On Fri, May 22, 2020 at 04:23:41PM +0100, Gabriele Cerami wrote: > 1) Code review is done on email. I suppose that may there be a bot that > interprets the tags added to commit messages and automates some tasks, > and maybe being independent from any organization is the main drive for > this, but have you discussed a switch to github PR or any other web-based > method like gerrit ? Assuming you have, can you point me to the thread > in the archive ?
It's been discussed a couple of times, but with no strong conclusion. The biggest issue with email is that mailing lists are hard to administer. Github PRs would be OK. I do not like Gerrit: https://benpfaff.org/writings/gerrit.html > 2) I'd like to understand better what git branching model are you using, > like e.g. long lived feature branches with occasional rebases and a > merge at the end ? Or maybe you prefer amended single patches ? The OVN branching model is documented, see Documentation/internals/release-process.rst. There is a lot of other information on contributions in and under that directory. > 3) I've noticed that ovn-northd.c is a single file with 12000 lines. Is > this caused by precise implementation choice, maintainability or > performance reasons ? Was ever proposed to modularize it ? Again, is > there a thread to look at ? It's just kind of grown over time. It's a rather simple program for its length--most of it is the moral equivalent of printf() calls--so this causes less trouble that one might guess. It could be easily split up into multiple pieces. I don't know whether that would be of great benefit. In the medium to long term I'd like to switch away from C for this program. String processing in C is silly. I've been prototyping a switch to the declarative programming language DDlog. The new version of the code that I've written for that is also more modular. > 4) I think I understand in tests the "unit" size seems to be set to a > command, so every test traverses all the functions in the execution path > that implement that command. The effect of this is that negative tests > also traverse all the functions before reaching an error. Have you ever > discussed using a unit test framework to shrink the unit to the size of > a function ? A lot of the tests do run only a single function. OVN has a number of test programs to facilitate that. I do not recall talking about switching test frameworks. > I ran some make checks and the feedback time for the component tests seem > comfortable. End to end obviously less. So I'm not sure it would provide > benefit, but "less than a full execution path" is the size that is > usually preferred for a "unit" If this is a naming thing, that is, you think the tests shouldn't be called "unit tests", I will personally not argue. If you want to propose more granular tests, that's great. I welcome more tests. _______________________________________________ discuss mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
