Re: [PATCH v3 1/1] runchecks: Generalize make C={1,2} to support multiple checkers

[Knut Omang <knut.omang@oracle.com>]

On Sun, 2018-01-07 at 08:12 -0200, Mauro Carvalho Chehab wrote:
> Em Fri, 05 Jan 2018 20:41:41 +0100
> Knut Omang <knut.omang@oracle.com> escreveu:
> 
> > On Fri, 2018-01-05 at 16:08 -0200, Mauro Carvalho Chehab wrote:
> > > Em Thu, 04 Jan 2018 21:15:31 +0100
> > > Knut Omang <knut.omang@oracle.com> escreveu:
> > >   
> > > > > I'm surprised the commit message and the provided documentation say
> > > > > nothing about using CHECK=foo on the command line. That already supports
> > > > > arbitrary checkers.     
> > > > 
> > > > The problem, highlighted by Jim Davis in
> > > > 
> > > > https://lkml.org/lkml/2017/11/20/638
> > > > 
> > > > is that the current solution isn't flexible enough - that discussion 
> > > > is what lead me to this reimplementation of what I originally intended 
> > > > to be a checkpatch only solution.
> > > >   
> > > > > How does this relate to that? Is this supposed to be
> > > > > a complete replacement? Or what?    
> > > > 
> > > > It has evolved into a complete replacement of the intention of CHECK.
> > > >   
> > > > > 'make help' also references $CHECK, and this patch doesn't update the
> > > > > help text.    
> > > > 
> > > > I realize now that this needs to be handled in some way due to the way I split the 
> > > > arguments with '--' - the intention was to keep it for bw compatibility.
> > > > 
> > > > It would be good to know if people rely on using CHECK with C={1,2} for 
> > > > anything beside the checkers supported by runchecks today  
> > > 
> > > I do. Here, I use:
> > > 
> > > $ make ARCH=i386  CF=-D__CHECK_ENDIAN__ CONFIG_DEBUG_SECTION_MISMATCH=y C=1 W=1
> > > CHECK='compile_checks' M=drivers/media
> > > 
> > > Where "compile_checks" is actually a small script that calls both
> > > smatch and sparse:
> > > 
> > > 	#!/bin/bash
> > > 	/devel/smatch/smatch -p=kernel $@  
> > 
> > I suppose you here refer to this:
> > https://blogs.oracle.com/linuxkernel/smatch-static-analysis-tool-overview,-by-dan-carpenter
> > 
> > Good idea! I'll have a look at how that plays with this.
> 
> Yes.
> 
> > 
> > > 	/devel/sparse/sparse $@
> > > 
> > > So, I'm not sure why we need something else.   
> > 
> > The core functionality is the selective suppression logic and output unification
> > which makes checking with automated build tools more flexible and 
> > applicable right away (not when every warning from every checker is fixed...)
> 
> If the idea is to use it only/mostly with automated build tools, then
> the better would be to call it only when explicitly requested, e. g.
> something like C=3, in order to avoid breaking the usecase where one
> would run its own script.

Funny you should mention C=3 - I have an idea for that, but not what you suggest...

> On my case, I use C=1 CHECK=compile_checks as part as my usual patch
> handling. 

This is exactly what I implemented this for - I do this myself.

> For every patch I apply on media, I call make again, to be
> sure that no warning/building errors were added, not only with gcc
> but also with smatch and sparse. 

I humbly think this should fit your use case perfectly ;-)
Once build bots use this across the line, you might even save time
reviewing other people's smatch/sparse errors in your code, and also get the benefit of
errors detected by checkpatch - without having to fix all checkpatch check types right
away, you might also as a maintainer decide that some are not desirable to fix,
yet still be able to get the benefit of automation.

Just to illustrate, this is the result for az6007.o in -rc6:

total: 0 errors, 13 warnings, 20 checks, 991 lines checked

> > > That said, I didn't look
> > > on its code, but looking on its diffstat:

No problem,

Thanks,
Knut

> > > 
> > >  Makefile                               |  23 +-
> > >  scripts/Makefile.build                 |   4 +-
> > >  scripts/runchecks                      | 734 ++++++++++++++++++++++++++-
> > >  scripts/runchecks.cfg                  |  63 ++-
> > >  scripts/runchecks_help.txt             |  43 ++-
> > > 
> > > Using a 734 lines python program just to do an exec on an external checker
> > > seems too much!  
> > 
> > Sure, if that was the case I would be the first to agree :-)
> > 
> > Thanks,
> > Knut
> > 
> > > Thanks,
> > > Mauro  
> 
> 
> 
> Thanks,
> Mauro