Re: [Linux-kernel-mentees] [PATCH] checkpatch: Improve SPDX license identifier check for script files

[Lukas Bulwahn <lukas.bulwahn@gmail.com>]

On Sat, 22 Aug 2020, Lukas Bulwahn wrote:

> 
> 
> On Thu, 20 Aug 2020, Mrinal Pandey wrote:
> 
> > On 20/08/09 12:52PM, Mrinal Pandey wrote:
> > > On 20/08/04 09:37PM, Lukas Bulwahn wrote:
> > > > 
> > > > 
> > > > On Tue, 4 Aug 2020, Mrinal Pandey wrote:
> > > > 
> > > > > On 20/08/03 12:59PM, Lukas Bulwahn wrote:
> > > > > > 
> > > > > > 
> > > > > > On Mon, 3 Aug 2020, Mrinal Pandey wrote:
> > > > > > 
> > > > > > > The diff content includes the SPDX licensing information but excludes the
> > > > > > > shebang when a change is made to a script file in commit 37f8173dd849
> > > > > > > ("locking/atomics: Flip fallbacks and  instrumentation") and commit
> > > > > > > 075c8aa79d54 ("selftests: forwarding: tc_actions.sh: add matchall mirror
> > > > > > > test"). In these cases checkpatch issues a false positive warning:
> > > > > > > "Misplaced SPDX-License-Identifier tag - use line 1 instead".
> > > > > > > 
> > > > > > > Currently, if checkpatch finds a shebang in line 1, it expects the
> > > > > > > license identifier in line 2. However, this doesn't work when a shebang
> > > > > > > isn't found on the line 1.
> > > > > > 
> > > > > > It does not work when the diff does not contain line 1, but only line 2,
> > > > > > because then the shebang check for line 1 cannot work.
> > > > > > 
> > > > > > > 
> > > > > > > I noticed this false positive, while running checkpatch on the set of
> > > > > > > commits from v5.7 to v5.8-rc1 of the kernel, on the said commits.
> > > > > > > This false positive exists in checkpatch since commit a8da38a9cf0e
> > > > > > > ("checkpatch: add test for SPDX-License-Identifier on wrong line #")
> > > > > > > when the corresponding rule was first added.
> > > > > > > 
> > > > > > > The alternatives considered to improve this check were looking the file
> > > > > > > to be a script by either examining the file extension or file permissions.
> > > > > > >
> > > > > > 
> > > > > > Make this sentence shorter. Try.
> > > > > >  
> > > > > > > The evaluation on former option resulted in 120 files which had a shebang
> > > > > > > in the first line but no file extension. This didn't look like a promising
> > > > > > > result and hence I dropped the idea of using this approach.
> > > > > > > 
> > > > > > > The evaluation on the latter approach shows that there are 53 files in the
> > > > > > > kernel which have an executable bit set but don't have a shebang in the
> > > > > > > first line.
> > > > > > > 
> > > > > > > At the first sight on these 53 files, it seems that they either have a
> > > > > > > wrong file permission set or could be reasonably extended with a shebang
> > > > > > > and SPDX license information. Thus, further cleanup in the repository
> > > > > > > would make the latter approach to work even more precisely.
> > > > > > > 
> > > > > > > Hence, I chose to check the file permissions to determine if the file is a
> > > > > > > script and notify checkpatch to expect SPDX on second line for such files.
> > > > > > >
> > > > > > 
> > > > > > There is no notification here. Think about better wording.
> > > > > >  
> > > > > > > Signed-off-by: Mrinal Pandey <mrinalmni@gmail.com>
> > > > > > > ---
> > > > > > >  scripts/checkpatch.pl | 3 +++
> > > > > > >  1 file changed, 3 insertions(+)
> > > > > > > 
> > > > > > > diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> > > > > > > index 4c820607540b..bae1dd824518 100755
> > > > > > > --- a/scripts/checkpatch.pl
> > > > > > > +++ b/scripts/checkpatch.pl
> > > > > > > @@ -3166,6 +3166,9 @@ sub process {
> > > > > > >  		}
> > > > > > >  
> > > > > > >  # check for using SPDX license tag at beginning of files
> > > > > > > +		if ($line =~ /^index\ .*\.\..*\ .*[7531]\d{0,2}$/) {
> > > > > > > +			$checklicenseline = 2;
> > > > > > > +		}
> > > > > > 
> > > > > > That check looks good now.
> > > > > > 
> > > > > > >  		if ($realline == $checklicenseline) {
> > > > > > >  			if ($rawline =~ /^[ \+]\s*\#\!\s*\//) {
> > > > > > >  				$checklicenseline = 2;
> > > > > > 
> > > > > > This is probably broken now. It should check for shebang in line 1 and 
> > > > > > then set checklicenseline to line 2, right?
> > > > > 
> > > > > Sir,
> > > > > 
> > > > > Should we remove this check? Earlier when I checked for file extension
> > > > > we had 120 cases where this check was also needed but now we have a
