Pdf reflow rework (#51)

* perf(pdf): native Pdfium-based reflow engine with auto-open

Migrate PDF-to-Markdown reflow generation from PDFBox to a custom
native implementation using Pdfium. This transition improves
processing speed by ~10x and significantly reduces memory overhead.

- Native JNI Bridge: Implemented `pdfium_bridge.cpp` using `dlopen`
  to hook into the existing `libpdfium.so` memory space.
- Optimized Extraction: Replaced character-by-character JNI calls
  with bulk array retrieval (`getPageFontSizes`, `getPageFontWeights`),
  drastically reducing JNI boundary overhead.
- Enhanced Accuracy: Improved Markdown formatting logic by using
  native font weight (bold) and relative font size variance (headers).
- Thread Safety: Refactored generator to process pages sequentially
  while synchronized with the global `PdfiumCore.lock` to ensure
  stability across concurrent UI operations.
- Seamless UX: Implemented a reactive auto-open system in the
  PDF viewer that tracks user intent and navigates to the reflow
  view immediately upon background task completion.

* Improved PDF to Markdown conversion and added cache cleanup.

- Implemented automated cleanup of imported file caches when deleting books.
- Enhanced `PdfToMarkdownGenerator` with support for font flags (italics), improved kerning, and smarter paragraph wrapping.
- Updated `NativePdfiumBridge` and C++ JNI code to extract font information flags from PDFium.

* Implemented seamless file switching and enhanced reflow transition logic.

Key changes include:
- Added `switchToFileSeamlessly` to `MainViewModel` to handle state transitions and navigation when switching between PDF and reflowed text views.
- Updated `generateAndImportReflowFile` to support automatic opening of the generated file at a specific page/chapter.
- Integrated `NavigationEvent` and `CompletableDeferred` to manage asynchronous navigation and state updates during file switches.
- Modified `EpubReaderScreen` and `PdfViewerScreen` to pass the current position when toggling between PDF and text modes.
- Updated `AppNavigation` to move navigation logic out of the `NavHost` and added loading overlays to viewers to improve UI feedback during transitions.

app/src/main/java/com/aryan/reader/pdf/NativePdfiumBridge.kt

@@ -0,0 +1,14 @@
+package com.aryan.reader.pdf
+
+object NativePdfiumBridge {
+    init {
+        System.loadLibrary("native-lib")
+    }
+
+    @JvmStatic external fun getFontSize(textPagePtr: Long, index: Int): Double
+    @JvmStatic external fun getFontWeight(textPagePtr: Long, index: Int): Int
+
+    @JvmStatic external fun getPageFontSizes(textPagePtr: Long, count: Int): FloatArray?
+    @JvmStatic external fun getPageFontWeights(textPagePtr: Long, count: Int): IntArray?
+    @JvmStatic external fun getPageFontFlags(textPagePtr: Long, count: Int): IntArray?
+}

app/src/main/java/com/aryan/reader/pdf/PdfToMarkdownGenerator.kt

@@ -3,11 +3,9 @@ package com.aryan.reader.pdf
 
 import android.content.Context
 import android.net.Uri
-import com.tom_roush.pdfbox.io.MemoryUsageSetting
-import com.tom_roush.pdfbox.pdmodel.PDDocument
-import com.tom_roush.pdfbox.pdmodel.PDPage
-import com.tom_roush.pdfbox.text.PDFTextStripper
-import com.tom_roush.pdfbox.text.TextPosition
+import io.legere.pdfiumandroid.PdfiumCore
+import io.legere.pdfiumandroid.suspend.PdfiumCoreKt
+import io.legere.pdfiumandroid.suspend.PdfDocumentKt
 import kotlinx.coroutines.Dispatchers
 import kotlinx.coroutines.withContext
 import timber.log.Timber
@@ -15,8 +13,6 @@ import java.io.File
 import kotlin.math.roundToInt
 
 object PdfToMarkdownGenerator {
-
-    // Unique delimiter to split pages reliably
     const val PAGE_DELIMITER = "\n\n[[PAGE_BREAK]]\n\n"
 
