The "Bazaar" style of project management, as described
Eric Raymond and typified by the Linux kernel development model, is
undoubtedly effective at producing quality software, at least in some
situations. It can also, however, be a harsh environment in which to
operate, as demonstrated by events in the kernel community over the 2.5
series, and especially over the last week.
Readers of the LWN.net weekly Kernel Page will have been following the
development of the IDE/ATA layer in the 2.5 series for some time. For the
rest, here is some quick background to provide context for the rest.
The IDE layer, of course, is the low-level code that handles the disk (and
CD) drives found on most Linux systems. This code operates under a number
of serious constraints. It must be fast - able to drive the hardware at
its maximum speed; the performance of a Linux system as a whole is highly
dependent on how fast its disks can go. It also must be absolutely
correct; users get grumpy when their data is lost or corrupted. And it
must deal with a wide variety of, um, "inexpensive" hardware that does not
always behave as the documentation and standards say it should. Hacking on
the IDE subsystem is not for the faint of heart.
In recent times the IDE maintainer has been Andre Hedrick. Andre has had
numerous communication problems with Linus (and others) which have made it
difficult for him to get patches into the kernel. It is also fashionable
in certain quarters to criticize the quality of Andre's code. But, it
should be said: Andre's IDE layer has proved, over time, to be rigidly
standards compliant and highly reliable.
Andre's inability to get patches into the kernel left a void in the 2.5
series, however. That void was filled by Marcin Dalecki, who started
posting his "IDE cleanup" patches back in February. The "cleanups" began
to look increasingly like a complete rework (and hostile takeover) of the
IDE code, and, with IDE 18, Marcin put
his name into the MAINTAINERS file.
Marcin's work has been controversial all along - especially after he
started removing features that people were using, and when the IDE layer
started breaking for some users. His approach was not subtle, and he
seemed untroubled by the concerns of the other Linux kernel hackers. After
Marcin, "Breakage is the price you have to pay for
Linus, for the most part, seemed to agree; he merged almost every patch
from Marcin through IDE 115, posted on
All this changed on August 16, when Linus, without fanfare, deleted the
entire 2.5 IDE subsystem and replaced it with the "foreport" of the 2.4 IDE
layer, done by Jens Axboe and others. The word from Linus is that Marcin
got tired of all the criticism and quit; Marcin, himself, has been silent
since then. It is telling, though, that Linus responded by simply deleting
and replacing the entire body of 2.5 IDE work, rather than trying to find
somebody who would continue that task. Either Linus came to agree with
other kernel hackers about the quality of the reworked IDE code, or he
concluded that nobody else would be willing to work with that code.
The end result is that six months worth of Marcin's work, in the form of
115 IDE patches, has just been dumped into the bit bucket.
And that is an example of the harsh side of participating in the
kernel bazaar. One can work for months, see that work apparently accepted,
then have it vanish in a moment. Linus has said numerous times that the
doesn't much care about the feelings of kernel hackers; he is far more
concerned about the quality of the code. This approach may well be part of
why Linus is a good manager for Linux development - in the end, the code
quality must remain high or the whole thing will collapse under its own
weight. But it also explains why kernel hackers occasionally get
frustrated and leave the kernel development community. The bazaar can be
fun and effective, but it's not always nice.
to post comments)