refactor(viewer): collapse five copies of get_results_dir into paths.py

viewer/ai_comparer.py

@@ -11,6 +11,7 @@
 from evalbench.generators.models import get_generator
 from evalbench.util.config import load_yaml_config
 from summarizer import get_summarizer
+from paths import get_results_dir
 
 # Setup logging
 logging.basicConfig(level=logging.INFO)
@@ -19,28 +20,6 @@
 # Global models dict for get_generator
 global_models = {"lock": threading.Lock(), "registered_models": {}}
 
-def get_results_dir():
-    # Try to read from environment variable
-    res_dir = os.environ.get("RESULTS_DIR")
-    if res_dir:
-        return res_dir
-
-    # Check multiple locations for results directory
-    results_dir_candidates = [
-        "/tmp_session_files/results",
-        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
-        os.path.join(os.getcwd(), "results"),
-        "/evalbench/results"
-    ]
-
-    for candidate in results_dir_candidates:
-        if os.path.exists(candidate):
-            logger.info(f"Found results directory at: {candidate}")
-            return candidate
-
-    logger.warning("Results directory not found in candidates, defaulting to current directory results")
-    return os.path.join(os.getcwd(), "results")
-
 def compare_evals(id1, id2):
     """Compares two evaluation runs using Gemini."""
     results_dir = get_results_dir()

viewer/dataset_quality.py

@@ -13,13 +13,13 @@
 
 import mesop as me
 
+from paths import get_results_dir
 from precompute_dataset_quality import (
     CACHE_FILENAME,
     DATASET_FORMAT_CONFIG_KEY,
     DATASET_QUALITY_FORMAT,
     SUMMARY_COMPARATOR,
     artifact_paths,
-    get_results_dir,
 )
 
 csv.field_size_limit(10**9)

viewer/main.py

@@ -12,6 +12,7 @@
 import dataset_quality
 from summarizer import summarize_eval_scoring
 from ai_comparer import compare_evals
+from paths import get_results_dir
 
 @me.stateclass
 class State:
@@ -116,26 +117,6 @@ def df_to_config(df: pd.DataFrame) -> dict:
 
 
 
-def get_results_dir():
-    # Try to read from environment variable
-    res_dir = os.environ.get("RESULTS_DIR")
-    if res_dir:
-        return res_dir
-
-    # Check multiple locations for results directory
-    results_dir_candidates = [
-        "/tmp_session_files/results",
-        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
-        os.path.join(os.getcwd(), "results"),
-    ]
-
-    for candidate in results_dir_candidates:
-        if os.path.exists(candidate) and os.path.isdir(candidate):
-            return candidate
-
-    return results_dir_candidates[1]  # Fallback to default
-
-
 # (mtime, value). Each is replaced as a whole so a concurrent reader never sees a
 # value that disagrees with the mtime it was built from.
 _TRENDS_DF_CACHE = (None, None)

viewer/paths.py

@@ -0,0 +1,21 @@
+import os
+
+
+def get_results_dir():
+    res_dir = os.environ.get("RESULTS_DIR")
+    if res_dir:
+        return res_dir
+
+    # Candidate 1 is the repo root, which is /evalbench in the container, so the
+    # deployed image resolves it to the same /evalbench/results the runs write to.
+    candidates = [
+        "/tmp_session_files/results",
+        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
+        os.path.join(os.getcwd(), "results"),
+    ]
+
+    for candidate in candidates:
+        if os.path.isdir(candidate):
+            return candidate
+
+    return candidates[1]

viewer/precompute_dataset_quality.py

@@ -13,6 +13,8 @@
 import logging
 import os
 
+from paths import get_results_dir
+
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 CACHE_FILENAME = "dataset_quality_cache.json"
@@ -31,26 +33,6 @@
 csv.field_size_limit(10**9)
 
 
-def get_results_dir():
-    # Try to read from environment variable
-    res_dir = os.environ.get("RESULTS_DIR")
-    if res_dir:
-        return res_dir
-
-    # Check multiple locations for results directory
-    results_dir_candidates = [
-        "/tmp_session_files/results",
-        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
-        os.path.join(os.getcwd(), "results"),
-    ]
-
-    for candidate in results_dir_candidates:
-        if os.path.exists(candidate) and os.path.isdir(candidate):
-            return candidate
-
-    return results_dir_candidates[1]  # Fallback to default
-
-
 def _read_configs(configs_file):
     """Return {config_key: value} plus the run_time carried on every row."""
     values = {}

viewer/precompute_trends.py

@@ -4,6 +4,8 @@
 import argparse
 import pandas as pd
 
+from paths import get_results_dir
+
 logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
 
 BATCH_SIZE = int(os.environ.get("PRECOMPUTE_BATCH_SIZE", 50))
@@ -15,26 +17,6 @@
 MAX_WORKERS = int(os.environ.get("PRECOMPUTE_WORKERS", 50))
 
 
-def get_results_dir():
-    # Try to read from environment variable
-    res_dir = os.environ.get("RESULTS_DIR")
-    if res_dir:
-        return res_dir
-
-    # Check multiple locations for results directory
-    results_dir_candidates = [
-        "/tmp_session_files/results",
-        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
-        os.path.join(os.getcwd(), "results"),
-    ]
-
-    for candidate in results_dir_candidates:
-        if os.path.exists(candidate) and os.path.isdir(candidate):
-            return candidate
-
-    return results_dir_candidates[1]  # Fallback to default
-
-
 def process_directory(d, results_dir):
     run_dir = os.path.join(results_dir, d)
     configs_file = os.path.join(run_dir, "configs.csv")

viewer/trends.py

@@ -3,25 +3,8 @@
 import mesop as me
 import pandas as pd
 from main import State, list_run_dirs, load_trends_df
+from paths import get_results_dir
 
-def get_results_dir():
-    # Try to read from environment variable
-    res_dir = os.environ.get("RESULTS_DIR")
-    if res_dir:
-        return res_dir
-
-    # Check multiple locations for results directory
-    results_dir_candidates = [
-        "/tmp_session_files/results",
-        os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
-        os.path.join(os.getcwd(), "results"),
-    ]
-
-    for candidate in results_dir_candidates:
-        if os.path.exists(candidate) and os.path.isdir(candidate):
-            return candidate
-
-    return results_dir_candidates[1]  # Fallback to default
 
 def generate_d3_chart(df, x_col, y_col, hue_col, title, ylabel):
     # job_id is unplotted but read by the tooltip in chart.js.