     suspend fun generateMarkdownFile(
@@ -26,109 +22,252 @@ object PdfToMarkdownGenerator {
         startPage: Int = 1,
         onProgress: (Float) -> Unit
     ): Boolean = withContext(Dispatchers.IO) {
+        val methodStartTime = System.currentTimeMillis()
+        Timber.tag("PdfToMdPerf").d("generateMarkdownFile NATIVE START | uri=$pdfUri | startPage=$startPage")
+
+        val pdfiumCore = PdfiumCoreKt(Dispatchers.Default)
+        val pfd = context.contentResolver.openFileDescriptor(pdfUri, "r")
+        if (pfd == null) {
+            Timber.tag("PdfToMdPerf").e("Failed to open ParcelFileDescriptor")
+            return@withContext false
+        }
+
         try {
-            context.contentResolver.openInputStream(pdfUri)?.use { inputStream ->
-                // Setup mixed memory usage to handle larger files without OOM
-                PDDocument.load(inputStream, MemoryUsageSetting.setupMixed(50 * 1024 * 1024)).use { doc ->
-                    val totalPages = doc.numberOfPages
+            val doc = pdfiumCore.newDocument(pfd)
+            val totalPages = doc.getPageCount()
+            Timber.tag("PdfToMdPerf").d("Document loaded natively. Total Pages: $totalPages")
 
-                    // Configure stripper for linear processing
-                    val stripper = MarkdownStripper(totalPages, onProgress)
-                    stripper.startPage = startPage
-                    stripper.endPage = totalPages
+            destFile.bufferedWriter().use { writer ->
+                for (pageIdx in (startPage - 1) until totalPages) {
+                    val pageMd = extractPageMarkdown(doc, pageIdx)
+                    writer.write(pageMd)
+                    writer.write(PAGE_DELIMITER)
 
-                    // Write directly to file stream (O(N) complexity)
-                    destFile.bufferedWriter().use { writer ->
-                        stripper.writeText(doc, writer)
+                    if (pageIdx % 5 == 0 || pageIdx == totalPages - 1) {
+                        onProgress((pageIdx + 1).toFloat() / totalPages.toFloat())
                     }
                 }
             }
+
+            doc.close()
+            pfd.close()
+
+            Timber.tag("PdfToMdPerf").d("generateMarkdownFile NATIVE SUCCESS | totalTime=${System.currentTimeMillis() - methodStartTime}ms")
             return@withContext true
         } catch (e: Exception) {
-            Timber.e(e, "Failed to generate Markdown from PDF")
+            Timber.e(e, "Failed to generate Markdown from PDF natively")
+            pfd.close()
             return@withContext false
         }
     }
 
-    private class MarkdownStripper(
-        private val totalPages: Int,
-        private val onProgress: (Float) -> Unit
-    ) : PDFTextStripper() {
-        private var currentPageBaseFontSize = 0f
+    private suspend fun extractPageMarkdown(doc: PdfDocumentKt, pageIdx: Int): String {
+        return try {
+            doc.openPage(pageIdx).use { page ->
+                page.openTextPage().use { textPage ->
+                    val charCount = textPage.textPageCountChars()
+                    if (charCount <= 0) return@use ""
 
-        init {
-            sortByPosition = true
-            suppressDuplicateOverlappingText = true
-            paragraphStart = ""
-            paragraphEnd = "\n\n"
-        }
+                    val text = textPage.textPageGetText(0, charCount) ?: ""
+                    val actualCount = minOf(charCount, text.length)
 
-        // Override endPage to update progress and insert delimiter
-        override fun endPage(page: PDPage?) {
-            super.endPage(page)
+                    val rawPtr = textPage.page.pagePtr
 
-            try {
-                // Insert our custom delimiter so importer can split chapters
-                output.write(PAGE_DELIMITER)
+                    val sizes: FloatArray?
+                    val weights: IntArray?
+                    val flags: IntArray?
 
-                // Update progress
-                val current = currentPageNo // inherited from PDFTextStripper
-                if (totalPages > 0) {
-                    onProgress(current.toFloat() / totalPages.toFloat())
+                    synchronized(PdfiumCore.lock) {
+                        sizes = NativePdfiumBridge.getPageFontSizes(rawPtr, actualCount)
+                        weights = NativePdfiumBridge.getPageFontWeights(rawPtr, actualCount)
+                        flags = NativePdfiumBridge.getPageFontFlags(rawPtr, actualCount)
+                    }
+
+                    if (sizes == null || weights == null || flags == null) {
+                        return@use text
+                    }
+
+                    buildMarkdown(text, sizes, weights, flags, actualCount)
                 }
-            } catch (e: Exception) {
-                Timber.e(e, "Error writing page delimiter")
             }
-        }
-
-        override fun startPage(page: PDPage?) {
-            currentPageBaseFontSize = 0f
-            super.startPage(page)
-        }
-
-        private fun calculateBaseFontSize(textPositions: List<TextPosition>) {
-            val sizeCounts = mutableMapOf<Float, Int>()
-            textPositions.forEach { pos ->
-                val size = pos.fontSizeInPt.roundToInt().toFloat()
-                sizeCounts[size] = (sizeCounts[size] ?: 0) + 1
-            }
-            currentPageBaseFontSize = sizeCounts.maxByOrNull { it.value }?.key ?: 12f
-        }
-
-        override fun writeString(text: String?, textPositions: MutableList<TextPosition>?) {
-            if (text.isNullOrBlank() || textPositions.isNullOrEmpty()) return
-
-            if (currentPageBaseFontSize == 0f) {
-                calculateBaseFontSize(textPositions)
-            }
-
-            val firstPos = textPositions[0]
-            val fontSize = firstPos.fontSizeInPt
-            val fontDescriptor = firstPos.font?.fontDescriptor
-
-            val isBold = fontDescriptor?.isForceBold == true ||
-                    (firstPos.font?.name?.contains("Bold", ignoreCase = true) == true)
-            val isItalic = fontDescriptor?.isItalic == true ||
-                    (firstPos.font?.name?.contains("Italic", ignoreCase = true) == true)
-
-            // Header detection logic
-            val isHeader = fontSize > currentPageBaseFontSize * 1.2
-            val isBigHeader = fontSize > currentPageBaseFontSize * 1.5
-
-            val sb = StringBuilder()
-
-            if (isBigHeader) sb.append("## ")
-            else if (isHeader) sb.append("### ")
-
-            if (isBold && !isHeader) sb.append("**")
-            if (isItalic) sb.append("*")
-
-            text.forEach { char -> sb.append(char) }
-
-            if (isItalic) sb.append("*")
-            if (isBold && !isHeader) sb.append("**")
-
-            writeString(sb.toString())
+        } catch (e: Exception) {
+            Timber.w(e, "Error extracting page $pageIdx")
+            ""
         }
     }
+
+    private data class TextSpan(
+        val text: String,
+        val size: Float,
+        val isBold: Boolean,
+        val isItalic: Boolean
+    )
+
+    private data class TextLine(
+        val spans: List<TextSpan>
+    )
+
+    private fun fixKerning(text: String): String {
+        val pattern = Regex("\\b(?:[A-Za-z0-9] ){2,}[A-Za-z0-9]\\b")
+        return pattern.replace(text) { matchResult ->
+            matchResult.value.replace(" ", "")
+        }
+    }
+
+    private fun buildMarkdown(text: String, sizes: FloatArray, weights: IntArray, flags: IntArray, count: Int): String {
+        if (count == 0) return ""
+
+        val sizeFrequency = HashMap<Int, Int>()
+        for (i in 0 until count) {
+            val s = sizes[i].roundToInt()
+            sizeFrequency[s] = (sizeFrequency[s] ?: 0) + 1
+        }
+        val baseSize = sizeFrequency.maxByOrNull { it.value }?.key ?: 12
+
+        val lines = mutableListOf<TextLine>()
+        @Suppress("CanBeVal") var currentSpans = mutableListOf<TextSpan>()
+        val currentSpanText = StringBuilder()
+
+        var currentSize = -1f
+        var currentBold = false
+        var currentItalic = false
+
+        for (i in 0 until count) {
+            val c = text[i]
+            if (c == '\u0000') continue
+
+            if (c == '\n' || c == '\r') {
+                if (c == '\n' && i > 0 && text[i - 1] == '\r') continue
+
+                if (currentSpanText.isNotEmpty()) {
+                    currentSpans.add(TextSpan(currentSpanText.toString(), currentSize, currentBold, currentItalic))
+                    currentSpanText.clear()
+                }
+                lines.add(TextLine(currentSpans.toList()))
