From 1c33fe6e84dd2dc732f2c56064494388bdee7bb2 Mon Sep 17 00:00:00 2001
From: Sam Scholten
Date: Tue, 28 Oct 2025 12:57:55 +1000
Subject: Fix time array monotonicity for very large arrays

- Generate time arrays with float64 precision before converting to float32
- This prevents precision loss that causes non-monotonic time arrays
- Affects datasets with >10 million samples (e.g., 200M points at 20 MHz)
- Changed empty time array warning to raise RuntimeError for better error handling
- Applied fix to both rd() and rd_chunked() functions
---
 src/transivent/io.py | 19 ++++++++++---------
 1 file changed, 10 insertions(+), 9 deletions(-)

(limited to 'src')

diff --git a/src/transivent/io.py b/src/transivent/io.py
index 5eb46f6..b272283 100644
--- a/src/transivent/io.py
+++ b/src/transivent/io.py
@@ -335,16 +335,16 @@ def rd(
     # x = x.astype(np.float32) # NB: data is already in physical units (V)
 
     # Use np.linspace for more robust time array generation
+    # Generate with float64 precision to handle very large arrays, then convert to float32
     num_points = len(x)
-    if num_points > 0:
-        t = np.linspace(0, (num_points - 1) * si, num_points, dtype=np.float32)
-    else:
-        t = np.array([], dtype=np.float32)
-        warnings.warn(
-            f"Generated an empty time array for file {rel_fp}. "
-            f"Length of signal: {len(x)}, sampling interval: {si}. "
-            "This might indicate an issue with input data or sampling interval."
+    if num_points == 0:
+        raise RuntimeError(
+            f"No data points found in file {rel_fp}. "
+            f"This indicates an empty or corrupted binary file."
         )
+    
+    t = np.linspace(0, (num_points - 1) * si, num_points, dtype=np.float64)
+    t = t.astype(np.float32)
 
     return t, x
 
@@ -445,8 +445,9 @@ def rd_chunked(
                     start_time,
                     start_time + (num_points - 1) * si,
                     num_points,
-                    dtype=np.float32,
+                    dtype=np.float64,
                 )
+                t_chunk = t_chunk.astype(np.float32)
 
                 yield t_chunk, x_chunk.astype(np.float32)
 
-- 
cgit v1.2.3