Hi Artyom, Artyom Poptsov <poptsov.art...@gmail.com> skribis:
> Currently one of the main showstoppers for the project is the > implementation of the tests: they're failing sometimes and sometimes get > stuck due to some kind of dead-lock or something. I'm tired of dragging > the current implementation with all its "quirks" from release to > release, so there's a burning need to fix it somehow or re-implement the > testing for the project altogether. So I guess I'm not going to make a > release untill the problem is fixed. > > Many problems with the tests are arising from the fact that Guile-SSH > needs multi-process tests (i.e. client-server tests) and SRFI-64 alone > that I use for tests seems to be a little bit insufficient on its own > for such tests, so I built some kind of a testing framework upon it. I think these kinds of tests are inherently hard to get right and to make reliable. These are more or less distributed computing issues: you need to distinguish between slow response and no response, and you might want to resort to some out-of-band synchronization mechanism (meaning that you don’t rely on Guile-SSH itself to synchronize, say, an SSH server and its client under test). So SRFI-64 is not always great, but I’m not sure it’s the real source of pain here. (Not that it really helps, though…) Ludo’.