knn test failure in b-next

[maelle]

I have saved the edgelist of the graph that's created at

rigraph/tests/testthat/test-structural-properties.R

Lines 930 to 934 in fd62234

	g4 <- sample_gnp(1000, p = 5 / 1000)
	r4 <- knn(g4)
	expect_equal(r4$knn[1000], 20 / 3)
	expect_equal(length(r4$knnk), 15)
	expect_equal(r4$knnk[12], 19 / 3)

bla.txt

b-next

devtools::load_all()
#> ℹ Loading igraph
g4 <- read_graph("bla.txt")
r4 <- knn(g4)
r4$knn[1000] # 20 / 3
#> [1] 5
length(r4$knnk) # 15
#> [1] 15
r4$knnk[12] # 19 / 3
#> [1] 5.638889

main

library("igraph")
#> 
#> Attaching package: 'igraph'
#> The following objects are masked from 'package:stats':
#> 
#>     decompose, spectrum
#> The following object is masked from 'package:base':
#> 
#>     union
g4 <- read_graph("bla.txt")
r4 <- knn(g4)
r4$knn[1000] # 20 / 3
#> [1] 5
length(r4$knnk) # 15
#> [1] 15
r4$knnk[12] # 19 / 3
#> [1] 5.638889

^{Created on 2025-10-03 with reprex v2.1.1}

So actually it is the same. To me it means that now for whatever reason in the test we are not drawing the same graph any more.

@schochastics would you recommend:

• updating the expectation
• splitting the test so that in the future we have better control of what the random graph is?

[schochastics]

I wonder if we should (eventually) move away from using random graphs in tests. Always prone to errors if something messes with the seeds.

For now I think that splitting the test might be the best idea