fix: unique-per-call probe file in copy_article_images()

When two or more workers render articles from the same source
directory in parallel, the fixed filename
  <path_dir(input_path)>/--find-assets.html
written by copy_article_images() collides between workers, surfacing as

  [EEXIST] Failed to copy '<built>.html' to '<src>/--find-assets.html':
    file already exists
  [ENOENT] Failed to remove '<src>/--find-assets.html':
    no such file or directory

depending on whether the failing worker raced on the file_copy or on
its sibling's deferred file_delete. The race is observable with as
few as two workers and bites reliably at >= 6 workers on a 400-
vignette package (see nlmixr2/nlmixr2lib PR #427, run 26225121646).

The probe file has to live next to the input Rmd so the relative
paths in the built HTML resolve when rmarkdown::find_external_resources
reads them back; tempdir() would break that. Generate a unique
filename per call via tempfile(pattern = "--find-assets-", tmpdir =
path_dir(input_path), fileext = ".html") so concurrent workers in the
same source directory each get their own probe.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: billdenney

NEWS.md

@@ -2,6 +2,8 @@
 
 * `build_articles()`, `build_site()`, and `build_site_github_pages()` gain an `n_cores` argument to build articles in parallel via `purrr::in_parallel()`. The default (`n_cores = 1L`) preserves the traditional serial build; values greater than 1 require the mirai package, and `Inf` autodetects via `parallel::detectCores()`.
 
+* The internal `copy_article_images()` helper no longer races on a fixed `--find-assets.html` probe file when articles in the same source directory are built in parallel; the probe filename is now made unique per call via `tempfile()`.
+
 * When previewing a site, it is now served via a local http server. This enables dynamic features such as search to work correctly (@shikokuchuo, #2975).
 
 * do not autolink code that is in a link (href) in Rd files (#2972)

R/build-article.R

@@ -260,11 +260,18 @@ copy_article_images <- function(built_path, input_path, output_path) {
   ext_src <- rmarkdown::find_external_resources(input_path)
 
   # temporarily copy the rendered html into the input path directory and scan
-  # again for additional external resources that may be been included by R code
-  tempfile <- path(path_dir(input_path), "--find-assets.html")
-  withr::defer(try(file_delete(tempfile)))
-  file_copy(built_path, tempfile)
-  ext_post <- rmarkdown::find_external_resources(tempfile)
+  # again for additional external resources that may be been included by R code.
+  # The probe file has to live next to the input Rmd so relative paths in the
+  # HTML resolve correctly; use tempfile() so concurrent workers in the same
+  # source directory don't race on a fixed filename.
+  probe_file <- tempfile(
+    pattern = "--find-assets-",
+    tmpdir = path_dir(input_path),
+    fileext = ".html"
+  )
+  withr::defer(try(file_delete(probe_file)))
+  file_copy(built_path, probe_file)
+  ext_post <- rmarkdown::find_external_resources(probe_file)
 
   ext <- rbind(ext_src, ext_post)
   ext <- ext[!duplicated(ext$path), ]