automoc4 corner issues - what lerned from qt-creator port

[Helio Chissini de Castro]

<missing patch>
Attached is the new way better patch to handle cases 1, 2 and 4 correctly. 
3 still have need use the automoc4_moc_headers

An interesting case for 3 is generated headers from .ui files, 

[]'s

On Tuesday 24 of March 2009 15:36:19 Helio Chissini de Castro wrote:
> Hi
>
> Since i finished the port of qt-creator build, i'm trying to identify why
> automoc4 wasn't been able to build without patches adding moc_* in some
> codes but qmake can.
>
> DFaure helped a lot to identify the issue, and we have four corner cases to
> be treated:
>
> 1 - w.cpp have an internal Q_OBJECT class and includes w.h with other class
> with Q_OBJECT, and have on code #include "moc_w.cpp" and #include "w.moc"
>
> 2 - w.cpp have an internal Q_OBJECT class and includes w.h with other class
> with Q_OBJECT, but have only  "w.moc"
>
> 3 - w.cpp, independent of having internal Q_OBJECT class, includes a
> different header, y.h with a Q_OBJECT class, but without correspondent
> y.cpp code
>
> 4 - w.cpp have an internal Q_OBJECT class and includes w.h with other class
> with Q_OBJECT but have no moc includes.
>
> qmake can handle all cases, based on addition of headers and sources on
> .pro, so where and why automoc4 fails:
>
> * On case 1: Autmoc4 works 100%
>
> * On case 2: Automoc4 generates w.moc, but not generates moc_w.cpp. This
> results in a linker error when using --no-undefined.
> The case is that automoc4 detects the existency of a w.moc header and
> simply discards the fact that exists an Q_OBJECT inside the cpp, so was a
> logic question.
> As automoc4 handles the case of having no moc includes, i did a patch that
> test case 2 and solve case 2. Patch attached, don't know if is the best
> approach
>
> * On case 3: There's no way to do it automatically. Automoc4 can solve in
> CMakeLists using AUTOMOC4_MOC_HEADERS( <target> y.h )
> Is similar as qmake parsing headers, so in this case we still need add the
> headers that not match for all the cases. A *perfect world* solution would
> be automoc4 parse the local includes in source and verify if are standalone
> and have Q_OBJECT, but is not a trivial change and should touch too much in
> the current automoc4 structure. But is doable
>
> * On case 4: The right behavior is add only w.moc, and of course not test
> the .cpp, not generating moc_w.cpp falling in case 2. I'm expanding my
> patch to cover this one.
>
> For who want to help or test, i borrowed a dfaure's example an tweaked to
> be Qt only and have w.cpp with all moc includes commented. Is located in
> <kdesvnserver>/trunk/branches/work/~helio/
> The default status is failing on case 3 and 4 leading to case 2.
>
> []'s

-- 
Helio Chissini de Castro
KDE Developer
Brasil/South America Primary Contact

-------------- next part --------------
A non-text attachment was scrubbed...
Name: kde4automoc-handle-non-includes.patch
Type: text/x-patch
Size: 3260 bytes
Desc: not available
Url : http://mail.kde.org/pipermail/kde-buildsystem/attachments/20090324/4e0be367/attachment.patch