TF-4265 Add exclude list support to AttachmentTextDetector

This commit is contained in:
dab246
2026-01-30 13:16:17 +07:00
committed by Dat H. Pham
parent 772a5e3219
commit 1071bb49cf
7 changed files with 330 additions and 85 deletions
@@ -886,7 +886,7 @@ class ComposerController extends BaseController
_sendButtonState = ButtonState.enabled;
return;
}
final attachmentKeywords = validateAttachmentReminder(
final attachmentKeywords = await validateAttachmentReminder(
emailSubject: subjectEmail.value ?? '',
emailContent: emailContent,
);
@@ -9,14 +9,15 @@ import 'package:tmail_ui_user/main/routes/route_navigation.dart';
extension AttachmentDetectionExtension on ComposerController {
List<String> validateAttachmentReminder({
Future<List<String>> validateAttachmentReminder({
required String emailContent,
required String emailSubject,
}) {
}) async {
try {
final fullContent = '$emailSubject $emailContent';
final plainText = HtmlUtils.extractPlainText(fullContent);
final keywords = AttachmentTextDetector.matchedKeywordsUnique(plainText);
final keywords =
await AttachmentTextDetector.matchedKeywordsUnique(plainText);
if (keywords.isEmpty) {
return [];
} else {
@@ -1,3 +1,6 @@
import 'package:flutter/foundation.dart';
import 'package:tmail_ui_user/features/composer/presentation/manager/detection_params.dart';
import 'package:tmail_ui_user/features/composer/presentation/manager/keyword_filter.dart';
import 'package:tmail_ui_user/main/localizations/language_code_constants.dart';
class AttachmentTextDetector {
@@ -47,6 +50,22 @@ class AttachmentTextDetector {
],
};
static RegExp? _combinedRegExp;
/// SYNC/ASYNC EXECUTION THRESHOLD
///
/// Value: 20,000 characters (approx. 4-5 pages of text).
///
/// Technical Rationale:
/// 1. Frame Budget (16ms): To maintain 60fps, operations on the UI thread must stay under 16ms.
/// 2. Isolate Overhead: `compute` takes about 2-4ms just to copy data and start the thread.
/// 3. Real-world Benchmark:
/// - For < 20k chars, Sync execution is usually faster (< 10ms) than the Async overhead.
/// - For > 20k chars, Regex processing might exceed 16ms on low-end devices, causing jank.
///
/// -> Below 20k: Run Sync (Instant response).
/// -> Above 20k: Run Async (Safety for UI).
static const int _kAsyncExecutionThreshold = 20000;
/// Detect if the text contains keywords suggesting there is an attachment.
/// [lang] is the language code (`en`, `fr`, `ru`, `vi`, `ar`, ...).
static bool containsAttachmentKeyword(String text, {required String lang}) {
@@ -95,48 +114,85 @@ class AttachmentTextDetector {
return result;
}
// Store a single combined RegExp pattern
static RegExp? _combinedRegExp;
/// Initializes a single optimized RegExp that contains all keywords.
/// Initializes the single optimized RegExp pattern.
static void _initializeRegExp() {
if (_combinedRegExp != null) return;
// Flatten all keywords from all languages into a single list
final allKeywords = _keywordsByLang.values.expand((list) => list).toSet().toList();
final allKeywords =
_keywordsByLang.values.expand((l) => l).toSet().toList();
// Sort keywords by length descending (longest first).
// Sort by length descending (Longest first).
// Crucial to ensure "attachment" is matched before "attach".
allKeywords.sort((a, b) => b.length.compareTo(a.length));
// Create a pattern like: (attachment|attach|file|files)(?![\p{L}])
// Join all keywords with OR operator (|)
final patternString = allKeywords.map(RegExp.escape).join('|');
// LOGIC:
// 1. RegExp.escape: Neutralizes special characters.
// 2. No Nested Quantifiers: Prevents catastrophic backtracking.
// 3. Linear Scan: Uses logical OR (|) for a single-pass scan.
final pattern = allKeywords.map(RegExp.escape).join('|');
// Pattern: (keyword1|keyword2)(?![\p{L}])
// (?![\p{L}]): Negative lookahead to ensure the next char is NOT a letter.
// Allows: numbers (file123), punctuation (file.), spaces (file ).
// Rejects: letter extensions (filetage).
