[eliotmcintire]

Each year the module picks which small map cells ignite inside every coarse cell that has a fire. It did this by searching the whole landscape table once or twice per fire, which was slow on large maps; it now sorts the candidates once. The fires, escapes and random numbers are exactly the same.

Affects: speed only; results identical.

Technical details

Defect: fireSense_ignitionPredict.R:196-206 (development f11d27e) did igs[chunkyPixels == n] and ig[pixelID %in% drawn] on the 1.26M-row fine-pixel table for every ignited coarse pixel (300+ a year). New placeIgnitions() (same file, after fineCoarseTable()) does one split() of the candidate rows and a pixelID -> row lookup. The sample() calls are the same, in the same order, on the same candidate vectors, so the random stream is unchanged. The sample(x, size) behaviour for a single candidate (draws from 1:x) is kept on purpose; it is a separate known issue, not fixed here. The module has no R/ folder and its helpers live in the module file, so the function is there; the test is tests/testthat/test-place-ignitions.R.

Tests: the new test compares against a labelled copy of the old code for 6 seeds, a coarse pixel with exactly one candidate, and coarse pixels with zero ignitions (identical output and .Random.seed). convertToPackage(buildDocuments = TRUE) + test_local: branch all pass (2 existing skips for the known single-candidate defect); base with only the new test file: 3 errors (no placeIgnitions), the rest pass.

Real 14.3 state (profile harness, 4 threads), seeds 11/2/3: identical() TRUE for ignitionsAndEscapes (335/344/328 ignitions), the probability rasters (seed 11) and .Random.seed. IgnitionPredictRun 2.2-2.3 s -> 0.4-0.5 s.

🤖 Generated with Claude Code

https://claude.ai/code/session_01CwcjqqK59FmTJscyi7xUqv