Red Hat Bugzilla – Bug 20286
RPM has cryptic and non standard error messages
Last modified: 2007-04-18 12:29:40 EDT
this bug is similar to bug: 2895 but this one is MINE, and has more detail, and additional requests. First I would like the RPM error messages to be of a standard format and tell which executable (rpm) has triggered the error. My build process is quite complex with manny levels of scripts running other scripts and I can not tell where the errors originate if the executable does not identfy itself. I have spent many a time running strings over executables looking to see which binary could have given me a particular error message. It is also important that if RPM runs a program which fails (like a dependency program or a build script) that RPM should throw its own error message into the STDERR so that I can track down the error to RPM and its children. I may not know that 'find-requires' is a child of RPM so if I see an error from 'find-requires' I need to narrow this down to an RPM error. This is verry necessary for any automated build monitoring system, like tinderbox. The system needs to have regular expressions to determine what 'build errors' look like. Currently there is no regular expression which can capture what an RPM error looks like. Second (this is like an earlier bug) the RPM errors are too cryptic. A bigger effort should be made to narrow causes of failure down to something a programmer can understand and give meaningful messages. I often have to add a substatial ammount of checking code in my executables, beyond what is strictly necessary to complete the job, to ensure that all errors are meaningful to the user. Each program action needs to have guards to check before the action is taken that it is likely to succeed. All of the common error causes should be explicitly checked for and notifiy the user with a clear error message.
As a specific and supporting example of this bug (on RedHat 7.3/rpm-4.0.4): # ls nss_db.spec nss_db.spec # rpm -bp nss_db.spec nss_db.spec: No such file or directory ...when what really happens is: # strace rpm -bp nss_db.spec [...] execve("/usr/lib/rpm/rpmb", ["/usr/lib/rpm/rpmb", "-bp", "nss_db.spec"], [/* 20 vars */]) = -1 ENOENT (No such file or directory) write(2, "nss_db.spec: No such file or dir"..., 39nss_db.spec: No such file or directory ) = 39 _exit(1) = ? Not only is this error message terse, it is misleading, for it is not nss_db.spec that does not exist, but rather /usr/lib/rpm/rpmb. Therefore, at the very least, it should say: /usr/lib/rpm/rpmb: No such file or directory Now, I understand why this error occurred, and, judging by past responses to bugs in this component, I'll probably get flamed by someone at redhat.com for suggesting this, but might it not say something like: rpm: Fatal error: could not exec(/usr/lib/rpm/rpmb): No such file or directory rpm: Maybe you need to install rpm-devel? I'm no error-writing expert, so I'm sure the above could be polished up a bit, but I think that it's fairly reasonable to expect that a program: 1) Give an error that accurately represents what went wrong 2) Clearly identify itself in error messages 3) Clearly identify what it was trying to do when it encountered the error It also seems wise, for certain errors that occur in well-known conditions, to give a polite suggestion for a way that the user can fix the problem.
Problem #1: the error messages are much more regular now. Problem #2: the build modes have been moved from rpm into rpmbuild, and the popt aliases that provided backward compatibility (somewhat imperfectly) have been removed as well. Details through "man rpm", Legacy section.