-
Notifications
You must be signed in to change notification settings - Fork 1k
Incorrect version is selected #399
Comments
Hi, thanks for posting an issue! 😄 And, eeek - that's the whole, unmodified trace output? If so, there's definitely a bug somewhere; it shouldn't be possible for the solver to move through those versions without reporting a failure reason. (There's no more verbose mode than that) |
It is both modified (project name, URLs are changed) and truncated (at Here is another one:
Both |
I added
It doesn't make sense for me: all versions are importable, and |
Ok, I tracked that one down (there was a lot of red herring on the way). Library package
Import path What I think can be improved:
|
Thanks for putting the time in on figuring this one out - you really dug in deep! 🎉
The question is, what more information would you want to see? Everything here kinda revolves around the solve tracer, which is tricky. Solve tracing has generally been sufficient so far - as you discovered, once you instrumented it and added that error message, the issue here became a lot clearer. It's sufficient, at least, to understand where the logical problem was. The problem is, it's output that's not designed for reprocessing or extension; the visualization it presents would be pretty torn up by just dumping data into the trace. But we kinda need the trace to contextualize the information we have. One class of things we might want to see is "warnings" - I have a standalone issue for them, which touches on the question of how they should be recombined with solve tracing: sdboyer/gps#77
Probably, yeah. Useless in the solver, I think, but potentially quite useful in the I've avoided it thus far because I have a longer-term plan with errors that I've been...well, it's probably time to get moving on it, now - sdboyer/gps#20. But I deferred using pkg/errors because it wasn't clear to me how it would integrate later. Not a great reason.
A fine question, and the one I was planning on figuring out 😄
Well, it depends on which part. The That said, we can't budge on relative imports that escape the tree. If they ascend past the project root, they have to be discarded. All of this, though, is ignoring the basic fact that we don't have handling for the |
What with gps moving into dep, and dep already depending on |
I think it should log as much as possible. The main use case for that information is a reporting of issues:
I don't think we could expect it from all users. Debug log should be enough, in my opinion. |
Sure, I'm on board with these principles, but I was asking a more specific question (in fulfillment of the first bullet, really). I'm trying to understand if, for this particular case, there's more logging you think to be necessary beyond the bit you added. |
I think "try" error should always be logged when
+1. So in this particular case bug should be fixed, not error message. 😆 |
yep! we may need to wait until we get gps merged in directly, but that should hopefully happen pretty soon, now. |
OK, new gps is merged in, so gonna fix this up tonight. |
dep
from commit 4773988,Gopkg.lock
does not contain this package.Can I make logging more verbose?
The text was updated successfully, but these errors were encountered: