Hello, On Mon 15 Dec 2025 at 12:01am GMT, Ian Jackson wrote:
> We've just had a bug where a test failed in CI due to a race > (fixed in https://salsa.debian.org/dgit-team/dgit/-/merge_requests/353 ) > > I think this kind of bug is pretty mkuch inevitable with our > contraption of sockets and stuff to test oracled with a fake manager > (currently script which spawns shell scripts which act as the > manager). > > This arrangement has the virtue that it can test quite a lot of the > oracled's socket and child handling arrangments. But it is quite > complicated and makes normal testing of normal interactions with the > oracled harder than it needs to be. > > I think we should consider having a new test-specific invocation mode > for oracled which we would use for most of the tests. > A mode which > 1. Is entirely synchronous > 2. Doesn't fork worker children, but just runs one worker > 3. Instead of running nc.openbsd to connect to the manager, > allows execution of an arbitrary response script > (ie, we provide the the thing currently passed to t2u-fake-manager) > 4. Happily exits when it gets EOF from the "manager" > > LMK what you think. I don't have a problem with adding code to tag2upload-oracled so that it also has this operating mode. I'm not sure that it will make the test suite more robust, however, because it seems like we would want to keep all our current tests even if we add new ones that new the new mode. That's because we do actually want to test the aspects of the Oracle that the old tests test, right? -- Sean Whitton
signature.asc
Description: PGP signature