_combinedRegExp = RegExp(
'($patternString)(?![\\p{L}])',
'($pattern)(?![\\p{L}])',
unicode: true,
caseSensitive: false, // Let Regex handle case-insensitivity
caseSensitive: false,
);
}
/// Detects unique keywords in the provided text based on predefined language lists.
static List<String> matchedKeywordsUnique(String text) {
if (text.isEmpty) return [];
// Ensure RegExp are initialized
static List<String> _processMatchedKeywordsInIsolate(DetectionParams params) {
// Re-initialize RegExp (Isolates do not share static memory with the main app)
_initializeRegExp();
final text = params.text;
final filters = params.filters;
if (text.isEmpty) return [];
final matches = _combinedRegExp!.allMatches(text);
final result = <String>{};
for (final match in matches) {
// match.group(0) returns the actual text found (e.g., "FILE", "File").
// We convert it to lowercase to normalize the result list.
if (match.group(0) != null) {
result.add(match.group(0)!.toLowerCase());
final keyword = match.group(0)!.toLowerCase();
// Apply Filter Pipeline
bool isAccepted = true;
for (final filter in filters) {
if (!filter.isValid(text, match)) {
isAccepted = false;
break; // Stop checking other filters to save CPU
}
}
if (isAccepted) {
result.add(keyword);
}
}
return result.toList();
}
/// Smart keyword detection (Hybrid Sync/Async).
/// Automatically decides whether to run on the UI Thread or a Background Thread.
/// [forceSync]: If true, it will be MANDATORY to run on the UI Thread (ignoring the threshold).
/// Use this parameter for unit tests to benchmark the algorithm more accurately.
static Future<List<String>> matchedKeywordsUnique(
String text, {
List<KeywordFilter> filters = const [],
bool forceSync = false,
}) async {
// If text is short, run Synchronously to avoid Isolate overhead cost.
if (forceSync || text.length < _kAsyncExecutionThreshold) {
return _processMatchedKeywordsInIsolate(
DetectionParams(text: text, filters: filters),
);
} else { // If text is long, use `compute` to offload to an Isolate, preventing UI freeze.
return await compute(
_processMatchedKeywordsInIsolate,
DetectionParams(text: text, filters: filters),
);
}
}
}
@@ -0,0 +1,10 @@
import 'package:tmail_ui_user/features/composer/presentation/manager/keyword_filter.dart';
/// Data Transfer Object used to package data sent to the Isolate.
/// Since Isolates cannot share memory, we need a simple object to pass parameters.
class DetectionParams {
final String text;
final List<KeywordFilter> filters;
DetectionParams({required this.text, required this.filters});
}
@@ -0,0 +1,55 @@
import 'package:tmail_ui_user/features/composer/presentation/manager/keyword_filter.dart';
/// Implementation of Exclude List Filter.
/// Optimized for memory usage by avoiding unnecessary String allocations.
class ExcludeListFilter implements KeywordFilter {
final Set<String> _excludes;
/// Constructor takes a List for convenience, but converts to Set for O(1) lookup.
ExcludeListFilter(List<String> rawExcludes)
: _excludes = rawExcludes.map((e) => e.toLowerCase()).toSet();
/// Static RegExp to check for whitespace.
/// Using a static instance avoids re-compilation in loops.
static final RegExp _whitespaceRegExp = RegExp(r'\s');
@override
bool isValid(String fullText, Match match) {
if (_excludes.isEmpty) return true;
// Extract the full token surrounding the match
final token = _getSurroundingToken(fullText, match.start, match.end);
final lowerToken = token.toLowerCase();
// Exact match with the token (e.g., "file-246")
if (_excludes.contains(lowerToken)) return false;
// Match with trailing punctuation removed (e.g., "file246." -> "file246")
final cleanToken = lowerToken.replaceAll(RegExp(r'[^\w\s]+$'), '');
if (_excludes.contains(cleanToken)) return false;
return true;
}
/// Extracts the surrounding token without creating garbage strings.
/// Instead of using .trim() inside a loop (which allocates memory),
/// we check characters directly.
String _getSurroundingToken(String text, int matchStart, int matchEnd) {
int start = matchStart;
int end = matchEnd;
// Expand to the left until whitespace or start of string
while (start > 0) {
// Check the character before the current start index
if (_whitespaceRegExp.hasMatch(text[start - 1])) break;
start--;
}
// Expand to the right until whitespace or end of string
while (end < text.length) {
if (_whitespaceRegExp.hasMatch(text[end])) break;
end++;
}
return text.substring(start, end);
}
}
@@ -0,0 +1,6 @@
/// Common interface for all keyword filters.
abstract class KeywordFilter {
/// Returns `true` if the keyword is valid.
/// Returns `false` if the keyword should be rejected.
bool isValid(String fullText, Match match);
}