Making -Wmissing-include-dirs an error?

[Neal Gompa]

Hey all,

Recently, one of the folks working on packaging stuff in Fedora KDE
nearly missed an issue caused by GCC emitting a warning about missing
include dirs:

>  cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or
> directory [-Wmissing-include-dirs]
>  cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or directory
> [-Wmissing-include-dirs] 
I did manage to figure out this meant we needed an additional build
dependency (qt6-qtbase-private-devel, FYI), but it made me think if
there's a reason this shouldn't be an error.

If it's an error, then at least we can evaluate these things and
ensure we have the right build inputs...

What do y'all think?

-- 
真実はいつも一つ！/ Always, there's only one truth!

[Florian Weimer]

* Neal Gompa:

>  Hey all,
>
>  Recently, one of the folks working on packaging stuff in Fedora KDE
>  nearly missed an issue caused by GCC emitting a warning about missing
>  include dirs:
>
> > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or directory
> [-Wmissing-include-dirs]
> > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or directory
> [-Wmissing-include-dirs]
>
>  I did manage to figure out this meant we needed an additional build
>  dependency (qt6-qtbase-private-devel, FYI), but it made me think if
>  there's a reason this shouldn't be an error. 
Was there a subsequent build failure due to the missing headers?  Could
you share a link to the full build log?

Thanks,
Florian

[Ian McInerney]

On Tue, Oct 10, 2023 at 11:59 AM Neal Gompa <ngompa13(a)gmail.com&gt; wrote:

>  Hey all,
>
>  Recently, one of the folks working on packaging stuff in Fedora KDE
>  nearly missed an issue caused by GCC emitting a warning about missing
>  include dirs:
>
>  > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or
>  directory [-Wmissing-include-dirs]
>  > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or
>  directory [-Wmissing-include-dirs]
>
>  I did manage to figure out this meant we needed an additional build
>  dependency (qt6-qtbase-private-devel, FYI), but it made me think if
>  there's a reason this shouldn't be an error.
>
>  If it's an error, then at least we can evaluate these things and
>  ensure we have the right build inputs...
>  
Missing an include directory isn't necessarily the problem though, it is
the missing headers that aren't present when they are included that would
be - and that should trigger a build error for the missing file. What
advantage does failing on this warning provide that the failure on the
include file missing doesn't?

-Ian


>  What do y'all think?
>
>  --
>  真実はいつも一つ！/ Always, there's only one truth!
>  _______________________________________________
>  devel mailing list -- devel(a)lists.fedoraproject.org
>  To unsubscribe send an email to devel-leave(a)lists.fedoraproject.org
>  Fedora Code of Conduct:
>  https://docs.fedoraproject.org/en-US/project/code-of-conduct/
>  List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
>  List Archives:
>  https://lists.fedoraproject.org/archives/list/[e-mail]
>  Do not reply to spam, report it:
>  https://pagure.io/fedora-infrastructure/new_issue
>

[Jonathan Wakely]

On Tue, 10 Oct 2023 at 12:17, Ian McInerney via devel
<devel(a)lists.fedoraproject.org&gt; wrote:
>  On Tue, Oct 10, 2023 at 11:59 AM Neal Gompa <ngompa13(a)gmail.com&gt; wrote:
> >
> > Hey all,
> >
> > Recently, one of the folks working on packaging stuff in Fedora KDE
> > nearly missed an issue caused by GCC emitting a warning about missing
> > include dirs:
> >
> > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or directory
> [-Wmissing-include-dirs]
> > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or
> directory [-Wmissing-include-dirs]
> >
> > I did manage to figure out this meant we needed an additional build
> > dependency (qt6-qtbase-private-devel, FYI), but it made me think if
> > there's a reason this shouldn't be an error.
> >
> > If it's an error, then at least we can evaluate these things and
> > ensure we have the right build inputs...
>
>
>  Missing an include directory isn't necessarily the problem though, it is the missing
> headers that aren't present when they are included that would be - and that should
> trigger a build error for the missing file. What advantage does failing on this warning
> provide that the failure on the include file missing doesn't? 
Typically, yes, I'd expect a failure. But it's possible for code to do:

#if __has_include(<foo.h>)
# include <foo.h>
// use features in that header
#else
// roll your own inferior version
#endif

And so the code could compile either way, but with results that differ
depending on whether <foo.h> is found or not. The extra -I include dir
could be given to the compiler so that if <foo.h> is present in that
dir, it will be found and used. If it's not present, there's no harm.
I don't expect this to be a very widely used pattern though.

The above doesn't necessarily argue for or against making it an error.
There are arguments either way. Making it an error might give some
maintainers a heads up that something's missing, but for other
packages it would break a valid use case.

[Gary Buhrmaster]

On Mon, Nov 13, 2023 at 5:16 PM Jonathan Wakely <jwakely(a)redhat.com&gt; wrote:

>  Typically, yes, I'd expect a failure. But it's possible for code to do:
>
>  #if __has_include(<foo.h>)
>  # include <foo.h>
>  // use features in that header
>  #else
>  // roll your own inferior version
>  #endif 

And in the particular case of the Qt private headers
(where this started) I have seen this, or an equivalent
construct, used to not support a feature at all.

And yes, I do believe Qt places some things into
private headers that they should probably provide
a public API to access, but that is a different
issue.

In *theory* the application would report that
feature X is disabled or unavailable during
configuration and/or during the checks,
and the packager would be expected to
realize the implications, but for large apps
such messages might be missed or not
understood.

I am slightly in favor of making missing
headers an error as long as it is easy
enough to override that to at least provide
a few more guard rails.

[Florian Weimer]

* Jonathan Wakely:

> > Missing an include directory isn't necessarily the problem
> though, it
> > is the missing headers that aren't present when they are included
> > that would be - and that should trigger a build error for the missing
> > file. What advantage does failing on this warning provide that the
> > failure on the include file missing doesn't?
>
>  Typically, yes, I'd expect a failure. But it's possible for code to do:
>
>  #if __has_include(<foo.h>)
>  # include <foo.h>
>  // use features in that header
>  #else
>  // roll your own inferior version
>  #endif 
I can see this might be a problem.

I wouldn't object if someone submitted a change proposal for this, but
they have to do the necessary work (full distribution rebuild to assess
the impact of the change, preferably with an instrumented/wrapped
toolchain to catch silently miscompiled autoconf probes).

Thanks,
Florian

[Daniel P. Berrangé]

On Tue, Oct 10, 2023 at 12:16:55PM +0100, Ian McInerney via devel wrote:
>  On Tue, Oct 10, 2023 at 11:59 AM Neal Gompa
> <ngompa13(a)gmail.com&gt; wrote:
>  
>  > Hey all,
>  >
>  > Recently, one of the folks working on packaging stuff in Fedora KDE
>  > nearly missed an issue caused by GCC emitting a warning about missing
>  > include dirs:
>  >
>  > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or
>  > directory [-Wmissing-include-dirs]
>  > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or
>  > directory [-Wmissing-include-dirs]
>  >
>  > I did manage to figure out this meant we needed an additional build
>  > dependency (qt6-qtbase-private-devel, FYI), but it made me think if
>  > there's a reason this shouldn't be an error.
>  >
>  > If it's an error, then at least we can evaluate these things and
>  > ensure we have the right build inputs...
>  >
>  
>  Missing an include directory isn't necessarily the problem though, it is
>  the missing headers that aren't present when they are included that would
>  be - and that should trigger a build error for the missing file. What
>  advantage does failing on this warning provide that the failure on the
>  include file missing doesn't? 
The header file may be present in another directory, but as an older
version that should not be used, resulting in building the wrong
featureset.

Last week I uncovered a bug in opensuse 15.4 where they had
patched libtirpc to move its headers directly into /usr/include,
but not correctly changed the pkg-config file which still pointed
to the non-existant /usr/include/libtirpc. 

This case was harmless, as /usr/include is on the default search
path, but it was none the less a genuine bug, and was caught due
our code using -Werror=missing-include-dirs in CI.

The reason for missing-include-dirs to not to be a build error by
default though, is that I suspect there are plenty of apps which
have sloppy build system setup that results in bogus dirs being
added.

With regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|

[Richard W.M. Jones]

On Tue, Oct 10, 2023 at 06:57:40AM -0400, Neal Gompa wrote:
>  Hey all,
>  
>  Recently, one of the folks working on packaging stuff in Fedora KDE
>  nearly missed an issue caused by GCC emitting a warning about missing
>  include dirs:
>  
>  > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or directory
> [-Wmissing-include-dirs]
>  > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or directory
> [-Wmissing-include-dirs]
>  
>  I did manage to figure out this meant we needed an additional build
>  dependency (qt6-qtbase-private-devel, FYI), but it made me think if
>  there's a reason this shouldn't be an error.
>  
>  If it's an error, then at least we can evaluate these things and
>  ensure we have the right build inputs...
>  
>  What do y'all think? 
I might be missing the point, but what issue did this cause?

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
virt-builder quickly builds VMs from scratch
http://libguestfs.org/virt-builder.1.html

[Neal Gompa]

On Tue, Oct 10, 2023 at 11:01 AM Richard W.M. Jones <rjones(a)redhat.com&gt; wrote:
>  On Tue, Oct 10, 2023 at 06:57:40AM -0400, Neal Gompa wrote:
>  > Hey all,
>  >
>  > Recently, one of the folks working on packaging stuff in Fedora KDE
>  > nearly missed an issue caused by GCC emitting a warning about missing
>  > include dirs:
>  >
>  > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3: No such file or directory
> [-Wmissing-include-dirs]
>  > > cc1plus: warning: /usr/include/qt6/QtCore/6.5.3/QtCore: No such file or
> directory [-Wmissing-include-dirs]
>  >
>  > I did manage to figure out this meant we needed an additional build
>  > dependency (qt6-qtbase-private-devel, FYI), but it made me think if
>  > there's a reason this shouldn't be an error.
>  >
>  > If it's an error, then at least we can evaluate these things and
>  > ensure we have the right build inputs...
>  >
>  > What do y'all think?
>
>  I might be missing the point, but what issue did this cause?
>  
I don't know if it did or didn't. For all I know, it could have led to
a miscompilation. It was fixed before we landed it as it was caught
while working on packaging.



-- 
真実はいつも一つ！/ Always, there's only one truth!