(joe) switch from multiprocessing.pool to internal wrapper, thanks to @McSinyx and @uranusjr for the warning about the weird ways in which different pythons are broken

Author: jbylund

src/pip/_internal/resolution/resolvelib/resolver.py

@@ -22,6 +22,7 @@
 )
 from pip._internal.utils.deprecation import deprecated
 from pip._internal.utils.filetypes import is_archive_file
+from pip._internal.utils.parallel import map_multithread, LACK_SEM_OPEN
 
 from .base import Candidate, Requirement
 from .factory import Factory
@@ -90,15 +91,21 @@ def resolve(
             reporter,
         )
 
-        def _maybe_find_candidates(req: Requirement) -> None:
-            ident = provider.identify(req)
-            try:
-                self.factory._finder.find_all_candidates(ident)
-            except AttributeError:
-                pass
 
-        with ThreadPool() as tp:
-            for _ in tp.imap_unordered(_maybe_find_candidates, collected.requirements):
+        if LACK_SEM_OPEN:
+            # if a working threading implementation unavailable do nothing
+            pass
+        else:
+            # otherwise greedily call/consume _finder.find_all_candidates in parallel in order to
+            # populate the lru cache such that future calls don't block on networking
+            def _maybe_find_candidates(req: Requirement) -> None:
+                ident = provider.identify(req)
+                try:
+                    self.factory._finder.find_all_candidates(ident)
+                except AttributeError:
+                    pass
+
+            for _ in map_multithread(_maybe_find_candidates, collected.requirements):
                 pass
 
         try: