fix(pip): allow for different extras for different target platforms

python/private/pypi/parse_requirements.bzl

@@ -88,6 +88,7 @@ def parse_requirements(
     evaluate_markers = evaluate_markers or (lambda _ctx, _requirements: {})
     options = {}
     requirements = {}
+    reqs_with_env_markers = {}
     for file, plats in requirements_by_platform.items():
         logger.trace(lambda: "Using {} for {}".format(file, plats))
         contents = ctx.read(file)
@@ -96,16 +97,47 @@ def parse_requirements(
         # needed for the whl_library declarations later.
         parse_result = parse_requirements_txt(contents)
 
+        tokenized_options = []
+        for opt in parse_result.options:
+            for p in opt.split(" "):
+                tokenized_options.append(p)
+
+        pip_args = tokenized_options + extra_pip_args
+        for plat in plats:
+            requirements[plat] = parse_result.requirements
+            for entry in parse_result.requirements:
+                requirement_line = entry[1]
+
+                # output all of the requirement lines that have a marker
+                if ";" in requirement_line:
+                    reqs_with_env_markers.setdefault(requirement_line, []).append(plat)
+            options[plat] = pip_args
+
+    # This may call to Python, so execute it early (before calling to the
+    # internet below) and ensure that we call it only once.
+    #
+    # NOTE @aignas 2024-07-13: in the future, if this is something that we want
+    # to do, we could use Python to parse the requirement lines and infer the
+    # URL of the files to download things from. This should be important for
+    # VCS package references.
+    env_marker_target_platforms = evaluate_markers(ctx, reqs_with_env_markers)
+    logger.trace(lambda: "Evaluated env markers from:\n{}\n\nTo:\n{}".format(
+        reqs_with_env_markers,
+        env_marker_target_platforms,
+    ))
+
+    requirements_by_platform = {}
+    for target_platform, reqs_ in requirements.items():
         # Replicate a surprising behavior that WORKSPACE builds allowed:
         # Defining a repo with the same name multiple times, but only the last
         # definition is respected.
         # The requirement lines might have duplicate names because lines for extras
         # are returned as just the base package name. e.g., `foo[bar]` results
         # in an entry like `("foo", "foo[bar] == 1.0 ...")`.
-        # Lines with different markers are not condidered duplicates.
+        # Lines with different markers are not considered duplicates.
         requirements_dict = {}
         for entry in sorted(
-            parse_result.requirements,
+            reqs_,
             # Get the longest match and fallback to original WORKSPACE sorting,
             # which should get us the entry with most extras.
             #
@@ -114,33 +146,28 @@ def parse_requirements(
             # should do this now.
             key = lambda x: (len(x[1].partition("==")[0]), x),
         ):
-            req = requirement(entry[1])
-            requirements_dict[(req.name, req.version, req.marker)] = entry
+            req_line = entry[1]
+            req = requirement(req_line)
 
-        tokenized_options = []
-        for opt in parse_result.options:
-            for p in opt.split(" "):
-                tokenized_options.append(p)
+            if req.marker:
+                plats = env_marker_target_platforms.get(req_line, [])
+                if not plats:
+                    continue
+                elif target_platform not in plats:
+                    continue
 
-        pip_args = tokenized_options + extra_pip_args
-        for plat in plats:
-            requirements[plat] = requirements_dict.values()
-            options[plat] = pip_args
+            # FIXME @aignas 2025-11-01: I don't think the `req.version` should be here
+            # because WORKSPACE would create files based on the `req.name`.
+            requirements_dict[(req.name, req.version)] = entry
 
-    requirements_by_platform = {}
-    reqs_with_env_markers = {}
-    for target_platform, reqs_ in requirements.items():
         extra_pip_args = options[target_platform]
 
-        for distribution, requirement_line in reqs_:
+        for distribution, requirement_line in requirements_dict.values():
             for_whl = requirements_by_platform.setdefault(
                 normalize_name(distribution),
                 {},
             )
 
-            if ";" in requirement_line:
-                reqs_with_env_markers.setdefault(requirement_line, []).append(target_platform)
-
             for_req = for_whl.setdefault(
                 (requirement_line, ",".join(extra_pip_args)),
                 struct(
@@ -153,19 +180,6 @@ def parse_requirements(
             )
             for_req.target_platforms.append(target_platform)
 
-    # This may call to Python, so execute it early (before calling to the
-    # internet below) and ensure that we call it only once.
-    #
-    # NOTE @aignas 2024-07-13: in the future, if this is something that we want
-    # to do, we could use Python to parse the requirement lines and infer the
-    # URL of the files to download things from. This should be important for
-    # VCS package references.
-    env_marker_target_platforms = evaluate_markers(ctx, reqs_with_env_markers)
-    logger.trace(lambda: "Evaluated env markers from:\n{}\n\nTo:\n{}".format(
-        reqs_with_env_markers,
-        env_marker_target_platforms,
-    ))
-
     index_urls = {}
     if get_index_urls:
         index_urls = get_index_urls(

tests/pypi/parse_requirements/parse_requirements_tests.bzl

@@ -766,7 +766,7 @@ def _test_get_index_urls_single_py_version(env):
     env.expect.that_collection(got).contains_exactly([
         struct(
             is_exposed = True,
-            is_multiple_versions = True,
+            is_multiple_versions = False,
             name = "foo",
             srcs = [
                 struct(