vertexTFCE "random" crashes?

[gdevenyi]

Having some "random" crashes with vertexTFCE. With rstudio I wans't getting an error dump, but I just attempted on the console and got this:

> model = vertexLm(left_thickness ~  M.F*poly(Age, 2), filtered.data)                                             
N: 316 P: 6
Beginning vertex loop: 40962 7
Done with vertex loop
> configureMincParallel("sge")
Sourcing configuration file: '/opt/quarantine/RMINC/git-develop/build/RMINC/parallel/sge_BatchJobs.R'
> TFCE = vertexTFCE(model, surface="surf_reg_model_left.obj", parallel=c("sge",100), weights = as.numeric(readLines("surf_reg_model_left_area.txt")))     

 *** caught segfault ***
address 0x10a63eab0, cause 'memory not mapped'

Traceback:
 1: .Call("RMINC_graph_tfce", PACKAGE = "RMINC", map, adjacencies,     E, H, nsteps, weights)
 2: graph_tfce(x, adjacencies, E, H, nsteps, weights)
 3: vertexTFCE.numeric(x[, col_ind], surface[[surface_ind]], E = E,     H = H, nsteps = nsteps, weights = weights)
 4: (function (col_ind, surface_ind) {    vertexTFCE.numeric(x[, col_ind], surface[[surface_ind]],         E = E, H = H, nsteps = nsteps, weights = weights)})(dots[[1L]][[1L]], dots[[2L]][[1L]])
 5: mapply(function(col_ind, surface_ind) {    vertexTFCE.numeric(x[, col_ind], surface[[surface_ind]],         E = E, H = H, nsteps = nsteps, weights = weights)}, seq_len(ncol(x)), surface_inds)
 6: vertexTFCE.matrix(lmod[, columns], surface = surface, weights = weights,     d = d, E = E, H = H, side = side, like_volume = like_vol)
 7: vertexTFCE.vertexLm(model, surface = "surf_reg_model_left.obj",     parallel = c("sge", 100), weights = as.numeric(readLines("surf_reg_model_left_area.txt")))
 8: vertexTFCE(model, surface = "surf_reg_model_left.obj", parallel = c("sge",     100), weights = as.numeric(readLines("surf_reg_model_left_area.txt")))

[gdevenyi]

Less random, seems to consistent with this dataset.

Tried removing weights, same result.

[cfhammill]

Are there any NAs in filtered.data?

[gdevenyi]

In the dataframe, yes, in the variables M.F and Age, no.

[cfhammill]

Hmm, not quite sure what's happening, thought it had to do with arguments getting passed wrong, but it looks like defaults should have protected from a seg fault. Could you send an RDS of the model and I'll try to debug. It's failing before it goes to the cluster so I should be able to solve it without the raw data.

[gdevenyi]

Sent.

[cfhammill]

Should be fixed in dev :+1:

[gdevenyi]

Fixed!