Merge branch 'main' into scratch/workarenax_benchmarks

patricebechard · patricebechard · commit 839f52f4b37b · 2025-11-10T08:39:26.000-05:00
diff --git a/.github/workflows/pypi.yml b/.github/workflows/pypi.yml
@@ -28,7 +28,8 @@ jobs:
 
       - name: Build a binary wheel and a source tarball (browsergym-webarena)
         run: python3 -m build browsergym/webarena/ --outdir dist/
-
+      - name: Build a binary wheel and a source tarball (browsergym-webarenalite)
+        run: python3 -m build browsergym/webarenalite/ --outdir dist/
       - name: Build a binary wheel and a source tarball (browsergym-webarena)
         run: python3 -m build browsergym/visualwebarena/ --outdir dist/
 
@@ -85,7 +86,7 @@ jobs:
           path: dist/
 
       - name: Sign the dists with Sigstore
-        uses: sigstore/gh-action-sigstore-python@v2.1.1
+        uses: sigstore/gh-action-sigstore-python@v3.1.0
         with:
           inputs: >-
             ./dist/*.tar.gz
diff --git a/browsergym/assistantbench/requirements.txt b/browsergym/assistantbench/requirements.txt
@@ -1,4 +1,4 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
 datasets
 scipy
 numpy
diff --git a/browsergym/core/src/browsergym/core/__init__.py b/browsergym/core/src/browsergym/core/__init__.py
@@ -1,4 +1,4 @@
-__version__ = "0.14.2"
+__version__ = "0.14.3.dev1"
 
 import playwright.sync_api
 
diff --git a/browsergym/core/src/browsergym/core/action/functions.py b/browsergym/core/src/browsergym/core/action/functions.py
@@ -183,7 +183,7 @@ def dblclick(
     add_demo_mode_effects(page, elem, bid, demo_mode=demo_mode, move_cursor=True)
 
     def do(force: bool):
-        elem.click(button=button, modifiers=modifiers, force=force, timeout=500)
+        elem.dblclick(button=button, modifiers=modifiers, force=force, timeout=500)
 
     call_fun(do, retry_with_force)
 
diff --git a/browsergym/experiments/requirements.txt b/browsergym/experiments/requirements.txt
@@ -1,3 +1,3 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
 tiktoken>=0.4
 dataclasses-json
diff --git a/browsergym/miniwob/requirements.txt b/browsergym/miniwob/requirements.txt
@@ -1 +1 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
diff --git a/browsergym/pyproject.toml b/browsergym/pyproject.toml
@@ -28,16 +28,17 @@ classifiers = [
     "Topic :: Scientific/Engineering :: Artificial Intelligence",
     "License :: OSI Approved :: Apache Software License",
 ]
-version="0.14.2"
+version="0.14.3.dev1"
 dependencies = [
-    "browsergym-core==0.14.2",
-    "browsergym-miniwob==0.14.2",
-    "browsergym-webarena==0.14.2",
-    "browsergym-visualwebarena==0.14.2",
-    "browsergym-assistantbench==0.14.2",
-    "browsergym-experiments==0.14.2",
+    "browsergym-core==0.14.3.dev1",
+    "browsergym-miniwob==0.14.3.dev1",
+    "browsergym-webarena==0.14.3.dev1",
+    "browsergym-visualwebarena==0.14.3.dev1",
+    "browsergym-assistantbench==0.14.3.dev1",
+    "browsergym-experiments==0.14.3.dev1",
     "browsergym-workarena>=0.4.1",
     "weblinx-browsergym>=0.0.2",
+    "browsergym-webarenalite==0.14.3.dev1"
 ]
 
 [tool.setuptools]
diff --git a/browsergym/visualwebarena/requirements.txt b/browsergym/visualwebarena/requirements.txt
@@ -1,4 +1,4 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
 browsergym-webarena
 libvisualwebarena==0.0.15
 requests
diff --git a/browsergym/webarena/requirements.txt b/browsergym/webarena/requirements.txt
@@ -1,2 +1,2 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
 libwebarena==0.0.4
diff --git a/browsergym/webarenalite/pyproject.toml b/browsergym/webarenalite/pyproject.toml
@@ -7,6 +7,7 @@ name = "browsergym-webarenalite"
 description = "WebArena Lite benchmark for BrowserGym"
 authors = [
     {name = "Aman Jaiswal"},
+    {name = "Leo Biosvert"},
 ]
 requires-python = ">3.7"
 license = {text = "Apache-2.0"}
@@ -29,10 +30,5 @@ path = "../core/src/browsergym/core/__init__.py"
 [tool.hatch.metadata.hooks.requirements_txt]
 files = ["requirements.txt"]
 
-[tool.hatch.build]
-include = [
-    "src/browsergym/webarenalite/test_webarena_lite.raw.json"
-]
-
 [tool.hatch.build.targets.wheel]
 packages = ["src/browsergym"]
diff --git a/browsergym/webarenalite/requirements.txt b/browsergym/webarenalite/requirements.txt
@@ -1,2 +1,3 @@
-browsergym-core==0.14.2
+browsergym-core==0.14.3.dev1
+browsergym-webarena==0.14.3.dev1
 libwebarena==0.0.4
diff --git a/browsergym/webarenalite/src/browsergym/webarenalite/evaluators.py b/browsergym/webarenalite/src/browsergym/webarenalite/evaluators.py
@@ -16,7 +16,7 @@
 from beartype import beartype
 from nltk.tokenize import word_tokenize  # type: ignore
 from playwright.sync_api import CDPSession, Page
-from webarena.evaluation_harness.helper_functions import (
+from .helper_functions import (
     PseudoPage,
     gitlab_get_project_memeber_role,
     llm_fuzzy_match,
diff --git a/docs/src/conf.py b/docs/src/conf.py
@@ -6,7 +6,7 @@
 copyright = "2024, ServiceNow Research"
 author = "ServiceNow Research"
 
