On Wed, 2023-11-29 at 09:45 +0100, Krzysztof Kozlowski wrote: > On 29/11/2023 09:35, Nuno Sá wrote: > > On Tue, 2023-11-28 at 10:03 -0800, Guenter Roeck wrote: > > > On 11/28/23 08:50, Krzysztof Kozlowski wrote: > > > > On 27/11/2023 17:03, Andy Shevchenko wrote: > > > > > On Mon, Nov 27, 2023 at 09:12:14AM +0100, Krzysztof Kozlowski wrote: > > > > > > On 27/11/2023 09:10, Krzysztof Kozlowski wrote: > > > > > > > > > > ... > > > > > > > > > > > Wait, this was not even unusual test, just standard compile, which means > > > > > > you did not do basic tests on your end. You must build your new driver > > > > > > with W=1, smatch, sparse and coccinelle before sending upstream. > > > > > > > > > > Well, sparse is lagging in development, for the last year it's at least two > > > > > times it broke kernel builds because of being not ready for the new stuff > > > > > used > > > > > in the kernel. Do we have anybody to sync this? I don't think so, hence > > > > > requiring this from developer is doubtful. Otherwise I agree, that basic > > > > > compilation with GCC/LLVM must be done. > > > > > > > > Sparse still detects several issues and handles lock annotations, so it > > > > is useful. But if you disagree with that part, I still insist on Smatch > > > > (which is actively developed and works great) and Coccinelle (also > > > > actively developed). > > > > > > > > > > Quite frankly, for my part I would be more than happy if people would read > > > and follow Documentation/hwmon/submitting-patches.rst. Most submitters don't > > > bother. That doesn't even mention building with W=1 (the much more optimistic > > > me who wrote that document several years ago thought that would be obvious), > > > much less running any source code analysis tools . Feel free to submit a patch > > > to strengthen the wording there. If you do that, it would have to be more > > > explicit > > > then "run smatch" or "run coccinelle" because hardly anyone would know how > > > to do that. > > > > > > > IMO, submitting patches to linux is already not the most straightforward thing in > > the > > True... > > > world. If we are now going to ask to run smatch, cocci, sparse and so on, we will > > scare even more developers from the community... I mean, the bots are also in > > place > > This is not related to Linux at all. When you develop any C or C++ code, > you run these tools. Upstream or downstream, does not matter. Why would > you not use automated, free and easy tools to detect errors in your > code? It's just a matter of professional approach to your code. > That's true but still are too many things to remember for every single change/driver one sends upstream. Yeah, I might just wrap b4 in a script to run more advanced checks on my patches before 'send'. > > to help with these kind of more advanced analysis, right? > > They do not come for free (someone is paying for them even if they are > for free to you) and they have delays in responses. > Yeah, but actually thanks to you, I discovered I can have my private branches covered by lkp (and I got the PR merged already) and I do not mind having 1/2 day delay for sending patches. So maybe that will help me avoid these kind of mistakes. - Nuno Sá >
