fix read of remote parquet files

R/api_check.R

@@ -903,14 +903,17 @@
 #' \itemize{
 #' \item{
 #' \code{.check_file()} throws an error if provided value is not a valid and
-#' existing file path.
+#' existing file path. With \code{allow_remote = TRUE}, remote files
+#' (\code{http}, \code{https}, \code{s3}) are accepted after the extension
+#' check, since they have no local path to be found or created.
 #' }
 #' }
 #' @keywords internal
 #' @noRd
 .check_file <- function(x, ...,
                         extensions = NULL,
                         file_exists = TRUE,
+                        allow_remote = FALSE,
                         local_msg = NULL,
                         msg = NULL) {
     # check parameter name
@@ -945,6 +948,11 @@
             local_msg = local_msg
         )
     }
+    # remote files are read by their consumer, so there is no local
+    # file to find or create
+    if (allow_remote && !.file_is_local(x)) {
+        return(invisible(x))
+    }
     if (file_exists) {
         existing_files <- file.exists(x)
         existing_dirs <- dir.exists(x)

R/sits_parquet.R

@@ -102,11 +102,17 @@ sits_to_parquet.default <- function(data, file) {
 #'   Files from another layout version are read, not refused. A warning is
 #'   raised.
 #'
-#' @param  file   Full path of the file to read
+#'   Remote files (\code{http} or \code{https} URLs) are downloaded by
+#'   \code{arrow} to a temporary file before being read.
+#'
+#' @param  file   Full path or URL of the file to read
 #'                (valid file name with extension ".parquet").
 #' @return        Time series (tibble of class "sits").
 #'
-#' @note Requires the \code{arrow} and \code{jsonlite} packages.
+#' @note Requires the \code{arrow} and \code{jsonlite} packages. Remote
+#'   files are downloaded with \code{utils::download.file}, which is limited
+#'   by \code{getOption("timeout")}. Large files may need a higher value,
+#'   e.g. \code{options(timeout = 600)}.
 #'
 #' @examples
 #' if (sits_run_examples()) {
@@ -121,10 +127,13 @@ sits_to_parquet.default <- function(data, file) {
 sits_from_parquet <- function(file) {
     .check_set_caller("sits_from_parquet")
     .check_require_packages(c("arrow", "jsonlite"))
-    .check_file(x = file, extensions = "parquet")
-    reader <- arrow::ParquetFileReader$create(file)
-    block <- reader$GetSchema()$metadata[["sits"]]
-    tbl <- arrow::read_parquet(file)
+    .check_file(x = file, extensions = "parquet", allow_remote = TRUE)
+    # read as arrow::Table
+    tbl <- arrow::read_parquet(file, as_data_frame = FALSE)
+    # extract metadata
+    block <- tbl$metadata[["sits"]]
+    # convert arow::Table to tibble
+    tbl <- tibble::as_tibble(tbl)
     # no block: infer the class, then rebuild through the same path
     if (!.has(block)) {
         warning(.conf("messages", "sits_from_parquet_no_metadata"),

tests/testthat/test-parquet.R

@@ -107,6 +107,30 @@ test_that("Parquet errors are informative", {
     expect_error(suppressWarnings(sits_from_parquet(plain_file)))
 })
 
+test_that("Samples can be read from a remote parquet file", {
+    skip_if_not_installed("arrow")
+    skip_if_not_installed("jsonlite")
+
+    # a remote file must still have the parquet extension
+    expect_error(sits_from_parquet("https://example.com/samples.csv"))
+
+    skip_on_cran()
+    skip_if_offline()
+
+    # define remote file (parquet from sits book)
+    url <- paste0(
+        "https://huggingface.co/datasets/gilbertocamara/samples_book/",
+        "resolve/main/samples_deforestation_rondonia.parquet"
+    )
+
+    # read the file
+    samples <- sits_from_parquet(url)
+
+    expect_s3_class(samples, "sits")
+    expect_gt(nrow(samples), 0L)
+    expect_true("time_series" %in% colnames(samples))
+})
+
 test_that("Class is inferred when the file has no sits metadata", {
     skip_if_not_installed("arrow")
     skip_if_not_installed("jsonlite")