All, I apologize for taking so long to respond to this bug report. I think it's not likely that this is actually an ICU problem, though perhaps a newer version of ICU with more static data means that a change to dovecot's configuration is required. I have questions to both of you below and some explanation of what I think might be happening. This is just a guess though as I haven't actually reproduced the problem myself.
Fabio, Can you explain how your settings came to be what they are? Did you change default values for these parameters, or did the software fail to work with the defaults in the debian package? You reported this: > I've switched back to dovecot Version 1.1.12-1 and it also NOT worked in > the first try. So I changed some settings and tried around with the > /etc/dovecot/dovecot.conf and the result was, that if I got under an > special limit with two settings the error occured again and when I rised > these settings to an higher value it works: > > login_process_size = 64 # SUGGESTED > login_process_size = 16 # WORKS > login_process_size = 8 # DIES WITH ERROR > > auth_process_size = 256 # SUGGESTED > auth_process_size = 32 # WORKS > auth_process_size = 16 # DIES WITH ERROR When you say a value is "SUGGESTED", where is it suggested? Were the "DIES WITH ERROR" values defaults that you had to change, or were the suggested values defaults that you were trying to lower? Did you come up with the "WORKS" values just by experimentation or in some other way? Steffen, > * 2009-04-02 10:19, Steffen Ille wrote: >> I've switched back to dovecot Version 1.1.12-1 and it also NOT worked in >> the first try. So I changed some settings and tried around with the >> /etc/dovecot/dovecot.conf and the result was, that if I got under an >> special limit with two settings the error occured again and when I rised >> these settings to an higher value it works: > > According to this sentence, I think this bug is related to libicu40 and not > to dovecot, this I'm reassigning the bug report to that package. > > Best regards. Could you please explain why you believe this is an ICU problem and not a dovecot problem? I am not familiar with dovecot or with the particular settings that the reporter is changing. From my uneducated perspective, it appears that setting certain thresholds too low causes a memory-related error, but I don't see how it could actually be a problem in the ICU library since that library hasn't actually been loaded yet. I'd have to understand what those parameters mean to really know whether failure loading a shared library because of a memory problem is an expected consequence of making them too low. Perhaps there's a setrlimit call being made in response to these values, which would suggest that things are behaving as expected. In that case, if the software works with the default settings, there's probably not actually bug, or if there is, it's a documentation problem. If the problem occurs with the default values, then the default values probably need to be increased. Perhaps this came about in connection with an upgrade to ICU just because a newer ICU data library has a larger amount of static data and thus requires more mappable memory space for the application that loads it. I will assign this back to dovecot-common if there is no further discussion, but please don't take this as my just trying to pass it back and avoid the issue. If it does turn out to be a problem with ICU and not with the dovecot configuration as I suspect, then I will certainly pursue it as required. I'll wait a little while for further discussion before I reassign the bug. If you decide it's really a dovecot issue, please feel free to assign then bug back to dovecot-common or wherever you believe is appropriate. -- Jay Berkenbilt <q...@debian.org> -- To UNSUBSCRIBE, email to debian-bugs-dist-requ...@lists.debian.org with a subject of "unsubscribe". Trouble? Contact listmas...@lists.debian.org
