Instead of reserving 0xF00 bytes for the fixed length portion of each CodeView symbol record (which is often much smaller) calculate the fixed portion of the record length to allow for the maximum string length possible. This prevents unnecessary truncation of the symbol name.
Although you'll probably never see names this long in hand-written code, it can occur in generated source code.
If we have the space we might as well let people use it if they want.
I updated the existing long-names.ll test to also check for long variable names.