-version = "0.14.2"
+version = "0.14.3.dev1"
 release = version
 
 # -- General configuration
diff --git a/tests/core/data/dblclick.html b/tests/core/data/dblclick.html
@@ -0,0 +1,16 @@
+<!DOCTYPE html>
+<html>
+<body>
+    <button id="dblclick-counter" ondblclick="incrementCounter()">
+        Double-click me! Count: <span id="count">0</span>
+    </button>
+
+    <script>
+        let count = 0;
+        function incrementCounter() {
+            count++;
+            document.getElementById('count').textContent = count;
+        }
+    </script>
+</body>
+</html>
diff --git a/tests/core/test_actions_highlevel.py b/tests/core/test_actions_highlevel.py
@@ -27,6 +27,7 @@
 TEXTBOX_URL = f"file://{__DATA_DIR}/textbox.html"
 EXAMPLE_URL = f"file://{__DATA_DIR}/example.html"
 HOVER_URL = f"file://{__DATA_DIR}/hover.html"
+DBLCLICK_URL = f"file://{__DATA_DIR}/dblclick.html"
 INEXISTANT_FILE_URL = f"file://{__DATA_DIR}/no_file_here.html"
 LONG_PAGE_URL = f"file://{__DATA_DIR}/long_page.html"
 TEXT_INPUT_URL = f"file://{__DATA_DIR}/input_type/text_input.html"
@@ -826,9 +827,111 @@ def get_fname_lname_elems(obs):
     env.close()
 
 
-@pytest.mark.skip(reason="Not implemented yet")
 def test_dblclick():
-    pass
+    action_set = HighLevelActionSet(subsets=["bid"])
+
+    env = gym.make(
+        "browsergym/openended",
+        task_kwargs={"start_url": DBLCLICK_URL},
+        headless=__HEADLESS,
+        slow_mo=__SLOW_MO,
+        timeout=__TIMEOUT,
+        action_mapping=action_set.to_python_code,
+    )
+
+    def get_counter_elem(obs):
+        soup = bs4.BeautifulSoup(flatten_dom_to_str(obs["dom_object"]), "lxml")
+        counter = soup.find("span", attrs={"id": "count"})
+        return counter
+
+    obs, info = env.reset()
+    counter = get_counter_elem(obs)
+
+    # counter starts at 0
+    assert not obs["last_action_error"]
+    assert counter.text.strip() == "0"
+
+    # test dblclick using bid
+    counter_button = counter.find_parent("button", attrs={"id": "dblclick-counter"})
+    action = f"""
+dblclick({repr(counter_button.get(BID_ATTR))})
+"""
+
+    obs, reward, terminated, truncated, info = env.step(action)
+    counter = get_counter_elem(obs)
+
+    # counter incremented to 1
+    assert not obs["last_action_error"]
+    assert counter.text.strip() == "1"
+
+    # test dblclick using bid again
+    counter_button = counter.find_parent("button", attrs={"id": "dblclick-counter"})
+    action = f"""
+dblclick({repr(counter_button.get(BID_ATTR))})
+"""
+
+    obs, reward, terminated, truncated, info = env.step(action)
+    counter = get_counter_elem(obs)
+
+    # counter incremented to 2
+    assert not obs["last_action_error"]
+    assert counter.text.strip() == "2"
+
+    env.close()
+
+
+def test_mouse_dblclick():
+    action_set = HighLevelActionSet(subsets=["coord"])
+
+    env = gym.make(
+        "browsergym/openended",
+        task_kwargs={"start_url": DBLCLICK_URL},
+        headless=__HEADLESS,
+        slow_mo=__SLOW_MO,
+        timeout=__TIMEOUT,
+        action_mapping=action_set.to_python_code,
+    )
+
+    def counter():
+        return env.unwrapped.page.inner_text("#count")
+
+    def get_counter_coords():
+        button = env.unwrapped.page.locator("button")
+        box = button.bounding_box()
+        center_x = box["x"] + box["width"] / 2
+        center_y = box["y"] + box["height"] / 2
+        return center_x, center_y
+
+    obs, info = env.reset()
+
+    # counter starts at 0    assert not obs["last_action_error"]
+    assert counter() == "0"
+
+    # test mouse_dblclick using coordinates
+    counter_x, counter_y = get_counter_coords()
+    action = f"""
+mouse_dblclick({counter_x}, {counter_y})
+"""
+
+    obs, reward, terminated, truncated, info = env.step(action)
+
+    # counter incremented to 1
+    assert not obs["last_action_error"]
+    assert counter() == "1"
+
+    counter_x, counter_y = get_counter_coords()
+
+    action = f"""
+mouse_dblclick({counter_x}, {counter_y})
+"""
+
+    obs, reward, terminated, truncated, info = env.step(action)
+
+    # counter incremented to 2
+    assert not obs["last_action_error"]
+    assert counter() == "2"
+
+    env.close()
 
 
 # copy/paste text using a sequence of keyboard_press actions

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,4 @@`
`1`		`-__version__ = "0.14.2"`
	`1`	`+__version__ = "0.14.3.dev1"`
`2`	`2`
`3`	`3`	`import playwright.sync_api`
`4`	`4`
Original file line number	Diff line number	Diff line change
`@@ -1 +1 @@`
`1`		`-browsergym-core==0.14.2`
	`1`	`+browsergym-core==0.14.3.dev1`
Original file line number	Diff line number	Diff line change
`@@ -1,2 +1,2 @@`
`1`		`-browsergym-core==0.14.2`
	`1`	`+browsergym-core==0.14.3.dev1`
`2`	`2`	`libwebarena==0.0.4`