16 messages in net.sourceforge.lists.courier-users[courier-users] Re: Erroneous build d...| From | Sent On | Attachments |
|---|---|---|
| Robin Bowes | Oct 28, 2002 6:01 pm | |
| Sam Varshavchik | Oct 28, 2002 7:32 pm | |
| Robin Bowes | Oct 29, 2002 2:15 am | |
| Juri Haberland | Oct 29, 2002 2:25 am | |
| Robin Bowes | Oct 29, 2002 4:43 am | |
| Sam Varshavchik | Oct 29, 2002 4:43 am | |
| Juri Haberland | Oct 29, 2002 4:59 am | |
| Robin Bowes | Oct 29, 2002 6:22 am | |
| Robin Bowes | Oct 29, 2002 6:59 am | |
| Sam Varshavchik | Oct 29, 2002 2:56 pm | |
| Robin Bowes | Oct 30, 2002 7:01 am | |
| Juha Saarinen | Oct 30, 2002 12:13 pm | |
| Robin Bowes | Oct 30, 2002 3:06 pm | |
| Jesse Keating | Oct 30, 2002 3:17 pm | |
| Robin Bowes | Oct 31, 2002 9:36 am | |
| Jesse Keating | Oct 31, 2002 10:56 am |
| Subject: | [courier-users] Re: Erroneous build dependencies in courier-imap-1.6.0-20021025 | |
|---|---|---|
| From: | Sam Varshavchik (mrs...@courier-mta.com) | |
| Date: | Oct 29, 2002 2:56:46 pm | |
| List: | net.sourceforge.lists.courier-users | |
Robin Bowes writes:
libstdc++.so.5 installed (from libstdc++-3.2-0.1) and I managed to install fam using the "--nodeps" option.
Congratulations -- you have a completely broken server.
Hmmm. What leads you to that conclusion? Ok, so it's not going to win any prizes for the cleanest package set but "completely broken"? I think not.
The package dependencies are not picked at random. --nodeps isn't an option that automatically waves a magic wand and makes everything all right. It's there for cases where whoever built the package screwed up, and did not specify the correct dependencies.
If you install a package that explicitly lists a dependency on a particular version of the standard C runtime library, and you force-install it, you now have a binary API incompatibility with the runtime C library. If you're lucky, the package will simply not work, and refuse to run right off the bat. But that's rare. The typical result are random, subtle faults; such as daemon processes crashing at random times, or locking up completely. A typical ABI change involves an extra, or a different argument to some library call. Now, the library will attempt to read the argument that's not provided by the userapp, and end up reading random garbage, with completely unpredictable results.
Anyone who uses --nodeps on a production server needs to have his head examined.
All of this is getting away from the thrust of my original point which is that courier-imap does not *need* FAM - it is an option that adds functionality, but is not essential.
It doesn't, but that has nothing to do with what's listed in the spec file. The spec file provides instructions for building the package on a released distribution. Since all recent versions of Red Hat have FAM, it is listed as a dependency, since having it installed results in enhanced functionality at run time.