Closed TheBizzle closed 6 years ago
👏 👏 👏 This looks awesome! I gave it a quick look-through and had a few comments, but I haven't (and probably won't) do a "full" review. Way to go Jason! 🎉
Thanks for dealing with my spaghetti code and stupid questions! I hope something useful came out of all my work, even if most of the implementation doesn't survive. If I had to do GSoC again with what I know now, I'd definitely do some things differently. I'm really happy to see this implemented and hope to keep pushing the people I work with towards using NLW.
No worries, Camden. I can't say that I even recall any stupid questions, and the early version of export-world
that you put together definitely helped me to see the big picture more clearly and steer the design towards what it has become. Writing a desktop-compatible export-world
is challenging in a lot of ways, and it had to be implemented in order to suss out a lot of those challenges.
Ultimately, the real treasure was the code we threw out along the way.™
This all looks great to me. I'm especially happy about tests for import/export across all our test models!
Since the tests are superseding the old model tests, I ran a quick comparison between how they behaved before when something was different between headless and Tortoise and how they behave now and the difference was negligible; they're still totally fine to use for tracking down a difference between the systems, which is what matters to me.
Umm... yeah... lots of stuff here.
What seems to me to be the most controversial thing—something that any developer on this project should be aware of—is that I have drastically changed
TestModels
, which has traditionally been the workhorse of our test suite. Now, it is even more of a workhorse, and hopefully not in a way that's seen as problematic.Basically, instead of just running the models, it now does
import-world
/export-world
roundtripping sorcery, as well. Roundtrippingexport-world
andimport-world
won't catch too many bugs without running full models—but caught a ton with them!—and the most sensible way to roundtrip on full models seems to me to be leveraging the info thatTestModels
provides. I originally had that as a separate test suite, but it didn't make sense to me to have the model tests basically running twice (once with exporting and once without), since those tests already take forever when run just once. So now I've just mashed them together.Essentially,
TestModels
will take even longer now, and it's not easy to disentangle the "pure" model-running from the roundtripping stuff, but I think that this is worth it, in that it provides very thorough testing for a major feature. Others might disagree, though.Also probably worthy of some defense: There's a "TODO" item in there for adding the model name to the export. In order to get that, though, I need the compiler to provide me with the model name. That is being added in NetLogo/NetLogo#1547. Once that's merged, then I can add the model name over here. That said, I don't think it makes sense to wait for that before merging this PR; the "model name" line is largely ignored by NetLogo implementations and consumers of export
.csv
files alike.