Well, besides the big testing of compiling an entire linux distribution, yes, it is good to have specific tests for some particular issues in the compiler (manageable pass/fail tests that can identify a problem quickly)
Yeah...I got hit by a regression in a bug fix in RH's GCC back in the 4.1.2 era. The revision/build number only moved by a dozen or 2, but they completely broke C++ anonymous namespaces at global scope (caused an ICE). Glad I didn't let Ops upgrade all of my boxes at the same time like they wanted...