Re: [Ksummit-2013-discuss] [PATCH] checkpatch: Add comment about updating Documentation/CodingStyle

[Josh Triplett <josh@joshtriplett.org>]

On Tue, Sep 03, 2013 at 08:39:58AM +0800, Fengguang Wu wrote:
> On Mon, Sep 02, 2013 at 02:11:36PM -0700, Joe Perches wrote:
> > On Mon, 2013-09-02 at 21:50 +0100, David Howells wrote:
> > > Josh Triplett <josh@joshtriplett.org> wrote:
> > > 
> > > > > There are many checkpatch rules (like semicolons) that
> > > > > are not in CodingStyle.
> > > > 
> > > > It's a rule of thumb, not a mandate.  In *general*, checkpatch.pl should
> > > > not be enforcing style rules that aren't documented in CodingStyle.
> > > 
> > > Except that it becomes a mandate when someone runs it automatically against
> > > every one of your patches and then sends you an email for each patch it finds
> > > a checkpatch niggle against...
> > 
> > I think that any robot sending such checkpatch-only
> > emails should be disabled.
> > 
> > I know of 2 email robots.
> > 
> > Fengguang Wu's very useful build robot
> > sends out emails on build failures.
> > I think that's great.
> 
> Thanks! Yes I'm now running checkpatch these days because some people
> suggested to me that some of the checkpatch warnings do help catch
> real bugs.
> 
> However I do try to avoid upsetting people with maybe-subjective
> warnings. A checkpatch report will only be sent when a small fraction
> of error types are detected. Comments are very welcome on how to
> improve this list:
> 
> MEMSET
> IN_ATOMIC
> UAPI_INCLUDE
> MALFORMED_INCLUDE       
> SIZEOF_ADDRESS  
> KREALLOC_ARG_REUSE      
> EXECUTE_PERMISSIONS     
> ERROR:BAD_SIGN_OFF      
> LO_MACRO
> HI_MACRO
> CSYNC
> SSYNC
> HOTPLUG_SECTION
> INDENTED_LABEL
> INLINE_LOCATION
> STORAGE_CLASS
> USLEEP_RANGE
> UNNECESSARY_CASTS
> ALLOC_SIZEOF_STRUCT
> KREALLOC_ARG_REUSE
> USE_FUNC
> LOCKDEP
> EXPORTED_WORLD_WRITABLE
> WHITESPACE_AFTER_LINE_CONTINUATION
> MISSING_VMLINUX_SYMBOL
> NEEDLESS_IF
> PRINTF_L

Looks like you have KREALLOC_ARG_REUSE in that list twice.

Other than that, those look sensible.  I'd suggest a couple more, which
*should* always make sense, and to the best of my knowledge don't tend
to generate false positives:

C99_COMMENTS
CONFIG_EXPERIMENTAL
CVS_KEYWORD
ELSE_AFTER_BRACE
GLOBAL_INITIALIZERS
INITIALISED_STATIC
INVALID_UTF8
LINUX_VERSION_CODE
MISSING_EOF_NEWLINE
PREFER_SEQ_PUTS
PRINTK_WITHOUT_KERN_LEVEL
REDUNDANT_CODE
RETURN_PARENTHESES
SIZEOF_PARENTHESIS
SPACE_BEFORE_TAB
TRAILING_SEMICOLON
TRAILING_WHITESPACE
USE_DEVICE_INITCALL
USE_RELATIVE_PATH

These *ought* to make sense, but I don't know their false positive rates:

HEXADECIMAL_BOOLEAN_TEST
ALLOC_ARRAY_ARGS
CONSIDER_KSTRTO
CONST_STRUCT
SPLIT_STRING

The following almost always make sense, but only on patches not
yet applied to a tree:

PATCH_PREFIX
MODIFIED_INCLUDE_ASM
CORRUPTED_PATCH
NOT_UNIFIED_DIFF
MISSING_SIGN_OFF

- Josh Triplett