Prevent non-atomic writes to repodata JSON files

Fixes: conda#3833

CHANGELOG.txt

@@ -17,6 +17,12 @@ Contributors:
 * @AndrewAnnex
 * @183amir
 * @njzjz
+* @sscherfke
+
+Bug fixes:
+----------
+
+* Prevent non-atomic writes to repodata JSON files  #3833
 
 
 2019-10-14 3.18.10:
@@ -79,7 +85,7 @@ Enhancements:
 -------------
 
 * add --use-channeldata argument to conda render/build.
-* Extract the part in the skeletons pypi responsible to get the package metadata to a free function. 
+* Extract the part in the skeletons pypi responsible to get the package metadata to a free function.
 * Creat unittests for the get_package_metadata (skeletons/pypi.py) and for the new functions.
 
 Bug fixes:

conda_build/index.py

@@ -14,7 +14,6 @@
 from os.path import abspath, basename, getmtime, getsize, isdir, isfile, join, splitext, dirname
 import subprocess
 import sys
-from tempfile import gettempdir
 import time
 from uuid import uuid4
 
@@ -419,7 +418,9 @@ def _filter_add_href(text, link, **kwargs):
 
 
 def _maybe_write(path, content, write_newline_end=False, content_is_binary=False):
-    temp_path = join(gettempdir(), str(uuid4()))
+    # Create the temp file next "path" so that we can use an atomic move, see
+    # https://github.com/conda/conda-build/issues/3833
+    temp_path = '%s.%s' % (path, uuid4())
 
     if not content_is_binary:
         content = ensure_binary(content)