Closed michaelkirk closed 4 years ago
I think that as long as it's clearly documented in proj and geo it's an acceptable increase in complexity – the reduced build footprint is worth it 👍
I think that as long as it's clearly documented in proj and geo
Cool, let me add some docs for the feature flag.
Ok - can you take a look at my doc changes @urschrei?
`
OK - i've got the feature plumbing waiting in the wings: https://github.com/georust/geo/pull/506
I think this is good to merge?
Oh wait - there's a CI failure. Investigating...
Oh wait - there's a CI failure. Investigating...
Test failure on bundled_proj only:
stderr:
thread 'main' panicked at 'assertion failed: `(left !== right)` (left: `499972.7074429854`, right: `500119.7035366755`, expect diff: `0.000001`, real diff: `146.99609369010432`)', src/proj.rs:16:1
I can't reproduce this failure locally.
It's a suspiciously small diff, so I thought maybe related to somehow inadvertently using a network grid, but I don't see how that's possible, given the feature is never enabled in this VM's build. Furthermore, I changed the test locally to use the grid and still wasn't able to reproduce the test failure.
Any theories @urschrei ?
I re-ran CI and the test passed. 👻
It's a transient failure, and I have no theory as to why it crops up sometimes (and we all know what that means: writing a property test to figure out what terrible thing I've subtly screwed up). But for now, let's merge it.
(I would say that's quite a big difference?)
(I would say that's quite a big difference?)
Yeah true.
In relative terms it's 0.03% - so definitely large enough to have a meaningful impact - but my understanding is that was within the realm of what the grid offered. I have no idea if it's grid related to be clear.
Hmm. I need to look at it more carefully, but generally you use a grid for survey-grade conversions where sub-10 cm accuracy is required. So ~150 m (is the datum in meters? I think so…😬) is in Blunder (or just "gross error": see III-5 territory.
bors r=urschrei
On Sep 6, 2020, at 10:34, Stephan Hügel notifications@github.com wrote:
Hmm. I need to look at it more carefully, but generally you use a grid for survey-grade conversions where sub-10 cm accuracy is required. So ~150 m (is the datum in meters? I think so…😬) is in Blunder (or just "gross error": see III-5 territory.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.
Build succeeded:
The network stuff adds quite a bit to proj dependencies and build times.
Considering the developer has to opt in to networking anyway by calling enable_network on the context, what would you think about making them jump through one more hoop to make things better for the default user?
With reqwest: builds 166 crates
Without reqwest: builds 81 crates
If you think it's a good idea, I can open a PR to plumb this feature through geo.
dependency tree: