Easier said than done. An interesting observation is the fact that everybody thought things were stable, and as far as any reports coming in, it seemed things were. However, once Remy declared 4.1.17 as stable, suddenly people actually started putting Catalina through it's paces(to see if they wanted to put it into production), and started finding problems. I think it was a good move, looking back.
I guess what I'm saying is that, unfortunately, the only way to be sure a release is really stable, is to convince many non-committers to stress-test it, and they only do that if they are told it is stable. Unless you want to come up with a stress / stability test procedure to be run against every build, that would be nice. - Jeff >>> [EMAIL PROTECTED] 12/19/02 7:06:21 PM >>> Remy, I checked the cvs and found there are some number of changes in the writer and response in coyote starting from 4.1.12. (See the diff from 4.1.12). Many of the changes are started at around 4.1.15/16. I think there is the problem in the roll out procedures, especially when a release is voted by the committer as stable and before the release manager declare it as GA. Stability test and code review should be thoroughly run. May be in 5.0, we could do better on this. - Punky P.S. I remember there was discussion in the release procedure or numbering guides. But I cannot find one in tomcat site. Is it just from http project? (http://httpd.apache.org/dev/release.html) Remy Maucherat wrote: > A bug exists (unfortunately) in Tomcat 4.1.16 and Tomcat 4.1.17 which > causes the servlet Writer to stay in an invalid state after an > IOException occurs (99% of the time caused by an abrupt client > disconnection). After this happens, the processor will never be able to > output data using the Writer, causing blank pages. This is more often > seen with JSPs. > > The bug affects Coyote HTTP/1.1, and may also affect Coyote JK 2, > although this is less likely. > > It is proposed that Tomcat 4.1.18, based on the Tomcat 4.1.17 code, with > the addition of the patch committed by Bill fixing JK 2 SSL support, as > well as the following patch (which I committed one hour ago): > -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
