Re: checkpatch.pl false positive? "ERROR: do not use assignment inif condition"

[Tilman Schmidt]

Andy Whitcroft schrieb:
> On Tue, Oct 20, 2009 at 08:58:20AM -0700, Randy Dunlap wrote:
>> On Tue, 20 Oct 2009 17:50:45 +0200 Tilman Schmidt wrote:
>>
>>> The command
>>> ./scripts/checkpatch.pl -f drivers/isdn/gigaset/bas-gigaset.c 
>>> produces a lot of "ERROR" messages like these:
>>>
>>> ERROR: do not use assignment in if condition
>>> #608: FILE: isdn/gigaset/bas-gigaset.c:608:
>>> +       if ((ret = usb_submit_urb(ucs->urb_cmd_in, GFP_ATOMIC)) != 0) {
>>>
>>> ERROR: do not use assignment in if condition
>>> #745: FILE: isdn/gigaset/bas-gigaset.c:745:
>>> +               if ((ucs->rcvbuf = kmalloc(l, GFP_ATOMIC)) == NULL) {
>>>
>>> ERROR: do not use assignment in if condition
>>> #753: FILE: isdn/gigaset/bas-gigaset.c:753:
>>> +               if ((rc = atread_submit(cs, BAS_TIMEOUT)) < 0) {
>>>
>>> As far as I can see there's nothing wrong with these lines. In particular,
>>> I cannot find anything in Documentation/CodingStyle that would prohibit an
>>> assignment inside an 'if' condition.
>> Yes, we don't try to list Every Possible Problem in CodingStyle,

Well, that's not very nice towards small time developers like me.
I try to follow CodingStyle as best I can, only to get smacked with
a checkpatch ERROR for something that isn't even hinted at there.
If a way of coding is such an abomination that it merits an ERROR
from checkpatch (meaning " change this or see your submission
rejected"), then it also merits being mentioned in CodingStyle.

>> but emails on lkml over the past several years try to discourage such
>> assignments inside if's because they can be confusing, difficult to read,
>> and generally are not helping to produce any better code output than
>> using:
>> 	ret = usb_submit_urb(args);
>> 	if (ret) {
>> 		foo;
>> 		bar;
>> 		blah();
>> 	}

I don't agree, but judging from your wording you don't want to hear any
arguments, so I'll leave it at that. I just want to know *beforehand*
what I have to do in order to get my code merged.

> What he said :).  We try and codify the preferred style where there is
> a preference.  That preference was made by reviewers as it is much
> harder to accidentally do the following when you meant the assignment or
> visa versa:

But does a mere preference by some reviewers warrant an ERROR message,
when even a line that exceeds the 80 character limit, which is
elaborately justified in CodingStyle, produces only a warning?

Again, I don't want to start a discussion about that preference. I know
now that it exists, and I am in no position to question it. But how many
of those preferences may yet turn up to bite me? If people are feeling
so strongly about a preference that those who do not honor it must be
slapped with a checkpatch ERROR then they should also find the energy
to add an appropriate warning about that to CodingStyle so that coders
have a chance to avoid it. That's only fair.

Thanks,
Tilman

-- 
Tilman Schmidt                    E-Mail: tilman@xxxxxxx
Bonn, Germany
Diese Nachricht besteht zu 100% aus wiederverwerteten Bits.
Ungeöffnet mindestens haltbar bis: (siehe Rückseite)

Attachment: signature.asc
Description: OpenPGP digital signature