[YBA] kernel compile errors with GCC >= 4.6

[Oleg Goldshmidt]

On Thu, Apr 5, 2012 at 10:57 AM, Nadav Har'El <nyh at math.technion.ac.il>wrote:

> On Wed, Apr 04, 2012, Oleg Goldshmidt wrote about "Re: [YBA] kernel
> compile errors with GCC >= 4.6":
> > On Wed, Apr 4, 2012 at 10:03 AM, Jonathan Ben Avraham <yba at tkos.co.il
> >wrote:
> >
> > > Dear linux-il colleagues,
> > > GCC 4.6 introduced many new warnings that cause -Werror to stop the
> > > compilation for some platforms, such as powerpc, in various files.
> >
> > Oh, I didn't even know they finally introduced -Werror... Good.
>
> Why is this good?
>

Well, one reason is that there have been calls from many directions to make
this happen. ;-) Now to why there were such calls?...

I don't really think I need to explain - least of all to you - why -Werror
is good. We all know that it does not let the developer to ignore warnings.
All warnings indicate problems that are waiting to happen once explicit or
implicit assumptions that the problem will not occur are broken. It is very
difficult for me to come up with an example where there is NO chance that
there may be a problem (even a seemingly innocuous "initialized but not
used" may appear because someone inadvertently removed one line more than
needed, started a compilation of a large and complicated product, and went
to grab a coffee - -Werror would force another look). On the other hand,
real life examples where paying attention to warnings saves the day
abound.[1]

Now let's discuss relevant use-cases. Say you lead an open source project -
not necessarily a Linux kernel, but something that has lots of contributors
whom you wish to trust to some extent, but you won't have the bandwidth to
check everything. Forcing everybody to use -Werror seems a very good idea
in this case.

Oh, you say, but what if a new version of the compiler reacts to more stuff
than the latest on your computer? So, someone will notice that first.
Typically on FOSS (e.g., kernel), it will be a developer who will either
patch or inform the maintainer.

Oh, but what if you deliver your code in source form to end users and they
- not developers or maintainers - need to compile it? And one of them
happens to use Fedora 23 rc4 beta with gcc-5.6.7? Well, you did provide a
Makefile that allows the use to do "make all WERROR=no"[2], right? And you
wrote it in the README, I am sure. Or, better yet, the procedure is
"./configure && make && sudo make install" and your ./configure script has
a --without-werror option, right?

Obviously, the kernel has such a facility that will help anyone who tries
to compile the kernel. Most likely an experienced person will encounter it
first, and if he is baffled at first then Linux-IL or equivalent will
render assistance. ;-) And we all hope YBA will send a patch...

[1] I am helping to evaluate a sizeable body of C++ code - a product -
these days. Don't let me start... The developers didn't use any warnings,
or maybe they originally did, saw lots of them, decided to ignore them,
switched everything off... My favourite of the last 2-3 days favourite was
a warning, in a few places, of non-void functions not returning anything.
Often this means that the return value is not checked for validity, etc.,
but this case was more interesting. Turned out the warnings were for output
operators for some user-defined classes that did not return a reference to
the stream. It so happened that everything was fine because these operators
were never used in an output chain (of the "cout << x << y;" type) - in the
current code. We would buy the code and modify it or just write client
applications or whatever. Just a bunch of -W options (I use a fair amount)
would just flush a huge number of warnings, some of which would be harmless
(under reasonable, but not all, conditions - there is always an assumption
involved, remember). If this were encountered at runtime it might take some
time to find.

[2] One kinda cute (though not necessarily the best) way to do it - only
with GNU make - is something like

override CFLAGS := -Wall -Wextra <and many more -W flags> -Werror $(CFLAGS)

which allows one to say "make CGLAGS+=-Wno-error" (or "make
CFLAGS+=-Wno-error=<x>") on the command line. This is the only way I know
of to *append* to a make variable via the command line ;-)

-- 
Oleg Goldshmidt | pub at goldshmidt.org <oleg at goldshmidt.org>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.cs.huji.ac.il/pipermail/linux-il/attachments/20120405/cace6ccb/attachment.html>