. 





-- 
Alexey