On 06.03.2020 22:54, Junio C Hamano wrote:
With this detailed explanation, I expected to see a test or two that demonstrates a breakage, but reading a stale value may not reproducibly give the same wrong result or crash the program, perhaps?
Let's put it this way: one of the tests hits the bug every single time, yet still the bug has gone unnoticed for years. So yes, it's not super reliable. I think I could make a test that crashes often enough, but the effort will probably not be justified. The problem here is rather apparent when a finger is pointed to it.
