Fix handling of wsl output paths

NEWS.md

@@ -1,5 +1,8 @@
 # cmdstanr (development version)
 
+* When using CmdStan through WSL, paths for output, diagnostic, profile, config, 
+and metric files now remain accessible to Windows R when an explicit output 
+directory is supplied. (#1110; related: #1113)
 * Lists of matrices/vectors and data frames can now be supplied for variables
 declared as `int` in the Stan program. Previously these worked only for `real`
 variables and errored for `int` ones. (#817)

R/args.R

@@ -61,12 +61,10 @@ CmdStanArgs <- R6::R6Class(
       self$using_tempdir <- is.null(output_dir)
       self$model_variables <- model_variables
       self$save_cmdstan_config <- save_cmdstan_config
-      if (os_is_wsl()) {
+      if (os_is_wsl() && self$using_tempdir) {
         # Want to ensure that any files under WSL are written to a tempdir within
         # WSL to avoid IO performance issues
-        self$output_dir <- ifelse(is.null(output_dir),
-                                  file.path(wsl_dir_prefix(), wsl_tempdir()),
-                                  wsl_safe_path(output_dir))
+        self$output_dir <- file.path(wsl_dir_prefix(), wsl_tempdir())
       } else {
         self$output_dir <- output_dir %||% tempdir(check = TRUE)
       }

R/fit.R

@@ -35,9 +35,6 @@ CmdStanFit <- R6::R6Class(
       if (!is.null(private$model_methods_env_$model_ptr)) {
         initialize_model_pointer(private$model_methods_env_, self$data_file(), 0)
       }
-      # Need to update the output directory path to one that can be accessed
-      # from Windows, for the post-processing of results
-      self$runset$args$output_dir <- wsl_safe_path(self$runset$args$output_dir, revert = TRUE)
       invisible(self)
     },
     num_procs = function() {

R/run.R

@@ -76,7 +76,6 @@ CmdStanRun <- R6::R6Class(
     },
     config_files = function(include_failed = FALSE) {
       files <- private$config_files_
-      files_win_path <- sapply(private$config_files_, wsl_safe_path, revert = TRUE)
       if (include_failed) {
         files
       } else {
@@ -86,7 +85,6 @@ CmdStanRun <- R6::R6Class(
     },
     metric_files = function(include_failed = FALSE) {
       files <- private$metric_files_
-      files_win_path <- sapply(private$metric_files_, wsl_safe_path, revert = TRUE)
       if (include_failed) {
         files
       } else {

tests/testthat/test-model-output_dir.R

@@ -4,6 +4,98 @@ local_output_sandbox <- function(pattern = "sandbox", .local_envir = parent.fram
   withr::local_tempdir(pattern = pattern, .local_envir = .local_envir)
 }
 
+test_that("WSL output paths stay host-native until command composition", {
+  # Use minimal method arguments so this test exercises path handling without
+  # launching CmdStan.
+  method_args <- list(
+    method = "sample",
+    save_metric = NULL,
+    validate = function(num_procs) invisible(),
+    compose = function(idx, args) args
+  )
+  # Cover system and non-system Windows drives as well as a WSL UNC path.
+  host_dirs <- c(
+    "C:/output",
+    "D:/output",
+    "//wsl$/Ubuntu/home/user/output"
+  )
+  wsl_dirs <- c(
+    "/mnt/c/output",
+    "/mnt/d/output",
+    "/home/user/output"
+  )
+  as_wsl_path <- function(path = NULL, revert = FALSE) {
+    if (is.null(path) || revert) {
+      return(path)
+    }
+    path <- sub("//wsl$/Ubuntu", "", path, fixed = TRUE)
+    for (i in seq_along(host_dirs)) {
+      path <- sub(host_dirs[i], wsl_dirs[i], path, fixed = TRUE)
+    }
+    path
+  }
+  # Simulate Windows R using WSL so this boundary test runs on every platform.
+  with_mocked_bindings(
+    {
+      args <- lapply(host_dirs, function(output_dir) {
+        CmdStanArgs$new(
+          model_name = "model",
+          exe_file = "model",
+          proc_ids = 1,
+          method_args = method_args,
+          output_dir = output_dir,
+          output_basename = "model"
+        )
+      })
+      output_files <- file.path(host_dirs, "model-1.csv")
+      expect_equal(
+        vapply(args, function(x) x$output_dir, character(1)),
+        host_dirs
+      )
+      expect_equal(
+        vapply(args, function(x) x$new_files("output"), character(1)),
+        output_files
+      )
+      cmdstan_output_files <- vapply(seq_along(args), function(i) {
+        command_args <- args[[i]]$compose_all_args(
+          output_file = output_files[i]
+        )
+        sub("file=", "", command_args[grepl("^file=", command_args)], fixed = TRUE)
+      }, character(1))
+      expect_equal(cmdstan_output_files, file.path(wsl_dirs, "model-1.csv"))
+
+      command_args <- args[[1]]$compose_all_args(
+        output_file = output_files[1],
+        profile_file = file.path(host_dirs[1], "model-profile-1.csv"),
+        latent_dynamics_file = file.path(host_dirs[1], "model-diagnostic-1.csv")
+      )
+      expect_in("diagnostic_file=/mnt/c/output/model-diagnostic-1.csv", command_args)
+      expect_in("profile_file=/mnt/c/output/model-profile-1.csv", command_args)
+
+      # Omitting output_dir must still use the faster WSL-native temp directory.
+      default_args <- CmdStanArgs$new(
+        model_name = "model",
+        exe_file = "model",
+        proc_ids = 1,
+        method_args = method_args,
+        output_basename = "model"
+      )
+      expect_equal(default_args$output_dir, "//wsl$/Ubuntu/tmp/cmdstanr")
+      expect_in(
+        "file=/tmp/cmdstanr/model-1.csv",
+        default_args$compose_all_args(
+          output_file = default_args$new_files("output")
+        )
+      )
+    },
+    os_is_wsl = function() TRUE,
+    wsl_safe_path = as_wsl_path,
+    wsl_dir_prefix = function(...) "//wsl$/Ubuntu",
+    wsl_tempdir = function() "/tmp/cmdstanr",
+    validate_cmdstan_args = function(self) invisible()
+  )
+})
+
 test_that("all fitting methods work with output_dir", {
   sandbox <- local_output_sandbox()
   for (method in c("sample", "optimize", "variational")) {
@@ -29,24 +121,22 @@ test_that("all fitting methods work with output_dir", {
       call_args$save_metric <- TRUE
     }
     fit <- do.call(testing_fit, call_args)
-    # WSL path manipulations result in a short path which slightly differs
-    # from the original tempdir(), so need to normalise both for comparison
+    # Normalize to account for platform-specific path representations.
     expect_equal(normalizePath(fit$runset$args$output_dir),
                  normalizePath(method_dir))
     files <- normalizePath(list.files(method_dir, full.names = TRUE))
+    expect_equal(files[grepl("\\.csv$", files)],
+                 normalizePath(fit$output_files()))
     if (method == "sample") {
       mult <- 3
       expect_equal(files[grepl("metric", files)],
-                   normalizePath(sapply(fit$metric_files(), wsl_safe_path, revert = TRUE,
-                                        USE.NAMES = FALSE)))
+                   normalizePath(fit$metric_files()))
       expect_equal(files[grepl("config", files)],
-                   normalizePath(sapply(fit$config_files(), wsl_safe_path, revert = TRUE,
-                                        USE.NAMES = FALSE)))
+                   normalizePath(fit$config_files()))
     } else {
       mult <- 2
       expect_equal(files[grepl("config", files)],
-                   normalizePath(sapply(fit$config_files(), wsl_safe_path, revert = TRUE,
-                                        USE.NAMES = FALSE)))
+                   normalizePath(fit$config_files()))
     }
     expect_equal(length(list.files(method_dir)), mult * fit$num_procs())
 
@@ -69,6 +159,84 @@ test_that("all fitting methods work with output_dir", {
     sum(grepl("diagnostic", files)),
     fit$num_procs()
   )
+  expect_equal(
+    normalizePath(fit$latent_dynamics_files()),
+    normalizePath(list.files(
+      file.path(sandbox, "sample"),
+      pattern = "diagnostic",
+      full.names = TRUE
+    ))
+  )
+})
+
+test_that("explicit WSL output paths are usable by Windows R", {
+  skip_if_not(os_is_wsl())
+  # Unlike the mocked test above, this exercises the full Windows/WSL workflow.
+  output_dir <- local_output_sandbox("wsl-output-dir")
+  mod <- testing_model("logistic_profiling")
+  utils::capture.output(
+    fit <- mod$sample(
+      data = testing_data("logistic"),
+      chains = 1,
+      parallel_chains = 1,
+      seed = 123,
+      refresh = 0,
+      output_dir = output_dir,
+      save_latent_dynamics = TRUE,
+      save_cmdstan_config = TRUE,
+      save_metric = TRUE
+    )
+  )
+  paths <- c(
+    fit$output_files(),
+    fit$latent_dynamics_files(),
+    fit$profile_files(),
+    fit$config_files(),
+    fit$metric_files()
+  )
+  expect_equal(file.exists(paths), rep(TRUE, length(paths)))
+  expect_equal(
+    normalizePath(dirname(paths)),
+    rep(normalizePath(output_dir), length(paths))
+  )
+  expect_output(fit$cmdstan_summary(), "Inference for Stan model")
+  expect_output(fit$cmdstan_diagnose(), "Processing complete")
+
+  # All generated file types should remain usable when moved by Windows R.
+  save_root <- local_output_sandbox("wsl-save-files")
+  save_dirs <- file.path(
+    save_root,
+    c("output", "diagnostic", "profile", "config", "metric")
+  )
+  for (dir in save_dirs) {
+    dir.create(dir)
+  }
+  saved_paths <- suppressMessages(c(
+    fit$save_output_files(save_dirs[1]),
+    fit$save_latent_dynamics_files(save_dirs[2]),
+    fit$save_profile_files(save_dirs[3]),
+    fit$save_config_files(save_dirs[4]),
+    fit$save_metric_files(save_dirs[5])
+  ))
+  expect_equal(file.exists(saved_paths), rep(TRUE, length(saved_paths)))
+})
+
+test_that("explicit WSL UNC output_dir remains supported", {
+  skip_if_not(os_is_wsl())
+  # This covers explicit output only; #1113's temporary input paths are separate.
+  output_dir <- repair_path(file.path(wsl_dir_prefix(), wsl_tempdir()))
+  withr::defer(unlink(output_dir, recursive = TRUE))
+  fit <- testing_fit(
+    "bernoulli",
+    method = "optimize",
+    output_dir = output_dir
+  )
+
+  expect_equal(file.exists(fit$output_files()), TRUE)
+  expect_equal(
+    normalizePath(dirname(fit$output_files())),
+    normalizePath(output_dir)
+  )
 })
 
 test_that("error if output_dir is invalid", {