Am Mittwoch, den 14.03.2018, 16:44 +0100 schrieb Richard Leitner: > On 03/14/2018 04:27 PM, Oliver Neukum wrote: > > Am Mittwoch, den 14.03.2018, 14:31 +0100 schrieb Richard Leitner: > > > > > Well, but it does not. Removing a redundant definition is a clear > > benefit. But you are not removing a definition. You are introducing > > a preprocessor constant. Why? > > What is its benefit? > > AFAIK pci_ids.h collects PCI vendor and device IDs in one single > point. As the PCI vendor ID of Netlogic is used in multiple files > IMHO it would be a good idea to add it to pci_ids.h and furthermore > remove it from arch/mips/include/asm/netlogic/xlp-hal/iomap.h (where > it's currently defined). > > Or am I getting things wrong? I think so, yes. We are giving names to constants as a form of comment or to change them at multiple places at once and consistently. So #define XYZ_NETDEV_RESET_RETRIES 2 makes clearly sense. So does #define XYZ_MAGIC_VALUE1 0xab4e because it tells you that you have a magic value. But you will never redefine a PCI vendor ID. In fact you must not. And if you have a comparison like dev->vID == 0x1234 if you change this to dev->vID == SOME_VENDOR_ID what good does this to you? You already knew it was a vendor ID. Now you can name it at a glance. So what? If you have a device you will have to check whether you have some OEM version. You will always go and check the raw number. And if you have a log and need to check whether the check will be true, you will have a number. Using a constant there is nothing but trouble. Yet one more grep. Regards Oliver
