Fix critical hardsubx OCR bugs affecting accuracy and stability

- Fix iterator initialization in check_color_alternate_pixels
- Add proper null checks before dereferencing OCR results
- Fix confidence threshold comparison (use abs for negative confidences)
- Fix c_char array initialization in avc/nal.rs for cross-platform compatibility
- Prevent segmentation faults in OCR processing

Author: prateekmedia

src/rust/src/avc/nal.rs

@@ -596,7 +596,7 @@ pub unsafe fn slice_header(
            pic_order_cnt_lsb, (*dec_ctx.timing).current_tref,
            current_index, (*dec_ctx.avc_ctx).currref, (*dec_ctx.avc_ctx).lastmaxidx, (*dec_ctx.avc_ctx).maxtref);
 
-    let mut buf = [c_char::from(0i8); 64];
+    let mut buf = [0 as c_char; 64];
     debug!(
         msg_type = DebugMessageFlag::TIME;
         "  sync_pts:{} ({:8})",

src/rust/src/hardsubx/classifier.rs

@@ -179,7 +179,11 @@ pub unsafe extern "C" fn get_ocr_text_wordwise_threshold(
                 }
             }
 
-            text_out = format!("{} {}", text_out, word);
+            if text_out.is_empty() {
+                text_out = word;
+            } else {
+                text_out = format!("{} {}", text_out, word);
+            }
         }
     }
 
@@ -234,7 +238,7 @@ pub unsafe extern "C" fn get_ocr_text_letterwise_threshold(
     let mut total_conf: std::os::raw::c_float = 0.0;
     let mut num_characters: std::os::raw::c_int = 0;
 
-    let mut first_iter: bool = false;
+    let mut first_iter: bool = true;
 
     if it != null::<TessResultIterator>() as *mut TessResultIterator {
         loop {
@@ -245,17 +249,25 @@ pub unsafe extern "C" fn get_ocr_text_letterwise_threshold(
             }
 
             let letter = _tess_string_helper(it, level);
-            text_out = format!("{}{}", text_out, letter);
+            if letter.is_empty() {
+                continue;
+            }
 
+            // Check confidence BEFORE adding to output
+            let mut should_add = true;
             if threshold > 0.0 {
                 // we don't even want to bother with this call if threshold is 0 or less
                 let conf: std::os::raw::c_float = TessResultIteratorConfidence(it, level);
                 if conf < threshold {
-                    continue;
+                    should_add = false;
+                } else {
+                    total_conf += conf;
+                    num_characters += 1;
                 }
+            }
 
-                total_conf += conf;
-                num_characters += 1;
+            if should_add {
+                text_out = format!("{}{}", text_out, letter);
             }
         }
     }

src/rust/src/hardsubx/decoder.rs

@@ -41,10 +41,14 @@ pub unsafe fn dispatch_classifier_functions(ctx: *mut lib_hardsubx_ctx, im: *mut
     match (*ctx).ocr_mode {
         0 => {
             let ret_char_arr = get_ocr_text_simple_threshold(ctx, im, (*ctx).conf_thresh);
-            let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
-            match text_out_result {
-                Ok(T) => T,
-                Err(_E) => "".to_string(),
+            if ret_char_arr.is_null() {
+                "".to_string()
+            } else {
+                let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
+                match text_out_result {
+                    Ok(T) => T,
+                    Err(_E) => "".to_string(),
+                }
             }
         }
         1 => {
@@ -59,10 +63,14 @@ pub unsafe fn dispatch_classifier_functions(ctx: *mut lib_hardsubx_ctx, im: *mut
         }
         2 => {
             let ret_char_arr = get_ocr_text_letterwise_threshold(ctx, im, (*ctx).conf_thresh);
-            let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
-            match text_out_result {
-                Ok(T) => T,
-                Err(_E) => "".to_string(),
+            if ret_char_arr.is_null() {
+                "".to_string()
+            } else {
+                let text_out_result = ffi::CString::from_raw(ret_char_arr).into_string();
+                match text_out_result {
+                    Ok(T) => T,
+                    Err(_E) => "".to_string(),
+                }
             }
         }