+                currentSpans.clear()
+                continue
+            }
+
+            val isSpace = c.isWhitespace()
+            val size = sizes[i]
+            val bold = weights[i] > 600
+            val italic = (flags[i] and 64) != 0
+
+            if (currentSpanText.isEmpty()) {
+                currentSize = size
+                currentBold = bold
+                currentItalic = italic
+                currentSpanText.append(c)
+            } else {
+                if (!isSpace && (currentSize != size || currentBold != bold || currentItalic != italic)) {
+                    currentSpans.add(TextSpan(currentSpanText.toString(), currentSize, currentBold, currentItalic))
+                    currentSpanText.clear()
+                    currentSize = size
+                    currentBold = bold
+                    currentItalic = italic
+                }
+                currentSpanText.append(c)
+            }
+        }
+
+        if (currentSpanText.isNotEmpty()) {
+            currentSpans.add(TextSpan(currentSpanText.toString(), currentSize, currentBold, currentItalic))
+        }
+        if (currentSpans.isNotEmpty()) {
+            lines.add(TextLine(currentSpans))
+        }
+
+        val validLines = lines.filter { it.spans.isNotEmpty() }
+        val lineLengths = validLines.map { line -> line.spans.sumOf { it.text.length } }.filter { it > 10 }.sorted()
+
+        val typicalLineLen = if (lineLengths.isNotEmpty()) {
+            lineLengths[(lineLengths.size * 0.8).toInt().coerceAtMost(lineLengths.size - 1)]
+        } else {
+            80
+        }
+
+        val wrapThreshold = (typicalLineLen * 0.85).toInt()
+
+        val sb = StringBuilder()
+
+        for (i in lines.indices) {
+            val line = lines[i]
+            if (line.spans.isEmpty()) {
+                sb.append("\n")
+                continue
+            }
+
+            val maxFontSize = line.spans.filter { it.text.isNotBlank() }.maxOfOrNull { it.size } ?: baseSize.toFloat()
+            val charBigHeader = maxFontSize > baseSize * 1.5f
+            val charHeader = maxFontSize > baseSize * 1.2f
+
+            var prefix = ""
+            if (charBigHeader) prefix = "## "
+            else if (charHeader) prefix = "### "
+
+            val rawLineText = line.spans.joinToString("") { it.text }
+            val trimmedRaw = rawLineText.trim()
+            val lineLen = trimmedRaw.length
+
+            val isList = trimmedRaw.startsWith("•") ||
+                    trimmedRaw.startsWith("- ") ||
+                    trimmedRaw.startsWith("▪") ||
+                    trimmedRaw.matches(Regex("^[0-9]+\\.\\s.*")) ||
+                    trimmedRaw.matches(Regex("^[a-zA-Z]\\)\\s.*"))
+
+            if (prefix.isNotEmpty() && !isList) {
+                sb.append(prefix)
+            }
+
+            for (span in line.spans) {
+                var spanText = span.text
+                spanText = fixKerning(spanText)
+
+                val leadingSpaces = spanText.takeWhile { it.isWhitespace() }
+                val trailingSpaces = spanText.takeLastWhile { it.isWhitespace() }
+                val trimmedText = spanText.trim()
+
+                if (trimmedText.isEmpty()) {
+                    sb.append(spanText)
+                    continue
+                }
+
+                sb.append(leadingSpaces)
+
+                var tag = ""
+                if (span.isBold && span.isItalic) tag = "***"
+                else if (span.isBold) tag = "**"
+                else if (span.isItalic) tag = "*"
+
+                sb.append(tag).append(trimmedText).append(tag)
+                sb.append(trailingSpaces)
+            }
+
+            var isParagraphBreak = false
+
+            if (prefix.isNotEmpty() || isList) {
+                isParagraphBreak = true
+            } else if (lineLen < wrapThreshold) {
+                isParagraphBreak = true
+            } else if (trimmedRaw.matches(Regex(".*[.!?\"'”’;:*]$"))) {
+                isParagraphBreak = true
+            } else {
+                val nextLine = lines.subList(i + 1, lines.size).firstOrNull { it.spans.isNotEmpty() }
+                if (nextLine != null) {
+                    val nextRaw = nextLine.spans.joinToString("") { it.text }.trimStart()
+                    if (nextRaw.startsWith("\"") || nextRaw.startsWith("“") || nextRaw.startsWith("-")) {
+                        isParagraphBreak = true
+                    }
+                }
+            }
+
+            if (isParagraphBreak) {
+                sb.append("\n\n")
+            } else {
+                sb.append("\n")
+            }
+        }
+
+        return sb.toString().replace(Regex("\\n{3,}"), "\n\n").trim()
+    }
 }

