| 1 | /*! |
| 2 | This module provides support for TZif binary files from the [Time Zone |
| 3 | Database]. |
| 4 | |
| 5 | These binary files are the ones commonly found in Unix distributions in the |
| 6 | `/usr/share/zoneinfo` directory. |
| 7 | |
| 8 | [Time Zone Database]: https://www.iana.org/time-zones |
| 9 | */ |
| 10 | |
| 11 | use core::ops::Range; |
| 12 | |
| 13 | #[cfg (feature = "alloc" )] |
| 14 | use alloc::{string::String, vec::Vec}; |
| 15 | |
| 16 | use crate::{ |
| 17 | civil::DateTime, |
| 18 | error::Error, |
| 19 | shared::{self, util::array_str::Abbreviation}, |
| 20 | timestamp::Timestamp, |
| 21 | tz::{ |
| 22 | posix::PosixTimeZone, timezone::TimeZoneAbbreviation, AmbiguousOffset, |
| 23 | Dst, Offset, TimeZoneOffsetInfo, TimeZoneTransition, |
| 24 | }, |
| 25 | }; |
| 26 | |
| 27 | /// The owned variant of `Tzif`. |
| 28 | #[cfg (feature = "alloc" )] |
| 29 | pub(crate) type TzifOwned = Tzif< |
| 30 | String, |
| 31 | Abbreviation, |
| 32 | Vec<shared::TzifLocalTimeType>, |
| 33 | Vec<i64>, |
| 34 | Vec<shared::TzifDateTime>, |
| 35 | Vec<shared::TzifDateTime>, |
| 36 | Vec<shared::TzifTransitionInfo>, |
| 37 | >; |
| 38 | |
| 39 | /// The static variant of `Tzif`. |
| 40 | pub(crate) type TzifStatic = Tzif< |
| 41 | &'static str, |
| 42 | &'static str, |
| 43 | &'static [shared::TzifLocalTimeType], |
| 44 | &'static [i64], |
| 45 | &'static [shared::TzifDateTime], |
| 46 | &'static [shared::TzifDateTime], |
| 47 | &'static [shared::TzifTransitionInfo], |
| 48 | >; |
| 49 | |
| 50 | /// A time zone based on IANA TZif formatted data. |
| 51 | /// |
| 52 | /// TZif is a binary format described by RFC 8536. Its typical structure is to |
| 53 | /// define a single time zone per file in the `/usr/share/zoneinfo` directory |
| 54 | /// on Unix systems. The name of a time zone is its file path with the |
| 55 | /// `/usr/share/zoneinfo/` prefix stripped from it. |
| 56 | /// |
| 57 | /// This type doesn't provide any facilities for dealing with files on disk |
| 58 | /// or the `/usr/share/zoneinfo` directory. This type is just for parsing the |
| 59 | /// contents of TZif formatted data in memory, and turning it into a data type |
| 60 | /// that can be used as a time zone. |
| 61 | #[derive (Debug)] |
| 62 | // not part of Jiff's public API |
| 63 | #[doc (hidden)] |
| 64 | // This ensures the alignment of this type is always *at least* 8 bytes. This |
| 65 | // is required for the pointer tagging inside of `TimeZone` to be sound. At |
| 66 | // time of writing (2024-02-24), this explicit `repr` isn't required on 64-bit |
| 67 | // systems since the type definition is such that it will have an alignment of |
| 68 | // at least 8 bytes anyway. But this *is* required for 32-bit systems, where |
| 69 | // the type definition at present only has an alignment of 4 bytes. |
| 70 | #[repr (align(8))] |
| 71 | pub struct Tzif<STR, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> { |
| 72 | inner: shared::Tzif<STR, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS>, |
| 73 | /// The POSIX time zone for this TZif data, if present. |
| 74 | /// |
| 75 | /// Note that this is also present on `shared::Tzif`, but uses the |
| 76 | /// `shared::PosixTimeZone` type, which isn't quite what we want here. |
| 77 | /// |
| 78 | /// For now we just duplicate it, which is slightly unfortunate. But this |
| 79 | /// is small and not a huge deal. Ideally we can clean this up later. |
| 80 | posix_tz: Option<PosixTimeZone<ABBREV>>, |
| 81 | } |
| 82 | |
| 83 | impl TzifStatic { |
| 84 | /// Converts from the shared-but-internal API for use in proc macros. |
| 85 | /// |
| 86 | /// This specifically works in a `const` context. And it requires that |
| 87 | /// caller to pass in the parsed `Tzif` in its fixed form along with the |
| 88 | /// variable length local time types and transitions. (Technically, the |
| 89 | /// TZ identifier and the designations are also variable length despite |
| 90 | /// being parsed of `TzifFixed`, but in practice they can be handled just |
| 91 | /// fine via `&'static str`.) |
| 92 | /// |
| 93 | /// Notice that the `types` and `transitions` are *not* from the `shared` |
