[132] in Kerberos_V5_Development
Re: error codes again
qjb@ATHENA.MIT.EDU (qjb@ATHENA.MIT.EDU)
Sun Jan 21 16:10:17 1990
Your right about that static string. I guess we should avoid
static variables period if we want a system that will run in a
multi-threading environment, as we have been saying all along.
That was a lapse of reasoning.... I agree that if we want a
string, it should be passed around in a structure.
I don't think you can convince me that the strings I am refering
to should be less than a run-time option. They should either be
availble to the clients or there should be a run-time option to
have the library print error messages. If this is left as a
compile time option and we have some application with no symbol
table, then we are going to be as stuck as we are now with
people not being able to tell the meaning of an error message.
The reason that a user of libkrb5.a might care where whether
space allocation problems were occuring is because if those
problems are the result of a coding error, the user will not
want to have to spend two days tracing the library sources in
saber to find out where the problem is.
The reason for having a string as part of an error code seems
obvious to me. As is the case with just about anything else,
the linear structure that numerical error codes gives us is not
adequate. If we don't allow some other dimension such as an
arbitrary string, we will either have a proliferation of
slightly different error codes that would be hard for
applications to check for or the current problem of error codes
that don't give enough information.
