On 22/09/2016 09:12, Markus Armbruster wrote:
> Paolo Bonzini <pbonz...@redhat.com> writes:
>
>> ---
>> scripts/checkpatch.pl | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
>> index dde3f5f..3afa19a 100755
>> --- a/scripts/checkpatch.pl
>> +++ b/scripts/checkpatch.pl
>> @@ -2407,7 +2407,7 @@ sub process {
>> # we have e.g. CONFIG_LINUX and CONFIG_WIN32 for common cases
>> # where they might be necessary.
>> if ($line =~ m@^.\s*\#\s*if.*\b__@) {
>> - ERROR("architecture specific defines should be
>> avoided\n" . $herecurr);
>> + WARN("architecture specific defines should be
>> avoided\n" . $herecurr);
>> }
>>
>> # Check that the storage class is at the beginning of a declaration
>
> git-grep finds almost 400 of them. We certainly want people to think
> twice (or thrice) before they add more. The question to discuss here is
> whether we want to force that thinking onto the list. If yes, keep
> ERROR. If no, downgrade to warn.
I actually count 450, but:
- about a 100 are in imported code (disas/libvixl,
include/standard-headers and linux-headers, disas)
- another 40-odd hits are __NR_* syscall numbers
- about 80 are in user-exec.c, block/raw-posix.c, util/oslib-posix.c,
util/qemu-openpty.c, util/qemu-thread-posix.c which is probably unavoidable
- another 30 are in tcg
So this already covers more than half the hits.
The patch is a bit of a heavy hammer, but I don't think it's an endemic
problem that warrants a complaint on the list. If we want to keep the
error, I think we should have:
- a symbol blacklist. For example __linux__ and _WIN32 can be trivially
replaced by CONFIG_LINUX and CONFIG_WIN32, and __GNUC__ is probably a
bad idea (but __clang__ not so much; clang defines __GNUC__ for an
absurdly old version).
- a file blacklist, for example I would not expect target-*/ and hw/
should not have __ symbols and in fact they hardly have any
and warn for everything else. Something for bite-sized tasks?
Paolo