| 94 | /// API, but rather, from the types defined in this module. They have to |
| 95 | /// be this way because there's a conversion step that occurs. In practice, |
| 96 | /// this sort of thing is embedded as a literal in source code via a proc |
| 97 | /// macro. Like this: |
| 98 | /// |
| 99 | /// ```text |
| 100 | /// static TZIF: Tzif<&str, &str, &[LocalTimeType], &[Transition]> = |
| 101 | /// Tzif::from_shared_const( |
| 102 | /// shared::TzifFixed { |
| 103 | /// name: Some("America/New_York"), |
| 104 | /// version: b'3', |
| 105 | /// checksum: 0xDEADBEEF, |
| 106 | /// designations: "ESTEDT", |
| 107 | /// posix_tz: None, |
| 108 | /// }, |
| 109 | /// &[ |
| 110 | /// shared::TzifLocalTimeType { |
| 111 | /// offset: -5 * 60 * 60, |
| 112 | /// is_dst: false, |
| 113 | /// designation: 0..3, |
| 114 | /// indicator: shared::TzifIndicator::LocalWall, |
| 115 | /// }.to_jiff(), |
| 116 | /// ], |
| 117 | /// &[ |
| 118 | /// shared::TzifTransition { |
| 119 | /// timestamp: 123456789, |
| 120 | /// type_index: 0, |
| 121 | /// }.to_jiff(-5, -5), |
| 122 | /// ], |
| 123 | /// ); |
| 124 | /// ``` |
| 125 | /// |
| 126 | /// Or something like that anyway. The point is, our `static` slices are |
| 127 | /// variable length and they need to be the right types. At least, I |
| 128 | /// couldn't see a simpler way to arrange this. |
| 129 | pub(crate) const fn from_shared_const( |
| 130 | sh: shared::TzifStatic, |
| 131 | ) -> TzifStatic { |
| 132 | let posix_tz = match sh.fixed.posix_tz { |
| 133 | None => None, |
| 134 | Some(posix_tz) => Some(PosixTimeZone::from_shared_const(posix_tz)), |
| 135 | }; |
| 136 | Tzif { inner: sh, posix_tz } |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | #[cfg (feature = "alloc" )] |
| 141 | impl TzifOwned { |
| 142 | /// Parses the given data as a TZif formatted file. |
| 143 | /// |
| 144 | /// The name given is attached to the `Tzif` value returned, but is |
| 145 | /// otherwise not significant. |
| 146 | /// |
| 147 | /// If the given data is not recognized to be valid TZif, then an error is |
| 148 | /// returned. |
| 149 | /// |
| 150 | /// In general, callers may assume that it is safe to pass arbitrary or |
| 151 | /// even untrusted data to this function and count on it not panicking |
| 152 | /// or using resources that aren't limited to a small constant factor of |
| 153 | /// the size of the data itself. That is, callers can reliably limit the |
| 154 | /// resources used by limiting the size of the data given to this parse |
| 155 | /// function. |
| 156 | pub(crate) fn parse( |
| 157 | name: Option<String>, |
| 158 | bytes: &[u8], |
| 159 | ) -> Result<Self, Error> { |
| 160 | let sh = |
| 161 | shared::TzifOwned::parse(name, bytes).map_err(Error::shared)?; |
| 162 | Ok(TzifOwned::from_shared_owned(sh)) |
| 163 | } |
| 164 | |
| 165 | /// Converts from the shared-but-internal API for use in proc macros. |
| 166 | /// |
| 167 | /// This is not `const` since it accepts owned values on the heap for |
| 168 | /// variable length data inside `Tzif`. |
| 169 | pub(crate) fn from_shared_owned(sh: shared::TzifOwned) -> TzifOwned { |
| 170 | let posix_tz = match sh.fixed.posix_tz { |
| 171 | None => None, |
| 172 | Some(posix_tz) => Some(PosixTimeZone::from_shared_owned(posix_tz)), |
| 173 | }; |
| 174 | Tzif { inner: sh, posix_tz } |
| 175 | } |
| 176 | } |
| 177 | |
| 178 | impl< |
| 179 | STR: AsRef<str>, |
| 180 | ABBREV: AsRef<str>, |
| 181 | TYPES: AsRef<[shared::TzifLocalTimeType]>, |
| 182 | TIMESTAMPS: AsRef<[i64]>, |
| 183 | STARTS: AsRef<[shared::TzifDateTime]>, |
| 184 | ENDS: AsRef<[shared::TzifDateTime]>, |
| 185 | INFOS: AsRef<[shared::TzifTransitionInfo]>, |
| 186 | > Tzif<STR, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> |
| 187 | { |
| 188 | /// Returns the name given to this TZif data in its constructor. |
| 189 | pub(crate) fn name(&self) -> Option<&str> { |
| 190 | self.inner.fixed.name.as_ref().map(|n| n.as_ref()) |
| 191 | } |
| 192 | |
| 193 | /// Returns the appropriate time zone offset to use for the given |
| 194 | /// timestamp. |
| 195 | pub(crate) fn to_offset(&self, timestamp: Timestamp) -> Offset { |
| 196 | match self.to_local_time_type(timestamp) { |
| 197 | Ok(typ) => Offset::from_seconds_unchecked(typ.offset), |
| 198 | Err(tz) => tz.to_offset(timestamp), |
| 199 | } |
| 200 | } |
| 201 | |
| 202 | /// Returns the appropriate time zone offset to use for the given |
| 203 | /// timestamp. |
| 204 | /// |
| 205 | /// This also includes whether the offset returned should be considered to |
| 206 | /// be DST or not, along with the time zone abbreviation (e.g., EST for |
| 207 | /// standard time in New York, and EDT for DST in New York). |
| 208 | pub(crate) fn to_offset_info( |
| 209 | &self, |
| 210 | timestamp: Timestamp, |
| 211 | ) -> TimeZoneOffsetInfo<'_> { |
| 212 | let typ = match self.to_local_time_type(timestamp) { |
| 213 | Ok(typ) => typ, |
| 214 | Err(tz) => return tz.to_offset_info(timestamp), |
| 215 | }; |
| 216 | let abbreviation = |
| 217 | TimeZoneAbbreviation::Borrowed(self.designation(typ)); |
| 218 | TimeZoneOffsetInfo { |
| 219 | offset: Offset::from_seconds_unchecked(typ.offset), |
| 220 | dst: Dst::from(typ.is_dst), |
| 221 | abbreviation, |
| 222 | } |
| 223 | } |
| 224 | |
| 225 | /// Returns the local time type for the timestamp given. |
| 226 | /// |
| 227 | /// If one could not be found, then this implies that the caller should |
| 228 | /// use the POSIX time zone returned in the error variant. |
| 229 | fn to_local_time_type( |
| 230 | &self, |
| 231 | timestamp: Timestamp, |
| 232 | ) -> Result<&shared::TzifLocalTimeType, &PosixTimeZone<ABBREV>> { |
| 233 | let timestamp = timestamp.as_second(); |
| 234 | // This is guaranteed because we always push at least one transition. |
| 235 | // This isn't guaranteed by TZif since it might have 0 transitions, |
| 236 | // but we always add a "dummy" first transition with our minimum |
| 237 | // `Timestamp` value. TZif doesn't do this because there is no |
| 238 | // universal minimum timestamp. (`i64::MIN` is a candidate, but that's |
| 239 | // likely to cause overflow in readers that don't do error checking.) |
| 240 | // |
| 241 | // The result of the dummy transition is that the code below is simpler |
| 242 | // with fewer special cases. |
| 243 | let timestamps = self.timestamps(); |
| 244 | assert!(!timestamps.is_empty(), "transitions is non-empty" ); |
| 245 | let index = if timestamp > *timestamps.last().unwrap() { |
| 246 | timestamps.len() - 1 |
| 247 | } else { |
| 248 | let search = self.timestamps().binary_search(×tamp); |
| 249 | match search { |
| 250 | // Since the first transition is always Timestamp::MIN, it's |
| 251 | // impossible for any timestamp to sort before it. |
| 252 | Err(0) => { |
| 253 | unreachable!("impossible to come before Timestamp::MIN" ) |
| 254 | } |
| 255 | Ok(i) => i, |
| 256 | // i points to the position immediately after the matching |
| 257 | // timestamp. And since we know that i>0 because of the i==0 |
| 258 | // check above, we can safely subtract 1. |
| 259 | Err(i) => i.checked_sub(1).expect("i is non-zero" ), |
| 260 | } |
| 261 | }; |
| 262 | // Our index is always in bounds. The only way it couldn't be is if |
| 263 | // binary search returns an Err(len) for a time greater than the |
| 264 | // maximum transition. But we account for that above by converting |
| 265 | // Err(len) to Err(len-1). |
| 266 | debug_assert!(index < timestamps.len()); |
| 267 | // RFC 8536 says: "Local time for timestamps on or after the last |
| 268 | // transition is specified by the TZ string in the footer (Section 3.3) |
| 269 | // if present and nonempty; otherwise, it is unspecified." |
| 270 | // |
| 271 | // Subtracting 1 is OK because we know self.transitions is not empty. |
| 272 | let index = if index < timestamps.len() - 1 { |
| 273 | // This is the typical case in "fat" TZif files: we found a |
| 274 | // matching transition. |
| 275 | index |
| 276 | } else { |
| 277 | match self.posix_tz() { |
| 278 | // This is the typical case in "slim" TZif files, where the |
| 279 | // last transition is, as I understand it, the transition at |
| 280 | // which a consistent rule started that a POSIX TZ string can |
| 281 | // fully describe. For example, (as of 2024-03-27) the last |
| 282 | // transition in the "fat" America/New_York TZif file is |
| 283 | // in 2037, where as in the "slim" version it is 2007. |
| 284 | // |
| 285 | // This is likely why some things break with the "slim" |
| 286 | // version: they don't support POSIX TZ strings (or don't |
| 287 | // support them correctly). |
| 288 | Some(tz) => return Err(tz), |
| 289 | // This case is technically unspecified, but I think the |
| 290 | // typical thing to do is to just use the last transition. |
| 291 | // I'm not 100% sure on this one. |
| 292 | None => index, |
| 293 | } |
| 294 | }; |
| 295 | Ok(self.local_time_type(index)) |
| 296 | } |
| 297 | |
| 298 | /// Returns a possibly ambiguous timestamp for the given civil datetime. |
| 299 | /// |
| 300 | /// The given datetime should correspond to the "wall" clock time of what |
| 301 | /// humans use to tell time for this time zone. |
| 302 | /// |
| 303 | /// Note that "ambiguous timestamp" is represented by the possible |
| 304 | /// selection of offsets that could be applied to the given datetime. In |
| 305 | /// general, it is only ambiguous around transitions to-and-from DST. The |
| 306 | /// ambiguity can arise as a "fold" (when a particular wall clock time is |
| 307 | /// repeated) or as a "gap" (when a particular wall clock time is skipped |
| 308 | /// entirely). |
| 309 | pub(crate) fn to_ambiguous_kind(&self, dt: DateTime) -> AmbiguousOffset { |
| 310 | // This implementation very nearly mirrors `to_local_time_type` |
| 311 | // above in the beginning: we do a binary search to find transition |
| 312 | // applicable for the given datetime. Except, we do it on wall clock |
| 313 | // times instead of timestamps. And in particular, each transition |
| 314 | // begins with a possibly ambiguous range of wall clock times |
| 315 | // corresponding to either a "gap" or "fold" in time. |
| 316 | let dtt = shared::TzifDateTime::new( |
| 317 | dt.year(), |
| 318 | dt.month(), |
| 319 | dt.day(), |
| 320 | dt.hour(), |
| 321 | dt.minute(), |
| 322 | dt.second(), |
| 323 | ); |
| 324 | let (starts, ends) = (self.civil_starts(), self.civil_ends()); |
| 325 | assert!(!starts.is_empty(), "transitions is non-empty" ); |
| 326 | let this_index = match starts.binary_search(&dtt) { |
| 327 | Err(0) => unreachable!("impossible to come before DateTime::MIN" ), |
| 328 | Ok(i) => i, |
| 329 | Err(i) => i.checked_sub(1).expect("i is non-zero" ), |
| 330 | }; |
| 331 | debug_assert!(this_index < starts.len()); |
| 332 | |
| 333 | let this_offset = self.local_time_type(this_index).offset; |
| 334 | // This is a little tricky, but we need to check for ambiguous civil |
| 335 | // datetimes before possibly using the POSIX TZ string. Namely, a |
| 336 | // datetime could be ambiguous with respect to the last transition, |
| 337 | // and we should handle that according to the gap/fold determined for |
| 338 | // that transition. We cover this case in tests in tz/mod.rs for the |
| 339 | // Pacific/Honolulu time zone, whose last transition begins with a gap. |
| 340 | match self.transition_kind(this_index) { |
| 341 | shared::TzifTransitionKind::Gap if dtt < ends[this_index] => { |
| 342 | // A gap/fold can only appear when there exists a previous |
| 343 | // transition. |
| 344 | let prev_index = this_index.checked_sub(1).unwrap(); |
| 345 | let prev_offset = self.local_time_type(prev_index).offset; |
| 346 | return AmbiguousOffset::Gap { |
| 347 | before: Offset::from_seconds_unchecked(prev_offset), |
| 348 | after: Offset::from_seconds_unchecked(this_offset), |
| 349 | }; |
| 350 | } |
| 351 | shared::TzifTransitionKind::Fold if dtt < ends[this_index] => { |
| 352 | // A gap/fold can only appear when there exists a previous |
| 353 | // transition. |
| 354 | let prev_index = this_index.checked_sub(1).unwrap(); |
| 355 | let prev_offset = self.local_time_type(prev_index).offset; |
| 356 | return AmbiguousOffset::Fold { |
| 357 | before: Offset::from_seconds_unchecked(prev_offset), |
| 358 | after: Offset::from_seconds_unchecked(this_offset), |
| 359 | }; |
| 360 | } |
| 361 | _ => {} |
| 362 | } |
| 363 | // The datetime given is not ambiguous with respect to any of the |
| 364 | // transitions in the TZif data. But, if we matched at or after the |
| 365 | // last transition, then we need to use the POSIX TZ string (which |
| 366 | // could still return an ambiguous offset). |
| 367 | if this_index == starts.len() - 1 { |
| 368 | if let Some(tz) = self.posix_tz() { |
| 369 | return tz.to_ambiguous_kind(dt); |
| 370 | } |
| 371 | // This case is unspecified according to RFC 8536. It means that |
| 372 | // the given datetime exceeds all transitions *and* there is no |
| 373 | // POSIX TZ string. So this can happen in V1 files for example. |
| 374 | // But those should hopefully be essentially non-existent nowadays |
| 375 | // (2024-03). In any case, we just fall through to using the last |
| 376 | // transition, which does seem likely to be wrong ~half the time |
| 377 | // in time zones with DST. But there really isn't much else we can |
| 378 | // do I think. |
| 379 | } |
| 380 | AmbiguousOffset::Unambiguous { |
| 381 | offset: Offset::from_seconds_unchecked(this_offset), |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | /// Returns the timestamp of the most recent time zone transition prior |
| 386 | /// to the timestamp given. If one doesn't exist, `None` is returned. |
| 387 | pub(crate) fn previous_transition( |
| 388 | &self, |
| 389 | ts: Timestamp, |
| 390 | ) -> Option<TimeZoneTransition> { |
| 391 | assert!(!self.timestamps().is_empty(), "transitions is non-empty" ); |
| 392 | let mut timestamp = ts.as_second(); |
| 393 | if ts.subsec_nanosecond() != 0 { |
| 394 | timestamp = timestamp.saturating_add(1); |
| 395 | } |
| 396 | let search = self.timestamps().binary_search(×tamp); |
| 397 | let index = match search { |
| 398 | Ok(i) | Err(i) => i.checked_sub(1)?, |
| 399 | }; |
| 400 | let index = if index == 0 { |
| 401 | // The first transition is a dummy that we insert, so if we land on |
| 402 | // it here, treat it as if it doesn't exist. |
| 403 | return None; |
| 404 | } else if index == self.timestamps().len() - 1 { |
| 405 | if let Some(ref posix_tz) = self.posix_tz() { |
| 406 | // Since the POSIX TZ must be consistent with the last |
| 407 | // transition, it must be the case that tzif_last <= |
| 408 | // posix_prev_trans in all cases. So the transition according |
| 409 | // to the POSIX TZ is always correct here. |
| 410 | // |
| 411 | // What if this returns `None` though? I'm not sure in which |
| 412 | // cases that could matter, and I think it might be a violation |
| 413 | // of the TZif format if it does. |
| 414 | // |
| 415 | // It can return `None`! In the case of a time zone that |
| 416 | // has eliminated DST, it might have historical time zone |
| 417 | // transitions but a POSIX time zone without DST. (For example, |
| 418 | // `America/Sao_Paulo`.) And thus, this would return `None`. |
| 419 | // So if it does, we pretend as if the POSIX time zone doesn't |
| 420 | // exist. |
| 421 | if let Some(trans) = posix_tz.previous_transition(ts) { |
| 422 | return Some(trans); |
| 423 | } |
| 424 | } |
| 425 | index |
| 426 | } else { |
| 427 | index |
| 428 | }; |
| 429 | let timestamp = self.timestamps()[index]; |
| 430 | let typ = self.local_time_type(index); |
| 431 | Some(TimeZoneTransition { |
| 432 | timestamp: Timestamp::constant(timestamp, 0), |
| 433 | offset: Offset::from_seconds_unchecked(typ.offset), |
| 434 | abbrev: self.designation(typ), |
| 435 | dst: Dst::from(typ.is_dst), |
| 436 | }) |
| 437 | } |
| 438 | |
| 439 | /// Returns the timestamp of the soonest time zone transition after the |
| 440 | /// timestamp given. If one doesn't exist, `None` is returned. |
| 441 | pub(crate) fn next_transition( |
| 442 | &self, |
| 443 | ts: Timestamp, |
| 444 | ) -> Option<TimeZoneTransition> { |
| 445 | assert!(!self.timestamps().is_empty(), "transitions is non-empty" ); |
| 446 | let timestamp = ts.as_second(); |
| 447 | let search = self.timestamps().binary_search(×tamp); |
| 448 | let index = match search { |
| 449 | Ok(i) => i.checked_add(1)?, |
| 450 | Err(i) => i, |
| 451 | }; |
| 452 | let index = if index == 0 { |
| 453 | // The first transition is a dummy that we insert, so if we land on |
| 454 | // it here, treat it as if it doesn't exist. |
| 455 | return None; |
| 456 | } else if index >= self.timestamps().len() - 1 { |
| 457 | if let Some(posix_tz) = self.posix_tz() { |
| 458 | // Since the POSIX TZ must be consistent with the last |
| 459 | // transition, it must be the case that next.timestamp <= |
| 460 | // posix_next_tans in all cases. So the transition according to |
| 461 | // the POSIX TZ is always correct here. |
| 462 | // |
| 463 | // What if this returns `None` though? I'm not sure in which |
| 464 | // cases that could matter, and I think it might be a violation |
| 465 | // of the TZif format if it does. |
| 466 | // |
| 467 | // In the "previous" case above, this could return `None` even |
| 468 | // when there are historical time zone transitions in the case |
| 469 | // of a time zone eliminating DST (e.g., `America/Sao_Paulo`). |
| 470 | // But unlike the previous case, if we get `None` here, then |
| 471 | // that is the real answer because there are no other known |
| 472 | // future time zone transitions. |
| 473 | return posix_tz.next_transition(ts); |
| 474 | } |
| 475 | self.timestamps().len() - 1 |
| 476 | } else { |
| 477 | index |
| 478 | }; |
| 479 | let timestamp = self.timestamps()[index]; |
| 480 | let typ = self.local_time_type(index); |
| 481 | Some(TimeZoneTransition { |
| 482 | timestamp: Timestamp::constant(timestamp, 0), |
| 483 | offset: Offset::from_seconds_unchecked(typ.offset), |
| 484 | abbrev: self.designation(typ), |
| 485 | dst: Dst::from(typ.is_dst), |
| 486 | }) |
| 487 | } |
| 488 | |
| 489 | fn designation(&self, typ: &shared::TzifLocalTimeType) -> &str { |
| 490 | // OK because we verify that the designation range on every local |
| 491 | // time type is a valid range into `self.designations`. |
| 492 | &self.designations()[typ.designation()] |
| 493 | } |
| 494 | |
| 495 | fn local_time_type( |
| 496 | &self, |
| 497 | transition_index: usize, |
| 498 | ) -> &shared::TzifLocalTimeType { |
| 499 | // OK because we require that `type_index` always points to a valid |
| 500 | // local time type. |
| 501 | &self.types()[usize::from(self.infos()[transition_index].type_index)] |
| 502 | } |
| 503 | |
| 504 | fn transition_kind( |
| 505 | &self, |
| 506 | transition_index: usize, |
| 507 | ) -> shared::TzifTransitionKind { |
| 508 | self.infos()[transition_index].kind |
| 509 | } |
| 510 | |
| 511 | fn posix_tz(&self) -> Option<&PosixTimeZone<ABBREV>> { |
| 512 | self.posix_tz.as_ref() |
| 513 | } |
| 514 | |
| 515 | fn designations(&self) -> &str { |
| 516 | self.inner.fixed.designations.as_ref() |
| 517 | } |
| 518 | |
| 519 | fn types(&self) -> &[shared::TzifLocalTimeType] { |
| 520 | self.inner.types.as_ref() |
| 521 | } |
| 522 | |
| 523 | fn timestamps(&self) -> &[i64] { |
| 524 | self.inner.transitions.timestamps.as_ref() |
| 525 | } |
| 526 | |
| 527 | fn civil_starts(&self) -> &[shared::TzifDateTime] { |
| 528 | self.inner.transitions.civil_starts.as_ref() |
| 529 | } |
| 530 | |
| 531 | fn civil_ends(&self) -> &[shared::TzifDateTime] { |
| 532 | self.inner.transitions.civil_ends.as_ref() |
| 533 | } |
| 534 | |
| 535 | fn infos(&self) -> &[shared::TzifTransitionInfo] { |
| 536 | self.inner.transitions.infos.as_ref() |
| 537 | } |
| 538 | } |
| 539 | |
| 540 | impl<STR: AsRef<str>, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> Eq |
| 541 | for Tzif<STR, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> |
| 542 | { |
| 543 | } |
| 544 | |
| 545 | impl<STR: AsRef<str>, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> PartialEq |
| 546 | for Tzif<STR, ABBREV, TYPES, TIMESTAMPS, STARTS, ENDS, INFOS> |
| 547 | { |
| 548 | fn eq(&self, rhs: &Self) -> bool { |
| 549 | self.inner.fixed.name.as_ref().map(|n: &STR| n.as_ref()) |
| 550 | == rhs.inner.fixed.name.as_ref().map(|n: &STR| n.as_ref()) |
| 551 | && self.inner.fixed.checksum == rhs.inner.fixed.checksum |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | impl shared::TzifLocalTimeType { |
| 556 | fn designation(&self) -> Range<usize> { |
| 557 | usize::from(self.designation.0)..usize::from(self.designation.1) |
| 558 | } |
| 559 | } |
| 560 | |
| 561 | impl core::fmt::Display for shared::TzifIndicator { |
| 562 | fn fmt(&self, f: &mut core::fmt::Formatter) -> core::fmt::Result { |
| 563 | match *self { |
| 564 | shared::TzifIndicator::LocalWall => write!(f, "local/wall" ), |
| 565 | shared::TzifIndicator::LocalStandard => write!(f, "local/std" ), |
| 566 | shared::TzifIndicator::UTStandard => write!(f, "ut/std" ), |
| 567 | } |
| 568 | } |
| 569 | } |
| 570 | |
| 571 | /// Does a quick check that returns true if the data might be in TZif format. |
| 572 | /// |
| 573 | /// It is possible that this returns true even if the given data is not in TZif |
| 574 | /// format. However, it is impossible for this to return false when the given |
| 575 | /// data is TZif. That is, a false positive is allowed but a false negative is |
| 576 | /// not. |
| 577 | #[cfg (feature = "tzdb-zoneinfo" )] |
| 578 | pub(crate) fn is_possibly_tzif(data: &[u8]) -> bool { |
| 579 | data.starts_with(b"TZif" ) |
| 580 | } |
| 581 | |
| 582 | #[cfg (all(test, feature = "alloc" ))] |
| 583 | mod tests { |
| 584 | use alloc::{string::ToString, vec}; |
| 585 | |
| 586 | #[cfg (not(miri))] |
| 587 | use crate::tz::testdata::TZIF_TEST_FILES; |
| 588 | |
| 589 | use super::*; |
| 590 | |
| 591 | /// This converts TZif data into a human readable format. |
| 592 | /// |
| 593 | /// This is useful for debugging (via `./scripts/jiff-debug tzif`), but we |
| 594 | /// also use it for snapshot testing to make reading the test output at |
| 595 | /// least *somewhat* comprehensible for humans. Otherwise, one needs to |
| 596 | /// read and understand Unix timestamps. That ain't going to fly. |
| 597 | /// |
| 598 | /// For this to work, we make sure everything in a `Tzif` value is |
| 599 | /// represented in some way in this output. |
| 600 | fn tzif_to_human_readable(tzif: &TzifOwned) -> String { |
| 601 | use std::io::Write; |
| 602 | |
| 603 | fn datetime(dt: shared::TzifDateTime) -> DateTime { |
| 604 | DateTime::constant( |
| 605 | dt.year(), |
| 606 | dt.month(), |
| 607 | dt.day(), |
| 608 | dt.hour(), |
| 609 | dt.minute(), |
| 610 | dt.second(), |
| 611 | 0, |
| 612 | ) |
| 613 | } |
| 614 | |
| 615 | let mut out = tabwriter::TabWriter::new(vec![]) |
| 616 | .alignment(tabwriter::Alignment::Left); |
| 617 | |
| 618 | writeln!(out, "TIME ZONE NAME" ).unwrap(); |
| 619 | writeln!(out, " {}" , tzif.name().unwrap_or("UNNAMED" )).unwrap(); |
| 620 | |
| 621 | writeln!(out, "TIME ZONE VERSION" ).unwrap(); |
| 622 | writeln!( |
| 623 | out, |
| 624 | " {}" , |
| 625 | char::try_from(tzif.inner.fixed.version).unwrap() |
| 626 | ) |
| 627 | .unwrap(); |
| 628 | |
| 629 | writeln!(out, "LOCAL TIME TYPES" ).unwrap(); |
| 630 | for (i, typ) in tzif.inner.types.iter().enumerate() { |
| 631 | writeln!( |
| 632 | out, |
| 633 | " {i:03}: \toffset={off} \t\ |
| 634 | designation={desig} \t{dst} \tindicator={ind}" , |
| 635 | off = Offset::from_seconds_unchecked(typ.offset), |
| 636 | desig = tzif.designation(&typ), |
| 637 | dst = if typ.is_dst { "dst" } else { "" }, |
| 638 | ind = typ.indicator, |
| 639 | ) |
| 640 | .unwrap(); |
| 641 | } |
| 642 | if !tzif.timestamps().is_empty() { |
| 643 | writeln!(out, "TRANSITIONS" ).unwrap(); |
| 644 | for i in 0..tzif.timestamps().len() { |
| 645 | let timestamp = Timestamp::constant(tzif.timestamps()[i], 0); |
| 646 | let dt = Offset::UTC.to_datetime(timestamp); |
| 647 | let typ = tzif.local_time_type(i); |
| 648 | let wall = |
| 649 | alloc::format!("{}" , datetime(tzif.civil_starts()[i])); |
| 650 | let ambiguous = match tzif.transition_kind(i) { |
| 651 | shared::TzifTransitionKind::Unambiguous => { |
| 652 | "unambiguous" .to_string() |
| 653 | } |
| 654 | shared::TzifTransitionKind::Gap => { |
| 655 | let end = datetime(tzif.civil_ends()[i]); |
| 656 | alloc::format!(" gap-until({end})" ) |
| 657 | } |
| 658 | shared::TzifTransitionKind::Fold => { |
| 659 | let end = datetime(tzif.civil_ends()[i]); |
| 660 | alloc::format!("fold-until({end})" ) |
| 661 | } |
| 662 | }; |
| 663 | |
| 664 | writeln!( |
| 665 | out, |
| 666 | " {i:04}: \t{dt:?}Z \tunix={ts} \twall={wall} \t\ |
| 667 | {ambiguous} \t\ |
| 668 | type={type_index} \t{off} \t\ |
| 669 | {desig} \t{dst}" , |
| 670 | ts = timestamp.as_second(), |
| 671 | type_index = tzif.infos()[i].type_index, |