> > > > > better heuristic which is going to work for all cases where license
> > > > > should be on line 2 irrespective of the fact that we know the first line
> > > > > or not.
> > > > >
> > > > 
> > > > Are you sure about that? Where is the evaluation that proves your point?
> > > > 
> > > > E.g., are all files that contain a shebang really with an executable flag?
> > > > 
> > > > Which commands did you run to check this?
> > > >  
> > > > > If I am missing out on something and we should not be removing this check,
> > > > > then I suggest placing the new heuristics below this block so that it doesn't
> > > > > interfere with the existing logic.
> > > > > 
> > > > > Please let me know which path should I go about and then I shall resend
> > > > > the patch with the modified commit message.
> > > > > 
> > > > 
> > > > Think about the strengths and weaknesses of the potential solutions, then 
> > > > show with some commands (as I did for example, for finding the first 
> > > > lines previously) that you can show that it practically makes a 
> > > > difference and you can numbers on those differences.
> > > > 
> > > > When you did that, send a new patch.
> > > > 
> > > > Lukas
> > > > 
> > > Sir,
> > > 
> > > I ran the evaluation as:
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ cat get_permissions.sh
> > > #!/bin/bash
> > > 
> > > for file in $(git ls-files)
> > > do
> > >         permissions="$(stat -c "%a %n" $file)"
> > >         echo "$permissions"
> > > done
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ sh get_permissions.sh | grep ^[7531] > temp
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ cut -d ' ' -f 2 temp > executables
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ cat first_line.sh
> > > #!/bin/bash
> > > file="executables"
> > > while IFS= read -r line
> > > do
> > >         firstline=`head -n 1 $line`
> > >         printf '%s:%s\n' "$firstline" "$line"
> > > done <"$file"
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ cat executables | wc -l
> > > 611
> > > 
> > > mrinalpandey@mrinalpandey:~/linux/linux$ sh first_line.sh | grep ^#! | wc -l
> > > head: error reading 'scripts/dtc/include-prefixes/arc': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/arm': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/arm64': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/c6x': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/dt-bindings': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/h8300': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/microblaze': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/mips': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/nios2': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/openrisc': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/powerpc': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/sh': Is a directory
> > > head: error reading 'scripts/dtc/include-prefixes/xtensa': Is a directory
> > > 540
> > > 
> > > We can see that there are 71 files where the executable bit is set but
> > > the first line is not a shebang. These include 13 directories which
> > > throw the error above. Remaining 58 files(earlier the number was 53)
> > > could be cleaned so that this heuristic works better as we saw. So, by
> > > checking only for the executable bit we can say that license should be
> > > on second line, we probably don't need to check for the shebang on line
> > > 1.
> > > Please let me know if the evaluation makes sense.
> > >
> 
> This evaluation makes sense to find the cases that should be cleaned up.
> 
> Either the executable flag is simply set wrongly and should be dropped or 
> it is actually a script and should get a shebang in the beginning.
> 
> I actually already started cleaning up. See:
> 
> https://lore.kernel.org/lkml/20200819081808.26796-1-lukas.bulwahn@gmail.com/
> 
> We can discuss how to continue this cleanup.
>

I had another look at the results of your script.

Just a minor improvement to that resulting list:

I think symbolic links in the repository are always of permission 777, and 
I think that is reasonable.

So maybe you can filter out the symbolic links in your get_permissions.sh?

Then, the list is probably down to a few 20 to 30 cases that should 
probably really be cleaned up.

Can you share that script and the results? Then, let us start cleaning 
up.

Lukas
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees