Coverage Summary for Class: CompiledFilters (com.georgeci.moneysurfer.feature.transaction.filter)

 package com.georgeci.moneysurfer.feature.transaction.filter
 
 import com.georgeci.moneysurfer.domain.model.CategorizedTransaction
 import com.georgeci.moneysurfer.domain.model.Transaction
 import com.georgeci.moneysurfer.domain.preferences.TransactionPeriodMode
 import com.georgeci.moneysurfer.domain.primitives.TransactionStatus
 import com.georgeci.moneysurfer.domain.primitives.TransactionType
 import com.georgeci.moneysurfer.domain.util.TransactionPeriodWindow
 import com.georgeci.moneysurfer.domain.util.periodWindow
 import com.georgeci.moneysurfer.domain.util.startOfIsoWeek
 import kotlinx.datetime.DateTimeUnit
 import kotlinx.datetime.LocalDate
 import kotlinx.datetime.minus
 import kotlinx.datetime.plus
 
 private const val MINOR_PER_MAJOR = 100L
 private const val MAX_DECIMALS = 2
 private const val DAYS_PER_WEEK = 7
 
 /** `Money.MAX_MINOR` is 10^15 minor units, i.e. 13 digits of major units. */
 private const val MAX_WHOLE_DIGITS = 13
 
 private const val DECEMBER = 12
 private const val DAYS_IN_DECEMBER = 31
 private const val DECIMAL_RADIX = 10
 
 /**
  * The one date window the list loads, whoever asked for it.
  *
  * [TransactionDateRange.FollowPeriod] defers to the pager ([mode] + [anchor]); every other range
  * takes over completely. The pager is hidden while it is not the source, so the two can never end
  * up as competing windows.
  */
 fun resolveWindow(
     dateRange: TransactionDateRange,
     mode: TransactionPeriodMode,
     anchor: LocalDate,
     today: LocalDate,
 ): TransactionPeriodWindow = when (dateRange) {
     TransactionDateRange.FollowPeriod -> periodWindow(mode, anchor)
     is TransactionDateRange.Preset -> presetWindow(dateRange.preset, today)
     is TransactionDateRange.Custom -> TransactionPeriodWindow(from = dateRange.from, to = dateRange.to)
 }
 
 private fun presetWindow(preset: TransactionDatePreset, today: LocalDate): TransactionPeriodWindow =
     when (preset) {
         TransactionDatePreset.Today -> TransactionPeriodWindow(today, today)
         TransactionDatePreset.Yesterday -> today.minus(1, DateTimeUnit.DAY)
             .let { TransactionPeriodWindow(it, it) }
         TransactionDatePreset.ThisWeek -> today.startOfIsoWeek()
             .let { TransactionPeriodWindow(it, it.plus(DAYS_PER_WEEK - 1, DateTimeUnit.DAY)) }
         TransactionDatePreset.ThisMonth -> periodWindow(TransactionPeriodMode.Month, today)
         TransactionDatePreset.LastMonth -> periodWindow(
             TransactionPeriodMode.Month,
             LocalDate(today.year, today.month, 1).minus(1, DateTimeUnit.MONTH),
         )
         TransactionDatePreset.ThisYear -> TransactionPeriodWindow(
             from = LocalDate(today.year, 1, 1),
             to = LocalDate(today.year, DECEMBER, DAYS_IN_DECEMBER),
         )
     }
 
 /**
  * An amount the user typed, as an inclusive range of minor units.
  *
  * Typed precision is the tolerance: "12" means "twelve-something" (1200..1299), "12.5" means
  * "twelve fifty-something" (1250..1259), "12.50" is exact. Without that rule a search for the
  * amount on a receipt would only ever hit whole-unit transactions, and the min/max fields would
  * exclude a €12.99 charge from a "max 12" bound the user meant as "no more than twelve-ish".
  */
 data class AmountRange(val fromMinor: Long, val toMinor: Long)
 
 /**
  * Parses [raw] as an amount, or returns null when it is not one.
  *
  * Currency symbols, spaces and thousands separators are dropped, and both `.` and `,` read as the
  * decimal separator — a filter field is not the place to be strict about how a number is written.
  * More than [MAX_DECIMALS] fraction digits is rejected rather than rounded: it is not an amount
  * this app can hold, so silently matching a truncated one would be a lie.
  */
 fun parseAmountQuery(raw: String): AmountRange? {
     val normalized = normalizeAmountInput(raw) ?: return null
     val whole = normalized.substringBefore('.')
     val fraction = normalized.substringAfter('.', missingDelimiterValue = "")
     val from = whole.toLong() * MINOR_PER_MAJOR + fraction.padEnd(MAX_DECIMALS, '0').toLong()
     // Everything the untyped digits could still be: "12" covers 12.00 through 12.99.
     val slack = pow10(MAX_DECIMALS - fraction.length) - 1
     return AmountRange(fromMinor = from, toMinor = from + slack)
 }
 
 /**
  * `raw` reduced to `digits[.digits]`, or null when it is not an amount at all. Rejects rather
  * than clamps anything wider than the domain's own cap ([MAX_WHOLE_DIGITS] major units), so the
  * arithmetic above cannot overflow on a pasted wall of digits.
  */
 private fun normalizeAmountInput(raw: String): String? {
     val cleaned = raw.trim()
         .filterNot { it.isWhitespace() || it == '\'' || it == '_' }
         .removePrefix("+")
         .replace(',', '.')
     val whole = cleaned.substringBefore('.')
     val fraction = cleaned.substringAfter('.', missingDelimiterValue = "")
     val wellFormed = cleaned.all { it.isDigit() || it == '.' } &&
         cleaned.count { it == '.' } <= 1 &&
         whole.isNotEmpty() &&
         whole.length <= MAX_WHOLE_DIGITS &&
         fraction.length <= MAX_DECIMALS
     return cleaned.takeIf { wellFormed }
 }
 
 private fun pow10(exponent: Int): Long = generateSequence(1L) { it * DECIMAL_RADIX }.elementAt(exponent)
 
 /**
  * A [TransactionFilters] with its three free-text amount fields parsed exactly once, ready to test
  * many rows.
  *
  * The min/max bounds and the search-amount are constant across a whole page, so parsing them per
  * row (as a naive `matches` loop would) repeats the same parse hundreds of times per rebuild.
  * [TransactionFilters.compile] does it once; the per-row [matches] then only compares.
  */
 class CompiledFilters internal constructor(
     private val filters: TransactionFilters,
     private val minMinor: Long?,
     private val maxMinor: Long?,
     private val queryAmount: AmountRange?,
     private val needle: String,
 ) {
     /**
      * Whether [row] survives these filters.
      *
      * The date window is *not* checked here — it is applied by the query (see [resolveWindow]), so
      * re-checking it would only duplicate a bound the database already honoured.
      *
      * `accountIds` is expected to be empty when the list is already scoped to one account; the
      * ViewModel clears it there, because intersecting a per-account list with a different account
      * would silently show nothing.
      */
     fun matches(row: CategorizedTransaction): Boolean {
         val transaction = row.transaction
         return matchesType(transaction) &&
             (filters.accountIds.isEmpty() || transaction.accountId in filters.accountIds) &&
             (filters.categoryIds.isEmpty() || transaction.categoryId in filters.categoryIds) &&
             matchesAmountBounds(transaction) &&
             (!filters.recurringOnly || transaction.recurringRuleId != null) &&
             (!filters.plannedOnly || transaction.status == TransactionStatus.PLANNED) &&
             matchesQuery(row)
     }
 
     /**
      * The stored type alone cannot answer this: both legs of a transfer are an ordinary
      * `EXPENSE`/`INCOME` pair, so reading only [Transaction.type] would leave `Expenses` showing
      * the outgoing leg the user picked the segment to hide. `transferId` is what separates money
      * that left the books from money that only moved sideways, and it keeps the four segments
      * disjoint.
      */
     private fun matchesType(transaction: Transaction): Boolean {
         val isTransferLeg = transaction.transferId != null
         return when (filters.type) {
             TransactionTypeFilter.All -> true
             TransactionTypeFilter.Expenses -> !isTransferLeg && transaction.type == TransactionType.EXPENSE
             TransactionTypeFilter.Income -> !isTransferLeg && transaction.type == TransactionType.INCOME
             TransactionTypeFilter.Transfer -> isTransferLeg
         }
     }
 
     /**
      * Bounds compare against the *magnitude*: the sign is the type filter's business, and a user
      * asking for "at least 50" means fifty of money, not −50 being below the bound.
      *
      * An unparseable bound is ignored rather than treated as zero, and a trailing separator reads
      * as the digits before it, so neither nonsense nor a half-typed "12." empties the list.
      */
     private fun matchesAmountBounds(transaction: Transaction): Boolean {
         val minor = transaction.money.abs().minor
         return (minMinor == null || minor >= minMinor) && (maxMinor == null || minor <= maxMinor)
     }
 
     /**
      * Free-text search over what the user can actually see or remember about a row: merchant, note,
      * category, tags — and the amount, which is matched numerically rather than as text so that
      * "12" finds €12.40 without also finding every note containing "12".
      */
     private fun matchesQuery(row: CategorizedTransaction): Boolean {
         if (needle.isEmpty()) return true
         val transaction = row.transaction
         val textHit = sequenceOf(transaction.merchant, transaction.note, row.categoryName.orEmpty())
             .plus(transaction.tags)
             .any { it.contains(needle, ignoreCase = true) }
         val amountHit = queryAmount != null &&
             transaction.money.abs().minor in queryAmount.fromMinor..queryAmount.toMinor
         return textHit || amountHit
     }
 }
 
 /** Parses the amount fields once so a per-row loop only compares — see [CompiledFilters]. */
 fun TransactionFilters.compile(): CompiledFilters {
     val trimmed = query.trim()
     return CompiledFilters(
         filters = this,
         minMinor = parseAmountQuery(minAmount)?.fromMinor,
         maxMinor = parseAmountQuery(maxAmount)?.toMinor,
         queryAmount = trimmed.takeIf { it.isNotEmpty() }?.let(::parseAmountQuery),
         needle = trimmed,
     )
 }
 
 /** Single-row convenience — compiles then tests. For a page of rows, [compile] once and reuse. */
 fun TransactionFilters.matches(row: CategorizedTransaction): Boolean = compile().matches(row)