| 672 | off = Offset::from_seconds_unchecked(typ.offset), |
| 673 | desig = tzif.designation(typ), |
| 674 | dst = if typ.is_dst { "dst" } else { "" }, |
| 675 | ) |
| 676 | .unwrap(); |
| 677 | } |
| 678 | } |
| 679 | if let Some(ref posix_tz) = tzif.posix_tz { |
| 680 | writeln!(out, "POSIX TIME ZONE STRING" ).unwrap(); |
| 681 | writeln!(out, " {}" , posix_tz).unwrap(); |
| 682 | } |
| 683 | String::from_utf8(out.into_inner().unwrap()).unwrap() |
| 684 | } |
| 685 | |
| 686 | /// DEBUG COMMAND |
| 687 | /// |
| 688 | /// Takes environment variable `JIFF_DEBUG_TZIF_PATH` as input, and treats |
| 689 | /// the value as a TZif file path. This test will open the file, parse it |
| 690 | /// as a TZif and then dump debug data about the file in a human readable |
| 691 | /// plain text format. |
| 692 | #[cfg (feature = "std" )] |
| 693 | #[test ] |
| 694 | fn debug_tzif() -> anyhow::Result<()> { |
| 695 | use anyhow::Context; |
| 696 | |
| 697 | let _ = crate::logging::Logger::init(); |
| 698 | |
| 699 | const ENV: &str = "JIFF_DEBUG_TZIF_PATH" ; |
| 700 | let Some(val) = std::env::var_os(ENV) else { return Ok(()) }; |
| 701 | let Ok(val) = val.into_string() else { |
| 702 | anyhow::bail!("{ENV} has invalid UTF-8" ) |
| 703 | }; |
| 704 | let bytes = |
| 705 | std::fs::read(&val).with_context(|| alloc::format!("{val:?}" ))?; |
| 706 | let tzif = Tzif::parse(Some(val.to_string()), &bytes)?; |
| 707 | std::eprint!("{}" , tzif_to_human_readable(&tzif)); |
| 708 | Ok(()) |
| 709 | } |
| 710 | |
| 711 | #[cfg (not(miri))] |
| 712 | #[test ] |
| 713 | fn tzif_parse_v2plus() { |
| 714 | for tzif_test in TZIF_TEST_FILES { |
| 715 | insta::assert_snapshot!( |
| 716 | alloc::format!("{}_v2+" , tzif_test.name), |
| 717 | tzif_to_human_readable(&tzif_test.parse()) |
| 718 | ); |
| 719 | } |
| 720 | } |
| 721 | |
| 722 | #[cfg (not(miri))] |
| 723 | #[test ] |
| 724 | fn tzif_parse_v1() { |
| 725 | for tzif_test in TZIF_TEST_FILES { |
| 726 | insta::assert_snapshot!( |
| 727 | alloc::format!("{}_v1" , tzif_test.name), |
| 728 | tzif_to_human_readable(&tzif_test.parse_v1()) |
| 729 | ); |
| 730 | } |
| 731 | } |
| 732 | |
| 733 | /// This tests walks the /usr/share/zoneinfo directory (if it exists) and |
| 734 | /// tries to parse every TZif formatted file it can find. We don't really |
| 735 | /// do much with it other than to ensure we don't panic or return an error. |
| 736 | /// That is, we check that we can parse each file, but not that we do so |
| 737 | /// correctly. |
| 738 | #[cfg (not(miri))] |
| 739 | #[cfg (feature = "tzdb-zoneinfo" )] |
| 740 | #[cfg (target_os = "linux" )] |
| 741 | #[test ] |
| 742 | fn zoneinfo() { |
| 743 | const TZDIR: &str = "/usr/share/zoneinfo" ; |
| 744 | |
| 745 | for result in walkdir::WalkDir::new(TZDIR) { |
| 746 | // Just skip if we got an error traversing the directory tree. |
| 747 | // These aren't related to our parsing, so it's some other problem |
| 748 | // (like the directory not existing). |
| 749 | let Ok(dent) = result else { continue }; |
| 750 | // This test can take some time in debug mode, so skip parsing |
| 751 | // some of the less frequently used TZif files. |
| 752 | let Some(name) = dent.path().to_str() else { continue }; |
| 753 | if name.contains("right/" ) || name.contains("posix/" ) { |
| 754 | continue; |
| 755 | } |
| 756 | // Again, skip if we can't read. Not my monkeys, not my circus. |
| 757 | let Ok(bytes) = std::fs::read(dent.path()) else { continue }; |
| 758 | if !is_possibly_tzif(&bytes) { |
| 759 | continue; |
| 760 | } |
| 761 | let tzname = dent |
| 762 | .path() |
| 763 | .strip_prefix(TZDIR) |
| 764 | .unwrap_or_else(|_| { |
| 765 | panic!("all paths in TZDIR have {TZDIR:?} prefix" ) |
| 766 | }) |
| 767 | .to_str() |
| 768 | .expect("all paths to be valid UTF-8" ) |
| 769 | .to_string(); |
| 770 | // OK at this point, we're pretty sure `bytes` should be a TZif |
| 771 | // binary file. So try to parse it and fail the test if it fails. |
| 772 | if let Err(err) = Tzif::parse(Some(tzname), &bytes) { |
| 773 | panic!("failed to parse TZif file {:?}: {err}" , dent.path()); |
| 774 | } |
| 775 | } |
| 776 | } |
| 777 | } |
| 778 | |