[Linux-kernel-mentees] checkpatch.pl investigation: MULTISTATEMENT_MACRO_USE_DO_WHILE issues

[Linux-kernel-mentees] checkpatch.pl investigation: MULTISTATEMENT_MACRO_USE_DO_WHILE issues

[Dwaipayan Ray]

Hi,
Checkpatch seems to generate some false positives on
certain macros.

For example running checkpatch on (kernel/trace/trace_export.c),

ERROR: Macros with multiple statements should be enclosed in a do - while loop
#49: FILE: kernel/trace/trace_export.c:49:
+#define __field_packed(type, container, item) type item;

ERROR: Macros with multiple statements should be enclosed in a do - while loop
#52: FILE: kernel/trace/trace_export.c:52:
+#define __array(type, item, size) type item[size];

and several other same errors.

Wrapping this in a do - while certainly doesn't make sense.
Removing the semicolon at the end of macro and appending semicolon
in the call changes the error to:

ERROR: Macros with complex values should be enclosed in parentheses
#49: FILE: kernel/trace/trace_export.c:49:
+#define __array_desc(type, container, item, size) type item[size]

This seems more reasonable. But the error isn't necessary in the first
place.

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

Re: [Linux-kernel-mentees] checkpatch.pl investigation: MULTISTATEMENT_MACRO_USE_DO_WHILE issues

[Lukas Bulwahn]

On Mon, 28 Sep 2020, Dwaipayan Ray wrote:

> Hi,
> Checkpatch seems to generate some false positives on
> certain macros.
> 
> For example running checkpatch on (kernel/trace/trace_export.c),
> 
> ERROR: Macros with multiple statements should be enclosed in a do - while loop
> #49: FILE: kernel/trace/trace_export.c:49:
> +#define __field_packed(type, container, item) type item;
> 
> ERROR: Macros with multiple statements should be enclosed in a do - while loop
> #52: FILE: kernel/trace/trace_export.c:52:
> +#define __array(type, item, size) type item[size];
>

I guess checkpatch.pl detects the semicolon and thinks it must be two 
statements. How about checking if the semicolon is actually followed by 
anything or not?

> and several other same errors.
>

Are they all in one file or spread around in the kernel everywhere?
 
> Wrapping this in a do - while certainly doesn't make sense.
> Removing the semicolon at the end of macro and appending semicolon
> in the call changes the error to:
> 
> ERROR: Macros with complex values should be enclosed in parentheses
> #49: FILE: kernel/trace/trace_export.c:49:
> +#define __array_desc(type, container, item, size) type item[size]
> 
> This seems more reasonable. But the error isn't necessary in the first
> place.
>

Considering format or semantic checks, I would always check if 
the better check can be implemented with clang-format, clang-analyzer or 
coccinelle.

checkpatch.pl is always just a quick heuristics; if other checking tools 
become part of standard check, then we could actually just refer to those 
checking tools instead.


Lukas

> Thanks,
> Dwaipayan.
> 
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

Re: [Linux-kernel-mentees] checkpatch.pl investigation: MULTISTATEMENT_MACRO_USE_DO_WHILE issues

[Dwaipayan Ray]

On Mon, Sep 28, 2020 at 9:44 PM Lukas Bulwahn <lukas.bulwahn@gmail.com> wrote:
>
>
>
> On Mon, 28 Sep 2020, Dwaipayan Ray wrote:
>
> > Hi,
> > Checkpatch seems to generate some false positives on
> > certain macros.
> >
> > For example running checkpatch on (kernel/trace/trace_export.c),
> >
> > ERROR: Macros with multiple statements should be enclosed in a do - while loop
> > #49: FILE: kernel/trace/trace_export.c:49:
> > +#define __field_packed(type, container, item) type item;
> >
> > ERROR: Macros with multiple statements should be enclosed in a do - while loop
> > #52: FILE: kernel/trace/trace_export.c:52:
> > +#define __array(type, item, size) type item[size];
> >
>
> I guess checkpatch.pl detects the semicolon and thinks it must be two
> statements. How about checking if the semicolon is actually followed by
> anything or not?
>
> > and several other same errors.
> >
>
> Are they all in one file or spread around in the kernel everywhere?
>

Yes, there were all in the same file. I have to run checkpatch on other
files extensively before I can confirm these issues exist elsewhere.

But yes, checking for a semicolon followed by some characters
does solve the problem, and changes the warning to a (Macros
with complex values...).

I do agree that such complex checks should be left to clang-format or
likewise.

Not sure if it is worthwhile merging, but here goes the diff:

---
diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 9e65d21456f1..8382977e95fd 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl

@@ -5351,7 +5367,7 @@ sub process {
                                if ($dstat =~ /^\s*if\b/) {

ERROR("MULTISTATEMENT_MACRO_USE_DO_WHILE",
                                              "Macros starting with if
should be enclosed by a do - while loop to avoid possible if/else
logic defects\n" . "$herectx");
-                               } elsif ($dstat =~ /;/) {
+                               } elsif ($dstat =~ /;.+/) {

ERROR("MULTISTATEMENT_MACRO_USE_DO_WHILE",
                                              "Macros with multiple
statements should be enclosed in a do - while loop\n" . "$herectx");
                                } else {
---
Thanks,
Dwaipayan.
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

Re: [Linux-kernel-mentees] checkpatch.pl investigation: MULTISTATEMENT_MACRO_USE_DO_WHILE issues

[Lukas Bulwahn]

On Mon, 28 Sep 2020, Dwaipayan Ray wrote:

> On Mon, Sep 28, 2020 at 9:44 PM Lukas Bulwahn <lukas.bulwahn@gmail.com> wrote:
> >
> >
> >
> > On Mon, 28 Sep 2020, Dwaipayan Ray wrote:
> >
> > > Hi,
> > > Checkpatch seems to generate some false positives on
> > > certain macros.
> > >
> > > For example running checkpatch on (kernel/trace/trace_export.c),
> > >
> > > ERROR: Macros with multiple statements should be enclosed in a do - while loop
> > > #49: FILE: kernel/trace/trace_export.c:49:
> > > +#define __field_packed(type, container, item) type item;
> > >
> > > ERROR: Macros with multiple statements should be enclosed in a do - while loop
> > > #52: FILE: kernel/trace/trace_export.c:52:
> > > +#define __array(type, item, size) type item[size];
> > >
> >
> > I guess checkpatch.pl detects the semicolon and thinks it must be two
> > statements. How about checking if the semicolon is actually followed by
> > anything or not?
> >
> > > and several other same errors.
> > >
> >
> > Are they all in one file or spread around in the kernel everywhere?
> >
> 
> Yes, there were all in the same file. I have to run checkpatch on other
> files extensively before I can confirm these issues exist elsewhere.
>

Okay, maybe I can support here with some local computing power from my 
side.
 
> But yes, checking for a semicolon followed by some characters
> does solve the problem, and changes the warning to a (Macros
> with complex values...).
>

Yes, unfortunately, macros with complex values is still wrong here...
 
> I do agree that such complex checks should be left to clang-format or
> likewise.
> 
> Not sure if it is worthwhile merging, but here goes the diff:
> 
> ---
> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
> index 9e65d21456f1..8382977e95fd 100755
> --- a/scripts/checkpatch.pl
> +++ b/scripts/checkpatch.pl
> 
> @@ -5351,7 +5367,7 @@ sub process {
>                                 if ($dstat =~ /^\s*if\b/) {
> 
> ERROR("MULTISTATEMENT_MACRO_USE_DO_WHILE",
>                                               "Macros starting with if
> should be enclosed by a do - while loop to avoid possible if/else
> logic defects\n" . "$herectx");
> -                               } elsif ($dstat =~ /;/) {
> +                               } elsif ($dstat =~ /;.+/) {
> 
> ERROR("MULTISTATEMENT_MACRO_USE_DO_WHILE",
>                                               "Macros with multiple
> statements should be enclosed in a do - while loop\n" . "$herectx");
>                                 } else {
> ---

How about sending a first proper commit at least here for the mentees?

I guess if we can run the checkpatch.pl evaluation on that commit, we can 
see and argue if that is worth the inclusion. I get that it solves the 
issues above, but it is tricky to say if something unrelated now is much 
worse...

A proper evaluation will show.


Lukas

> Thanks,
> Dwaipayan.
> 
_______________________________________________
Linux-kernel-mentees mailing list
Linux-kernel-mentees@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees