Weekly Edition
Return to the Kernel page
| |||||||||||||
pr_unwanted()?
One of a kernel developer's best friends is the printk() function,
which works much like printf() in user-space programs. There
are some differences, though, including the existence of various levels of
logging. The convention used is a little funny, with the logging level
being a short string prepended to the format string. So a warning might be
printed this way:
printk(KERN_WARNING "Core meltdown imminent\n");
This form is not universally loved, though; some call it verbose, making it hard to make lines fit within 80 columns, and the severity string is easy to forget. As an alternative, the 2.6.28-rc5 kernel saw the addition of a set of pr_*() macros, written by Martin Schwidefsky, which are designed to make life a little easier. So, for example, the above warning could be rewritten as:
pr_warning("Software patents detected\n");
These macros sat in relative obscurity for a few development cycles until Joe Perches decided to switch a number of printk() statements in the core kernel. That led to an outburst from Peter Zijlstra and the eventual reverting of the change. Peter says:
I might be strange, but if I want to print something in C I write
print[fk]() and be done with it, there's no reason what so ever to
introduce fancy wankery for this. We try to stick to ANSI-C as
much as possible, we've got kalloc,kfree,strcmp,strnlen and all the
other 'regular' C bits, deviating from that serves no purpose but
seed confusion.
Chances are, there won't be any more such conversions near that part of the kernel. But the pr_*() macros won't be going away either. Their real purpose, perhaps, was best expressed by Arjan van de Ven: "pr_ is really just for 'I am a driver and want a single line message out in a standardized format'. Nothing wrong with that." (Log in to post comments)
pr_unwanted()? Posted Dec 24, 2009 8:53 UTC (Thu) by flewellyn (subscriber, #5047) [Link]
I think their real purpose was actually best expressed by Al Viro:
"leveraging the synergy between make-work job and commit count masturbation"
pr_unwanted()? Posted Dec 24, 2009 9:21 UTC (Thu) by liljencrantz (subscriber, #28458) [Link]
Less typing and more explicit code sounds like two very reasonable reasons to use a bit of code to me.
pr_unwanted()? Posted Dec 24, 2009 16:31 UTC (Thu) by grmd (subscriber, #4391) [Link]
The trouble is it's the thin end of the wedge; it's not any more specific,
and it means that developers needlessly have something extra to learn.
Having worked on project that layer macros upon macros it rapidly becomes a
pr_unwanted()? Posted Dec 24, 2009 19:14 UTC (Thu) by vonbrand (subscriber, #4458) [Link] "More explicit" as long as there are very few functions/macros around. If there are more than a dozen, you start wondering if what you think it does is right... and the kernel is already huge, with its own set of non-standard stuff and rules. Better leave them out.
|
|||||||||||||