[Freeipa-devel] Build system refactoring - design document

[Petr Spacek]

On 11.10.2016 10:04, Jan Cholasta wrote:
> On 11.10.2016 09:36, Petr Spacek wrote:
>> On 11.10.2016 09:00, Jan Cholasta wrote:
>>> Hi,
>>>
>>> On 7.10.2016 11:56, Petr Spacek wrote:
>>>> Dear FreeIPA developers and packagers,
>>>>
>>>> you can find first version of the Build system refactoring design document
>>>> on:
>>>> http://www.freeipa.org/page/V4/Build_system_refactoring
>>>>
>>>> If you do not care about implementation details, please be so kind and
>>>> quickly
>>>> scan through chapter
>>>> http://www.freeipa.org/page/V4/Build_system_refactoring#Feature_Management
>>>>
>>>> I'm not an FreeIPA packager so I might miss some important thing which needs
>>>> to be configurable.
>>>
>>> 1) There should be a --with-python switch to select the version of Python to
>>> use in our command line tools and/or during build. The default would be
>>> "python", i.e. the default Python interpreter found in the path.
>>
>> Okay. Can we pick some descriptive name?
>>
>> --with-default-python
>> or
>> --with-<something-else>-python?
>>
>> I think that it would be confusing if we had just
>> --with-python
>> --with-python2
>> --with-python3
> 
> If the default values are "python", "python2", "python3" respectively, I don't

These need to be full paths. I hope that some autoconf detection logic will
help with autodetection.


> think it would be confusing, since most of the time you only need to specify
> --with-python, if anything.

Okay, let me be explicit: It *is* confusing for me. Would you be okay with
--with-default-python ?


> Do we even need --with-python2 and --with-python3? I think they would only
> make sense if you had multiple Python minor versions installed and wanted to
> make packages for all of them.

AFAIK autoconf-style is to provide these for options where path to external
binary is needed. I would like to keep these conventions to avoid NIH syndrome
in the new build system.

Also, --without-python2/--without-python3 is needed anyway to disable this
part of build on systems without Python X version. I want to keep this
explicit (as with any other optional part of the build).


>> Besides that, I would make --with-default-python to accept either "2" or "3"
>> (and thus use path specified by --with-python? option).
>>
>>
>>
>>> 2) There is --with-pylint, --with-jslint, but no --with-po-validate.
>>
>> Let me clarify: I plan to use --with for things which have paths or other
>> parameters, --enable for booleans.
> 
> I see, that was not clear to me, I confused the two.
> 
>>
>> Where po-validate belongs? AFAIK target validate-po in install/po/Makefile is
>> calling script ../../ipatests/i18n.py which is in IPA source tree anyway.
>>
>> Do you want to have a --enable/--disable switch for these PO checks?
> 
> Not really.
> 
>>
>>
>>> 3) I would prefer that if pylint (or jslint or python-polib) is not installed
>>> the build would fail instead of silently skipping the lint. Let it be a wilful
>>> decision of the packager whether to run the lint or not.
>>
>> Yes, that is my intent. It will not skip anything automatically.
> 
> Right.
> 
>>
>>
>>
>>> 4) It is explicitly stated that I can turn off features using
>>> --without-feature. But how do I disable building server components?
>>
>> I've added explicit mention of --disable-feature:
>> http://www.freeipa.org/index.php?title=V4%2FBuild_system_refactoring&type=revision&diff=14311&oldid=14310
>>
> 
> Thanks.
> 
>>
>>
>>>> Also, I would appreciate ideas how to handle build versioning:
>>>> http://www.freeipa.org/page/V4/Build_system_refactoring#Versioning
>>>>
>>>> My main questions are:
>>>> * What is triggering IPA upgrade?
>>>> * Would it be sufficient to bump release in RPM? (I mean - theoretically.
>>>> Could the code be modified to detect this?)
>>>>
>>>> Here I'm trying to avoid unnecessary rebuilds caused by changes to
>>>> IPA_VENDOR_VERSION during each build.
>>>
>>> How exactly is IPA_VENDOR_VERSION causing unnecessary rebuilds? I can see it
>>> is written only to ipapython/version.py:
>>>
>>> $ git grep -E '\bIPA_VENDOR_VERSION\b'
>>> Makefile:IPA_VENDOR_VERSION=$(IPA_VERSION)$(IPA_VENDOR_VERSION_SUFFIX)
>>> Makefile:       sed -i -e "s:__VENDOR_VERSION__:$(IPA_VENDOR_VERSION):"
>>> ipapython/version.py
>>
>> My bad, I should write 'IPA*VERSION*'.
>>
>> Especially unconditional write to version.m4 is problematic but unconditional
>> writes to other files slows things as well, just not that much.
> 
> Could this be worked around by writing into a temporary file, comparing it
> with the real file and mv'ing the temporary file over the real file only if
> the differ?

Right now IPA_VERSION_IS_GIT_SNAPSHOT is always appends timestamp to the
string so the move would happen always ...

This is why I'm trying to understand where the versions are used and if they
really have to change at every (devel) build.

-- 
Petr^2 Spacek