Date: Wed, 20 Oct 1999 14:58:16 -0700 (PDT) From: Linus Torvalds <torvalds@transmeta.com> To: Gerard Roudier <groudier@club-internet.fr> Subject: Re: PATCH 2.3.23 pre 2 compile fixes On Wed, 20 Oct 1999, Gerard Roudier wrote: > > Just donnot backout Donald's patches that seems broken and I bet you that > everything will be just fine, or at least not worse that other breakages > that sometimes occur during kernel development. What? It's not about "seems broken" It's about the issue that there are real and definite bugs, and if a lot of things changed there is no good way to find out exactly what change caused the problem - especially not with the problem popping up for people who do not necessarily know C (or the device) enough to make a informed judgement other than "version X works for me, version Y does not". The whole point of open source is to expose the development, and NOT have the mentality that "it will be fixed in the next release". There should be many small incremental releases, because whatever Donald or others say, especially with drivers you are often in the situation that you cannot from looking at the source see whether something is broken or not. So it needs to be released often, and TESTED often. Which implies that the test-drivers should be part of the standard development kernel, because if they aren't, they aren't going to get very wide testing. For example, what has happened multiple times is that the 1% for whom some particular old network driver does not work will try out Donalds new drivers, and what do you know? It works for them! And people think that that means that the new driver has to be much better than the old one, right? Wrong. The new driver is NOT necessarily better at all. Not only has it been tested by much fewer people, it has been tested by a SELF-SELECTED group of people. Which may mean that the new driver fails horribly for a lot of people where the old driver was fine - because the new driver effectively has ZERO testing for common hardware that worked fine with the old driver. This is not worth discussing further. Timely incremental changes are just so OBVIOUSLY better to anybody who has done any real maintenance that the argument is pointless. It's true in non-Linux settings too - why do you think commercial software companies have regression tests and a large testbed of different machines that are always active? If Donald doesn't do the nice incremental patches, then somebody else will end up doing them. But that also means that Donald loses the right to then complain about others doing the work that he somehow considers "his". Linus - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/