app/src/main/java/com/aryan/reader/pdf/PdfViewerScreen.kt

@@ -27,11 +27,7 @@ import android.annotation.SuppressLint
 import android.app.Activity
 import android.content.Context
 import android.content.pm.PackageManager
-import androidx.compose.foundation.gestures.awaitEachGesture
-import androidx.compose.foundation.gestures.awaitFirstDown
-import androidx.compose.foundation.gestures.waitForUpOrCancellation
 import android.graphics.Bitmap
-import kotlin.math.max
 import android.graphics.RectF
 import android.net.Uri
 import android.os.Build
@@ -59,8 +55,11 @@ import androidx.compose.foundation.Image
 import androidx.compose.foundation.background
 import androidx.compose.foundation.border
 import androidx.compose.foundation.clickable
+import androidx.compose.foundation.gestures.awaitEachGesture
+import androidx.compose.foundation.gestures.awaitFirstDown
 import androidx.compose.foundation.gestures.detectDragGestures
 import androidx.compose.foundation.gestures.detectDragGesturesAfterLongPress
+import androidx.compose.foundation.gestures.waitForUpOrCancellation
 import androidx.compose.foundation.interaction.MutableInteractionSource
 import androidx.compose.foundation.layout.Arrangement
 import androidx.compose.foundation.layout.Box
@@ -137,10 +136,8 @@ import androidx.compose.material3.OutlinedTextField
 import androidx.compose.material3.RadioButton
 import androidx.compose.material3.Scaffold
 import androidx.compose.material3.Slider
-import androidx.compose.material3.SnackbarDuration
 import androidx.compose.material3.SnackbarHost
 import androidx.compose.material3.SnackbarHostState
-import androidx.compose.material3.SnackbarResult
 import androidx.compose.material3.Surface
 import androidx.compose.material3.Tab
 import androidx.compose.material3.TabRow
@@ -286,6 +283,7 @@ import java.io.ByteArrayOutputStream
 import java.io.File
 import java.net.HttpURLConnection
 import java.net.URL
+import kotlin.math.max
 import kotlin.math.min
 import kotlin.math.roundToInt
 import kotlin.random.Random
@@ -1845,26 +1843,6 @@ fun PdfViewerScreen(
         onToggleBookmark(currentPage)
     }
 
-    LaunchedEffect(reflowInfo) {
-        if (reflowInfo?.state == WorkInfo.State.SUCCEEDED &&
-            reflowInfo?.tags?.contains("book_$bookId") == true) {
-
-            val result = snackbarHostState.showSnackbar(
-                message = "Text View generation complete!",
-                actionLabel = "OPEN",
-                duration = SnackbarDuration.Long
-            )
-
-            if (result == SnackbarResult.ActionPerformed) {
-                snackbarHostState.currentSnackbarData?.dismiss()
-                val item = uiState.recentFiles.find { it.bookId == reflowBookId }
-                if (item != null) {
-                    viewModel.onRecentFileClicked(item)
-                }
-            }
-        }
-    }
-
     LaunchedEffect(pdfUri) { debugPdfLinks(context, pdfUri, pdfiumCore, this) }
 
     LaunchedEffect(currentBookId) {
@@ -4979,13 +4957,14 @@ fun PdfViewerScreen(
                                                 if (hasReflowFile) {
                                                     val item = uiState.recentFiles.find { it.bookId == reflowBookId }
                                                     if (item != null) {
-                                                        viewModel.onRecentFileClicked(item)
+                                                        viewModel.switchToFileSeamlessly(item, currentPage)
                                                     }
                                                 } else {
                                                     viewModel.generateAndImportReflowFile(
                                                         pdfBookId = bookId,
                                                         pdfUri = pdfUri,
-                                                        originalTitle = originalFileName
+                                                        originalTitle = originalFileName,
+                                                        autoOpenPage = currentPage
                                                     )
                                                 }
                                             },

app/src/main/java/com/aryan/reader/pdf/ReflowWorker.kt

@@ -20,29 +20,49 @@ class ReflowWorker(
 ) : CoroutineWorker(context, params) {
 
     override suspend fun doWork(): Result = withContext(Dispatchers.IO) {
-        val bookId = inputData.getString(KEY_BOOK_ID) ?: return@withContext Result.failure()
-        val pdfUriString = inputData.getString(KEY_PDF_URI) ?: return@withContext Result.failure()
+        val workStartTime = System.currentTimeMillis()
+        Timber.tag("PdfToMdPerf").d("=== ReflowWorker START ===")
+
+        val bookId = inputData.getString(KEY_BOOK_ID) ?: run {
+            Timber.tag("PdfToMdPerf").e("FAILURE: KEY_BOOK_ID is null")
+            return@withContext Result.failure()
+        }
+
+        val pdfUriString = inputData.getString(KEY_PDF_URI) ?: run {
+            Timber.tag("PdfToMdPerf").e("FAILURE: KEY_PDF_URI is null | bookId=$bookId")
+            return@withContext Result.failure()
+        }
         val originalTitle = inputData.getString(KEY_ORIGINAL_TITLE) ?: "Document"
         val reflowBookId = "${bookId}_reflow"
 
+        Timber.tag("PdfToMdPerf").d("Input data | bookId=$bookId | reflowBookId=$reflowBookId | pdfUri=$pdfUriString | originalTitle=$originalTitle")
+
         val destFile = File(applicationContext.filesDir, "${bookId}_reflow.md")
         val pdfUri = pdfUriString.toUri()
 
-        Timber.tag("ReflowWorker").d("Starting background reflow for $originalTitle.")
+        Timber.tag("PdfToMdPerf").d("Dest file path: ${destFile.absolutePath} | exists=${destFile.exists()}")
+        Timber.tag("PdfToMdPerf").d("Starting PdfToMarkdownGenerator.generateMarkdownFile...")
+        val genStartTime = System.currentTimeMillis()
 
-        // Delegate entire process to Generator (it now handles the loop and progress)
         val success = PdfToMarkdownGenerator.generateMarkdownFile(
             applicationContext,
             pdfUri,
             destFile,
-            startPage = 1 // Always start from beginning for full regeneration
+            startPage = 1
         ) { progress ->
-            // Report progress
+            if ((progress * 10).toInt() % 1 == 0) {
+                Timber.tag("PdfToMdPerf").d("Progress: ${(progress * 100).toInt()}%")
+            }
             setProgressAsync(workDataOf(KEY_PROGRESS to progress))
         }
 
+        Timber.tag("PdfToMdPerf").d("generateMarkdownFile completed | success=$success | time=${System.currentTimeMillis() - genStartTime}ms")
+
         if (success && destFile.exists()) {
-            Timber.tag("ReflowWorker").d("Reflow complete. Importing to database.")
+            val fileSizeKB = destFile.length() / 1024
+            Timber.tag("PdfToMdPerf").d("Reflow SUCCESS | outputFileSize=${fileSizeKB}KB")
+            Timber.tag("PdfToMdPerf").d("Starting database import...")
+            val dbStartTime = System.currentTimeMillis()
 
             val repo = RecentFilesRepository(applicationContext)
 
@@ -63,13 +83,16 @@ class ReflowWorker(
             )
 
             repo.addRecentFile(newItem)
+            Timber.tag("PdfToMdPerf").d("Database import completed in ${System.currentTimeMillis() - dbStartTime}ms")
 
-            // 100% Progress
             setProgressAsync(workDataOf(KEY_PROGRESS to 1.0f))
 
+            val totalTime = System.currentTimeMillis() - workStartTime
+            Timber.tag("PdfToMdPerf").d("=== ReflowWorker SUCCESS === | totalTime=${totalTime}ms | totalTimeSec=${totalTime / 1000}s")
             return@withContext Result.success()
         } else {
-            Timber.e("Reflow failed or was incomplete.")
+            val totalTime = System.currentTimeMillis() - workStartTime
+            Timber.tag("PdfToMdPerf").e("=== ReflowWorker FAILURE === | success=$success | fileExists=${destFile.exists()} | totalTime=${totalTime}ms")
             return@withContext Result.failure()
         }
     }