1 // Copyright 2012-2015 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
11 //! Really Bad Markup Language (rbml) is an internal serialization format of rustc.
12 //! This is not intended to be used by users.
14 //! Originally based on the Extensible Binary Markup Language
15 //! (ebml; http://www.matroska.org/technical/specs/rfc/index.html),
16 //! it is now a separate format tuned for the rust object metadata.
20 //! RBML document consists of the tag, length and data.
21 //! The encoded data can contain multiple RBML documents concatenated.
23 //! **Tags** are a hint for the following data.
24 //! Tags are a number from 0x000 to 0xfff, where 0xf0 through 0xff is reserved.
25 //! Tags less than 0xf0 are encoded in one literal byte.
26 //! Tags greater than 0xff are encoded in two big-endian bytes,
27 //! where the tag number is ORed with 0xf000. (E.g. tag 0x123 = `f1 23`)
29 //! **Lengths** encode the length of the following data.
30 //! It is a variable-length unsigned isize, and one of the following forms:
32 //! - `80` through `fe` for lengths up to 0x7e;
33 //! - `40 ff` through `7f ff` for lengths up to 0x3fff;
34 //! - `20 40 00` through `3f ff ff` for lengths up to 0x1fffff;
35 //! - `10 20 00 00` through `1f ff ff ff` for lengths up to 0xfffffff.
37 //! The "overlong" form is allowed so that the length can be encoded
38 //! without the prior knowledge of the encoded data.
39 //! For example, the length 0 can be represented either by `80`, `40 00`,
40 //! `20 00 00` or `10 00 00 00`.
41 //! The encoder tries to minimize the length if possible.
42 //! Also, some predefined tags listed below are so commonly used that
43 //! their lengths are omitted ("implicit length").
45 //! **Data** can be either binary bytes or zero or more nested RBML documents.
46 //! Nested documents cannot overflow, and should be entirely contained
47 //! within a parent document.
51 //! Most RBML tags are defined by the application.
52 //! (For the rust object metadata, see also `rustc::metadata::common`.)
53 //! RBML itself does define a set of predefined tags however,
54 //! intended for the auto-serialization implementation.
56 //! Predefined tags with an implicit length:
58 //! - `U8` (`00`): 1-byte unsigned integer.
59 //! - `U16` (`01`): 2-byte big endian unsigned integer.
60 //! - `U32` (`02`): 4-byte big endian unsigned integer.
61 //! - `U64` (`03`): 8-byte big endian unsigned integer.
62 //! Any of `U*` tags can be used to encode primitive unsigned integer types,
63 //! as long as it is no greater than the actual size.
64 //! For example, `u8` can only be represented via the `U8` tag.
66 //! - `I8` (`04`): 1-byte signed integer.
67 //! - `I16` (`05`): 2-byte big endian signed integer.
68 //! - `I32` (`06`): 4-byte big endian signed integer.
69 //! - `I64` (`07`): 8-byte big endian signed integer.
70 //! Similar to `U*` tags. Always uses two's complement encoding.
72 //! - `Bool` (`08`): 1-byte boolean value, `00` for false and `01` for true.
74 //! - `Char` (`09`): 4-byte big endian Unicode scalar value.
75 //! Surrogate pairs or out-of-bound values are invalid.
77 //! - `F32` (`0a`): 4-byte big endian unsigned integer representing
78 //! IEEE 754 binary32 floating-point format.
79 //! - `F64` (`0b`): 8-byte big endian unsigned integer representing
80 //! IEEE 754 binary64 floating-point format.
82 //! - `Sub8` (`0c`): 1-byte unsigned integer for supplementary information.
83 //! - `Sub32` (`0d`): 4-byte unsigned integer for supplementary information.
84 //! Those two tags normally occur as the first subdocument of certain tags,
85 //! namely `Enum`, `Vec` and `Map`, to provide a variant or size information.
86 //! They can be used interchangeably.
88 //! Predefined tags with an explicit length:
90 //! - `Str` (`10`): A UTF-8-encoded string.
92 //! - `Enum` (`11`): An enum.
93 //! The first subdocument should be `Sub*` tags with a variant ID.
94 //! Subsequent subdocuments, if any, encode variant arguments.
96 //! - `Vec` (`12`): A vector (sequence).
97 //! - `VecElt` (`13`): A vector element.
98 //! The first subdocument should be `Sub*` tags with the number of elements.
99 //! Subsequent subdocuments should be `VecElt` tag per each element.
101 //! - `Map` (`14`): A map (associated array).
102 //! - `MapKey` (`15`): A key part of the map entry.
103 //! - `MapVal` (`16`): A value part of the map entry.
104 //! The first subdocument should be `Sub*` tags with the number of entries.
105 //! Subsequent subdocuments should be an alternating sequence of
106 //! `MapKey` and `MapVal` tags per each entry.
108 //! - `Opaque` (`17`): An opaque, custom-format tag.
109 //! Used to wrap ordinary custom tags or data in the auto-serialized context.
110 //! Rustc typically uses this to encode type informations.
112 //! First 0x20 tags are reserved by RBML; custom tags start at 0x20.
114 #![crate_name = "rbml"]
115 #![unstable(feature = "rustc_private", issue = "27812")]
116 #![crate_type = "rlib"]
117 #![crate_type = "dylib"]
118 #![doc(html_logo_url = "https://www.rust-lang.org/logos/rust-logo-128x128-blk-v2.png",
119 html_favicon_url = "https://doc.rust-lang.org/favicon.ico",
120 html_root_url = "https://doc.rust-lang.org/nightly/",
121 html_playground_url = "https://play.rust-lang.org/",
122 test(attr(deny(warnings))))]
124 #![feature(rustc_private)]
125 #![feature(staged_api)]
127 #![cfg_attr(test, feature(test))]
129 extern crate serialize;
132 extern crate serialize as rustc_serialize; // Used by RustcEncodable
143 pub use self::EbmlEncoderTag::*;
144 pub use self::Error::*;
149 /// Common data structures
150 #[derive(Clone, Copy)]
157 impl<'doc> Doc<'doc> {
158 pub fn new(data: &'doc [u8]) -> Doc<'doc> {
166 pub fn get<'a>(&'a self, tag: usize) -> Doc<'a> {
167 reader::get_doc(*self, tag)
170 pub fn is_empty(&self) -> bool {
171 self.start == self.end
174 pub fn as_str_slice<'a>(&'a self) -> &'a str {
175 str::from_utf8(&self.data[self.start..self.end]).unwrap()
178 pub fn as_str(&self) -> String {
179 self.as_str_slice().to_string()
183 pub struct TaggedDoc<'a> {
188 #[derive(Copy, Clone, Debug)]
189 pub enum EbmlEncoderTag {
190 // tags 00..1f are reserved for auto-serialization.
191 // first NUM_IMPLICIT_TAGS tags are implicitly sized and lengths are not encoded.
192 EsU8 = 0x00, // + 1 byte
193 EsU16 = 0x01, // + 2 bytes
194 EsU32 = 0x02, // + 4 bytes
195 EsU64 = 0x03, // + 8 bytes
196 EsI8 = 0x04, // + 1 byte
197 EsI16 = 0x05, // + 2 bytes
198 EsI32 = 0x06, // + 4 bytes
199 EsI64 = 0x07, // + 8 bytes
200 EsBool = 0x08, // + 1 byte
201 EsChar = 0x09, // + 4 bytes
202 EsF32 = 0x0a, // + 4 bytes
203 EsF64 = 0x0b, // + 8 bytes
204 EsSub8 = 0x0c, // + 1 byte
205 EsSub32 = 0x0d, // + 4 bytes
206 // 0x0e and 0x0f are reserved
208 EsEnum = 0x11, // encodes the variant id as the first EsSub*
209 EsVec = 0x12, // encodes the # of elements as the first EsSub*
211 EsMap = 0x14, // encodes the # of pairs as the first EsSub*
217 const NUM_TAGS: usize = 0x1000;
218 const NUM_IMPLICIT_TAGS: usize = 0x0e;
220 #[cfg_attr(rustfmt, rustfmt_skip)]
221 static TAG_IMPLICIT_LEN: [i8; NUM_IMPLICIT_TAGS] = [
235 IoError(std::io::Error),
236 ApplicationError(String),
239 impl fmt::Display for Error {
240 fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
241 // FIXME: this should be a more useful display form
242 fmt::Debug::fmt(self, f)
245 // --------------------------------------
251 use std::mem::transmute;
256 use super::{ApplicationError, EsVec, EsMap, EsEnum, EsSub8, EsSub32, EsVecElt, EsMapKey,
257 EsU64, EsU32, EsU16, EsU8, EsI64, EsI32, EsI16, EsI8, EsBool, EsF64, EsF32,
258 EsChar, EsStr, EsMapVal, EsOpaque, EbmlEncoderTag, Doc, TaggedDoc, Error,
259 IntTooBig, InvalidTag, Expected, NUM_IMPLICIT_TAGS, TAG_IMPLICIT_LEN};
261 pub type DecodeResult<T> = Result<T, Error>;
264 macro_rules! try_or {
265 ($e:expr, $r:expr) => (
269 debug!("ignored error: {:?}", e);
276 #[derive(Copy, Clone)]
282 pub fn tag_at(data: &[u8], start: usize) -> DecodeResult<Res> {
283 let v = data[start] as usize;
291 val: ((v & 0xf) << 8) | data[start + 1] as usize,
295 // every tag starting with byte 0xf0 is an overlong form, which is prohibited.
301 fn vuint_at_slow(data: &[u8], start: usize) -> DecodeResult<Res> {
305 val: (a & 0x7f) as usize,
311 val: ((a & 0x3f) as usize) << 8 | (data[start + 1] as usize),
317 val: ((a & 0x1f) as usize) << 16 | (data[start + 1] as usize) << 8 |
318 (data[start + 2] as usize),
324 val: ((a & 0x0f) as usize) << 24 | (data[start + 1] as usize) << 16 |
325 (data[start + 2] as usize) << 8 |
326 (data[start + 3] as usize),
330 Err(IntTooBig(a as usize))
333 pub fn vuint_at(data: &[u8], start: usize) -> DecodeResult<Res> {
334 if data.len() - start < 4 {
335 return vuint_at_slow(data, start);
338 // Lookup table for parsing EBML Element IDs as per
339 // http://ebml.sourceforge.net/specs/ The Element IDs are parsed by
340 // reading a big endian u32 positioned at data[start]. Using the four
341 // most significant bits of the u32 we lookup in the table below how
342 // the element ID should be derived from it.
344 // The table stores tuples (shift, mask) where shift is the number the
345 // u32 should be right shifted with and mask is the value the right
346 // shifted value should be masked with. If for example the most
347 // significant bit is set this means it's a class A ID and the u32
348 // should be right shifted with 24 and masked with 0x7f. Therefore we
349 // store (24, 0x7f) at index 0x8 - 0xF (four bit numbers where the most
350 // significant bit is set).
352 // By storing the number of shifts and masks in a table instead of
353 // checking in order if the most significant bit is set, the second
354 // most significant bit is set etc. we can replace up to three
355 // "and+branch" with a single table lookup which gives us a measured
356 // speedup of around 2x on x86_64.
357 static SHIFT_MASK_TABLE: [(usize, u32); 16] = [(0, 0x0),
375 let ptr = data.as_ptr().offset(start as isize) as *const u32;
376 let val = u32::from_be(*ptr);
378 let i = (val >> 28) as usize;
379 let (shift, mask) = SHIFT_MASK_TABLE[i];
381 val: ((val >> shift) & mask) as usize,
382 next: start + ((32 - shift) >> 3),
387 pub fn tag_len_at(data: &[u8], tag: Res) -> DecodeResult<Res> {
388 if tag.val < NUM_IMPLICIT_TAGS && TAG_IMPLICIT_LEN[tag.val] >= 0 {
390 val: TAG_IMPLICIT_LEN[tag.val] as usize,
394 vuint_at(data, tag.next)
398 pub fn doc_at<'a>(data: &'a [u8], start: usize) -> DecodeResult<TaggedDoc<'a>> {
399 let elt_tag = try!(tag_at(data, start));
400 let elt_size = try!(tag_len_at(data, elt_tag));
401 let end = elt_size.next + elt_size.val;
406 start: elt_size.next,
412 pub fn maybe_get_doc<'a>(d: Doc<'a>, tg: usize) -> Option<Doc<'a>> {
413 let mut pos = d.start;
415 let elt_tag = try_or!(tag_at(d.data, pos), None);
416 let elt_size = try_or!(tag_len_at(d.data, elt_tag), None);
417 pos = elt_size.next + elt_size.val;
418 if elt_tag.val == tg {
421 start: elt_size.next,
429 pub fn get_doc<'a>(d: Doc<'a>, tg: usize) -> Doc<'a> {
430 match maybe_get_doc(d, tg) {
433 error!("failed to find block with tag {:?}", tg);
439 pub fn docs<'a>(d: Doc<'a>) -> DocsIterator<'a> {
440 DocsIterator { d: d }
443 pub struct DocsIterator<'a> {
447 impl<'a> Iterator for DocsIterator<'a> {
448 type Item = (usize, Doc<'a>);
450 fn next(&mut self) -> Option<(usize, Doc<'a>)> {
451 if self.d.start >= self.d.end {
455 let elt_tag = try_or!(tag_at(self.d.data, self.d.start), {
456 self.d.start = self.d.end;
459 let elt_size = try_or!(tag_len_at(self.d.data, elt_tag), {
460 self.d.start = self.d.end;
464 let end = elt_size.next + elt_size.val;
467 start: elt_size.next,
472 return Some((elt_tag.val, doc));
476 pub fn tagged_docs<'a>(d: Doc<'a>, tag: usize) -> TaggedDocsIterator<'a> {
483 pub struct TaggedDocsIterator<'a> {
484 iter: DocsIterator<'a>,
488 impl<'a> Iterator for TaggedDocsIterator<'a> {
491 fn next(&mut self) -> Option<Doc<'a>> {
492 while let Some((tag, doc)) = self.iter.next() {
501 pub fn with_doc_data<T, F>(d: Doc, f: F) -> T
502 where F: FnOnce(&[u8]) -> T
504 f(&d.data[d.start..d.end])
507 pub fn doc_as_u8(d: Doc) -> u8 {
508 assert_eq!(d.end, d.start + 1);
512 pub fn doc_as_u64(d: Doc) -> u64 {
514 // For performance, we read 8 big-endian bytes,
515 // and mask off the junk if there is any. This
516 // obviously won't work on the first 8 bytes
517 // of a file - we will fall of the start
518 // of the page and segfault.
521 b.clone_from_slice(&d.data[d.end - 8..d.end]);
522 let data = unsafe { (*(b.as_ptr() as *const u64)).to_be() };
523 let len = d.end - d.start;
525 data & ((1 << (len * 8)) - 1)
531 for b in &d.data[d.start..d.end] {
532 result = (result << 8) + (*b as u64);
539 pub fn doc_as_u16(d: Doc) -> u16 {
543 pub fn doc_as_u32(d: Doc) -> u32 {
548 pub fn doc_as_i8(d: Doc) -> i8 {
552 pub fn doc_as_i16(d: Doc) -> i16 {
556 pub fn doc_as_i32(d: Doc) -> i32 {
560 pub fn doc_as_i64(d: Doc) -> i64 {
564 pub struct Decoder<'a> {
569 impl<'doc> Decoder<'doc> {
570 pub fn new(d: Doc<'doc>) -> Decoder<'doc> {
577 fn next_doc(&mut self, exp_tag: EbmlEncoderTag) -> DecodeResult<Doc<'doc>> {
578 debug!(". next_doc(exp_tag={:?})", exp_tag);
579 if self.pos >= self.parent.end {
580 return Err(Expected(format!("no more documents in current node!")));
582 let TaggedDoc { tag: r_tag, doc: r_doc } = try!(doc_at(self.parent.data, self.pos));
583 debug!("self.parent={:?}-{:?} self.pos={:?} r_tag={:?} r_doc={:?}-{:?}",
590 if r_tag != (exp_tag as usize) {
591 return Err(Expected(format!("expected EBML doc with tag {:?} but found tag {:?}",
595 if r_doc.end > self.parent.end {
596 return Err(Expected(format!("invalid EBML, child extends to {:#x}, parent to \
601 self.pos = r_doc.end;
605 fn push_doc<T, F>(&mut self, exp_tag: EbmlEncoderTag, f: F) -> DecodeResult<T>
606 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
608 let d = try!(self.next_doc(exp_tag));
609 let old_parent = self.parent;
610 let old_pos = self.pos;
613 let r = try!(f(self));
614 self.parent = old_parent;
619 fn _next_sub(&mut self) -> DecodeResult<usize> {
620 // empty vector/map optimization
621 if self.parent.is_empty() {
625 let TaggedDoc { tag: r_tag, doc: r_doc } = try!(doc_at(self.parent.data, self.pos));
626 let r = if r_tag == (EsSub8 as usize) {
627 doc_as_u8(r_doc) as usize
628 } else if r_tag == (EsSub32 as usize) {
629 doc_as_u32(r_doc) as usize
631 return Err(Expected(format!("expected EBML doc with tag {:?} or {:?} but found \
637 if r_doc.end > self.parent.end {
638 return Err(Expected(format!("invalid EBML, child extends to {:#x}, parent to \
643 self.pos = r_doc.end;
644 debug!("_next_sub result={:?}", r);
648 // variable-length unsigned integer with different tags.
649 // `first_tag` should be a tag for u8 or i8.
650 // `last_tag` should be the largest allowed integer tag with the matching signedness.
651 // all tags between them should be valid, in the order of u8, u16, u32 and u64.
652 fn _next_int(&mut self,
653 first_tag: EbmlEncoderTag,
654 last_tag: EbmlEncoderTag)
655 -> DecodeResult<u64> {
656 if self.pos >= self.parent.end {
657 return Err(Expected(format!("no more documents in current node!")));
660 let TaggedDoc { tag: r_tag, doc: r_doc } = try!(doc_at(self.parent.data, self.pos));
661 let r = if first_tag as usize <= r_tag && r_tag <= last_tag as usize {
662 match r_tag - first_tag as usize {
663 0 => doc_as_u8(r_doc) as u64,
664 1 => doc_as_u16(r_doc) as u64,
665 2 => doc_as_u32(r_doc) as u64,
666 3 => doc_as_u64(r_doc),
670 return Err(Expected(format!("expected EBML doc with tag {:?} through {:?} but \
676 if r_doc.end > self.parent.end {
677 return Err(Expected(format!("invalid EBML, child extends to {:#x}, parent to \
682 self.pos = r_doc.end;
683 debug!("_next_int({:?}, {:?}) result={:?}", first_tag, last_tag, r);
687 pub fn read_opaque<R, F>(&mut self, op: F) -> DecodeResult<R>
688 where F: FnOnce(&mut opaque::Decoder, Doc) -> DecodeResult<R>
690 let doc = try!(self.next_doc(EsOpaque));
693 let mut opaque_decoder = opaque::Decoder::new(doc.data, doc.start);
694 try!(op(&mut opaque_decoder, doc))
700 pub fn position(&self) -> usize {
704 pub fn advance(&mut self, bytes: usize) {
709 impl<'doc> serialize::Decoder for Decoder<'doc> {
711 fn read_nil(&mut self) -> DecodeResult<()> {
715 fn read_u64(&mut self) -> DecodeResult<u64> {
716 self._next_int(EsU8, EsU64)
718 fn read_u32(&mut self) -> DecodeResult<u32> {
719 Ok(try!(self._next_int(EsU8, EsU32)) as u32)
721 fn read_u16(&mut self) -> DecodeResult<u16> {
722 Ok(try!(self._next_int(EsU8, EsU16)) as u16)
724 fn read_u8(&mut self) -> DecodeResult<u8> {
725 Ok(doc_as_u8(try!(self.next_doc(EsU8))))
727 fn read_uint(&mut self) -> DecodeResult<usize> {
728 let v = try!(self._next_int(EsU8, EsU64));
729 if v > (::std::usize::MAX as u64) {
730 Err(IntTooBig(v as usize))
736 fn read_i64(&mut self) -> DecodeResult<i64> {
737 Ok(try!(self._next_int(EsI8, EsI64)) as i64)
739 fn read_i32(&mut self) -> DecodeResult<i32> {
740 Ok(try!(self._next_int(EsI8, EsI32)) as i32)
742 fn read_i16(&mut self) -> DecodeResult<i16> {
743 Ok(try!(self._next_int(EsI8, EsI16)) as i16)
745 fn read_i8(&mut self) -> DecodeResult<i8> {
746 Ok(doc_as_u8(try!(self.next_doc(EsI8))) as i8)
748 fn read_int(&mut self) -> DecodeResult<isize> {
749 let v = try!(self._next_int(EsI8, EsI64)) as i64;
750 if v > (isize::MAX as i64) || v < (isize::MIN as i64) {
751 debug!("FIXME \\#6122: Removing this makes this function miscompile");
752 Err(IntTooBig(v as usize))
758 fn read_bool(&mut self) -> DecodeResult<bool> {
759 Ok(doc_as_u8(try!(self.next_doc(EsBool))) != 0)
762 fn read_f64(&mut self) -> DecodeResult<f64> {
763 let bits = doc_as_u64(try!(self.next_doc(EsF64)));
764 Ok(unsafe { transmute(bits) })
766 fn read_f32(&mut self) -> DecodeResult<f32> {
767 let bits = doc_as_u32(try!(self.next_doc(EsF32)));
768 Ok(unsafe { transmute(bits) })
770 fn read_char(&mut self) -> DecodeResult<char> {
771 Ok(char::from_u32(doc_as_u32(try!(self.next_doc(EsChar)))).unwrap())
773 fn read_str(&mut self) -> DecodeResult<String> {
774 Ok(try!(self.next_doc(EsStr)).as_str())
778 fn read_enum<T, F>(&mut self, name: &str, f: F) -> DecodeResult<T>
779 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
781 debug!("read_enum({})", name);
783 let doc = try!(self.next_doc(EsEnum));
785 let (old_parent, old_pos) = (self.parent, self.pos);
787 self.pos = self.parent.start;
789 let result = try!(f(self));
791 self.parent = old_parent;
796 fn read_enum_variant<T, F>(&mut self, _: &[&str], mut f: F) -> DecodeResult<T>
797 where F: FnMut(&mut Decoder<'doc>, usize) -> DecodeResult<T>
799 debug!("read_enum_variant()");
800 let idx = try!(self._next_sub());
801 debug!(" idx={}", idx);
806 fn read_enum_variant_arg<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
807 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
809 debug!("read_enum_variant_arg(idx={})", idx);
813 fn read_enum_struct_variant<T, F>(&mut self, _: &[&str], mut f: F) -> DecodeResult<T>
814 where F: FnMut(&mut Decoder<'doc>, usize) -> DecodeResult<T>
816 debug!("read_enum_struct_variant()");
817 let idx = try!(self._next_sub());
818 debug!(" idx={}", idx);
823 fn read_enum_struct_variant_field<T, F>(&mut self,
828 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
830 debug!("read_enum_struct_variant_arg(name={}, idx={})", name, idx);
834 fn read_struct<T, F>(&mut self, name: &str, _: usize, f: F) -> DecodeResult<T>
835 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
837 debug!("read_struct(name={})", name);
841 fn read_struct_field<T, F>(&mut self, name: &str, idx: usize, f: F) -> DecodeResult<T>
842 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
844 debug!("read_struct_field(name={}, idx={})", name, idx);
848 fn read_tuple<T, F>(&mut self, tuple_len: usize, f: F) -> DecodeResult<T>
849 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
851 debug!("read_tuple()");
852 self.read_seq(move |d, len| {
853 if len == tuple_len {
856 Err(Expected(format!("Expected tuple of length `{}`, found tuple of length \
864 fn read_tuple_arg<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
865 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
867 debug!("read_tuple_arg(idx={})", idx);
868 self.read_seq_elt(idx, f)
871 fn read_tuple_struct<T, F>(&mut self, name: &str, len: usize, f: F) -> DecodeResult<T>
872 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
874 debug!("read_tuple_struct(name={})", name);
875 self.read_tuple(len, f)
878 fn read_tuple_struct_arg<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
879 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
881 debug!("read_tuple_struct_arg(idx={})", idx);
882 self.read_tuple_arg(idx, f)
885 fn read_option<T, F>(&mut self, mut f: F) -> DecodeResult<T>
886 where F: FnMut(&mut Decoder<'doc>, bool) -> DecodeResult<T>
888 debug!("read_option()");
889 self.read_enum("Option", move |this| {
890 this.read_enum_variant(&["None", "Some"], move |this, idx| {
894 _ => Err(Expected(format!("Expected None or Some"))),
900 fn read_seq<T, F>(&mut self, f: F) -> DecodeResult<T>
901 where F: FnOnce(&mut Decoder<'doc>, usize) -> DecodeResult<T>
903 debug!("read_seq()");
904 self.push_doc(EsVec, move |d| {
905 let len = try!(d._next_sub());
906 debug!(" len={}", len);
911 fn read_seq_elt<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
912 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
914 debug!("read_seq_elt(idx={})", idx);
915 self.push_doc(EsVecElt, f)
918 fn read_map<T, F>(&mut self, f: F) -> DecodeResult<T>
919 where F: FnOnce(&mut Decoder<'doc>, usize) -> DecodeResult<T>
921 debug!("read_map()");
922 self.push_doc(EsMap, move |d| {
923 let len = try!(d._next_sub());
924 debug!(" len={}", len);
929 fn read_map_elt_key<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
930 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
932 debug!("read_map_elt_key(idx={})", idx);
933 self.push_doc(EsMapKey, f)
936 fn read_map_elt_val<T, F>(&mut self, idx: usize, f: F) -> DecodeResult<T>
937 where F: FnOnce(&mut Decoder<'doc>) -> DecodeResult<T>
939 debug!("read_map_elt_val(idx={})", idx);
940 self.push_doc(EsMapVal, f)
943 fn error(&mut self, err: &str) -> Error {
944 ApplicationError(err.to_string())
951 use std::io::prelude::*;
952 use std::io::{self, SeekFrom, Cursor};
955 use super::{EsVec, EsMap, EsEnum, EsSub8, EsSub32, EsVecElt, EsMapKey, EsU64, EsU32, EsU16,
956 EsU8, EsI64, EsI32, EsI16, EsI8, EsBool, EsF64, EsF32, EsChar, EsStr, EsMapVal,
957 EsOpaque, NUM_IMPLICIT_TAGS, NUM_TAGS};
962 pub type EncodeResult = io::Result<()>;
965 pub struct Encoder<'a> {
966 pub writer: &'a mut Cursor<Vec<u8>>,
967 size_positions: Vec<u64>,
968 relax_limit: u64, // do not move encoded bytes before this position
971 fn write_tag<W: Write>(w: &mut W, n: usize) -> EncodeResult {
973 w.write_all(&[n as u8])
974 } else if 0x100 <= n && n < NUM_TAGS {
975 w.write_all(&[0xf0 | (n >> 8) as u8, n as u8])
977 Err(io::Error::new(io::ErrorKind::Other, &format!("invalid tag: {}", n)[..]))
981 fn write_sized_vuint<W: Write>(w: &mut W, n: usize, size: usize) -> EncodeResult {
983 1 => w.write_all(&[0x80 | (n as u8)]),
984 2 => w.write_all(&[0x40 | ((n >> 8) as u8), n as u8]),
985 3 => w.write_all(&[0x20 | ((n >> 16) as u8), (n >> 8) as u8, n as u8]),
986 4 => w.write_all(&[0x10 | ((n >> 24) as u8), (n >> 16) as u8, (n >> 8) as u8, n as u8]),
987 _ => Err(io::Error::new(io::ErrorKind::Other, &format!("isize too big: {}", n)[..])),
991 pub fn write_vuint<W: Write>(w: &mut W, n: usize) -> EncodeResult {
993 return write_sized_vuint(w, n, 1);
996 return write_sized_vuint(w, n, 2);
999 return write_sized_vuint(w, n, 3);
1002 return write_sized_vuint(w, n, 4);
1004 Err(io::Error::new(io::ErrorKind::Other, &format!("isize too big: {}", n)[..]))
1007 impl<'a> Encoder<'a> {
1008 pub fn new(w: &'a mut Cursor<Vec<u8>>) -> Encoder<'a> {
1011 size_positions: vec![],
1016 pub fn start_tag(&mut self, tag_id: usize) -> EncodeResult {
1017 debug!("Start tag {:?}", tag_id);
1018 assert!(tag_id >= NUM_IMPLICIT_TAGS);
1020 // Write the enum ID:
1021 try!(write_tag(self.writer, tag_id));
1023 // Write a placeholder four-byte size.
1024 let cur_pos = try!(self.writer.seek(SeekFrom::Current(0)));
1025 self.size_positions.push(cur_pos);
1026 let zeroes: &[u8] = &[0, 0, 0, 0];
1027 self.writer.write_all(zeroes)
1030 pub fn end_tag(&mut self) -> EncodeResult {
1031 let last_size_pos = self.size_positions.pop().unwrap();
1032 let cur_pos = try!(self.writer.seek(SeekFrom::Current(0)));
1033 try!(self.writer.seek(SeekFrom::Start(last_size_pos)));
1034 let size = (cur_pos - last_size_pos - 4) as usize;
1036 // relax the size encoding for small tags (bigger tags are costly to move).
1037 // we should never try to move the stable positions, however.
1038 const RELAX_MAX_SIZE: usize = 0x100;
1039 if size <= RELAX_MAX_SIZE && last_size_pos >= self.relax_limit {
1040 // we can't alter the buffer in place, so have a temporary buffer
1041 let mut buf = [0u8; RELAX_MAX_SIZE];
1043 let last_size_pos = last_size_pos as usize;
1044 let data = &self.writer.get_ref()[last_size_pos + 4..cur_pos as usize];
1045 buf[..size].clone_from_slice(data);
1048 // overwrite the size and data and continue
1049 try!(write_vuint(self.writer, size));
1050 try!(self.writer.write_all(&buf[..size]));
1052 // overwrite the size with an overlong encoding and skip past the data
1053 try!(write_sized_vuint(self.writer, size, 4));
1054 try!(self.writer.seek(SeekFrom::Start(cur_pos)));
1057 debug!("End tag (size = {:?})", size);
1061 pub fn wr_tag<F>(&mut self, tag_id: usize, blk: F) -> EncodeResult
1062 where F: FnOnce() -> EncodeResult
1064 try!(self.start_tag(tag_id));
1069 pub fn wr_tagged_bytes(&mut self, tag_id: usize, b: &[u8]) -> EncodeResult {
1070 assert!(tag_id >= NUM_IMPLICIT_TAGS);
1071 try!(write_tag(self.writer, tag_id));
1072 try!(write_vuint(self.writer, b.len()));
1073 self.writer.write_all(b)
1076 pub fn wr_tagged_u64(&mut self, tag_id: usize, v: u64) -> EncodeResult {
1077 let bytes: [u8; 8] = unsafe { mem::transmute(v.to_be()) };
1078 // tagged integers are emitted in big-endian, with no
1080 let leading_zero_bytes = v.leading_zeros() / 8;
1081 self.wr_tagged_bytes(tag_id, &bytes[leading_zero_bytes as usize..])
1085 pub fn wr_tagged_u32(&mut self, tag_id: usize, v: u32) -> EncodeResult {
1086 self.wr_tagged_u64(tag_id, v as u64)
1090 pub fn wr_tagged_u16(&mut self, tag_id: usize, v: u16) -> EncodeResult {
1091 self.wr_tagged_u64(tag_id, v as u64)
1095 pub fn wr_tagged_u8(&mut self, tag_id: usize, v: u8) -> EncodeResult {
1096 self.wr_tagged_bytes(tag_id, &[v])
1100 pub fn wr_tagged_i64(&mut self, tag_id: usize, v: i64) -> EncodeResult {
1101 self.wr_tagged_u64(tag_id, v as u64)
1105 pub fn wr_tagged_i32(&mut self, tag_id: usize, v: i32) -> EncodeResult {
1106 self.wr_tagged_u32(tag_id, v as u32)
1110 pub fn wr_tagged_i16(&mut self, tag_id: usize, v: i16) -> EncodeResult {
1111 self.wr_tagged_u16(tag_id, v as u16)
1115 pub fn wr_tagged_i8(&mut self, tag_id: usize, v: i8) -> EncodeResult {
1116 self.wr_tagged_bytes(tag_id, &[v as u8])
1119 pub fn wr_tagged_str(&mut self, tag_id: usize, v: &str) -> EncodeResult {
1120 self.wr_tagged_bytes(tag_id, v.as_bytes())
1123 // for auto-serialization
1124 fn wr_tagged_raw_bytes(&mut self, tag_id: usize, b: &[u8]) -> EncodeResult {
1125 try!(write_tag(self.writer, tag_id));
1126 self.writer.write_all(b)
1129 fn wr_tagged_raw_u64(&mut self, tag_id: usize, v: u64) -> EncodeResult {
1130 let bytes: [u8; 8] = unsafe { mem::transmute(v.to_be()) };
1131 self.wr_tagged_raw_bytes(tag_id, &bytes)
1134 fn wr_tagged_raw_u32(&mut self, tag_id: usize, v: u32) -> EncodeResult {
1135 let bytes: [u8; 4] = unsafe { mem::transmute(v.to_be()) };
1136 self.wr_tagged_raw_bytes(tag_id, &bytes)
1139 fn wr_tagged_raw_u16(&mut self, tag_id: usize, v: u16) -> EncodeResult {
1140 let bytes: [u8; 2] = unsafe { mem::transmute(v.to_be()) };
1141 self.wr_tagged_raw_bytes(tag_id, &bytes)
1144 fn wr_tagged_raw_u8(&mut self, tag_id: usize, v: u8) -> EncodeResult {
1145 self.wr_tagged_raw_bytes(tag_id, &[v])
1148 fn wr_tagged_raw_i64(&mut self, tag_id: usize, v: i64) -> EncodeResult {
1149 self.wr_tagged_raw_u64(tag_id, v as u64)
1152 fn wr_tagged_raw_i32(&mut self, tag_id: usize, v: i32) -> EncodeResult {
1153 self.wr_tagged_raw_u32(tag_id, v as u32)
1156 fn wr_tagged_raw_i16(&mut self, tag_id: usize, v: i16) -> EncodeResult {
1157 self.wr_tagged_raw_u16(tag_id, v as u16)
1160 fn wr_tagged_raw_i8(&mut self, tag_id: usize, v: i8) -> EncodeResult {
1161 self.wr_tagged_raw_bytes(tag_id, &[v as u8])
1164 pub fn wr_bytes(&mut self, b: &[u8]) -> EncodeResult {
1165 debug!("Write {:?} bytes", b.len());
1166 self.writer.write_all(b)
1169 pub fn wr_str(&mut self, s: &str) -> EncodeResult {
1170 debug!("Write str: {:?}", s);
1171 self.writer.write_all(s.as_bytes())
1174 /// Returns the current position while marking it stable, i.e.
1175 /// generated bytes so far wouldn't be affected by relaxation.
1176 pub fn mark_stable_position(&mut self) -> u64 {
1177 let pos = self.writer.seek(SeekFrom::Current(0)).unwrap();
1178 if self.relax_limit < pos {
1179 self.relax_limit = pos;
1185 impl<'a> Encoder<'a> {
1186 // used internally to emit things like the vector length and so on
1187 fn _emit_tagged_sub(&mut self, v: usize) -> EncodeResult {
1188 if v as u8 as usize == v {
1189 self.wr_tagged_raw_u8(EsSub8 as usize, v as u8)
1190 } else if v as u32 as usize == v {
1191 self.wr_tagged_raw_u32(EsSub32 as usize, v as u32)
1193 Err(io::Error::new(io::ErrorKind::Other,
1194 &format!("length or variant id too big: {}", v)[..]))
1198 pub fn emit_opaque<F>(&mut self, f: F) -> EncodeResult
1199 where F: FnOnce(&mut opaque::Encoder) -> EncodeResult
1201 try!(self.start_tag(EsOpaque as usize));
1204 let mut opaque_encoder = opaque::Encoder::new(self.writer);
1205 try!(f(&mut opaque_encoder));
1208 self.mark_stable_position();
1213 impl<'a> serialize::Encoder for Encoder<'a> {
1214 type Error = io::Error;
1216 fn emit_nil(&mut self) -> EncodeResult {
1220 fn emit_uint(&mut self, v: usize) -> EncodeResult {
1221 self.emit_u64(v as u64)
1223 fn emit_u64(&mut self, v: u64) -> EncodeResult {
1224 if v as u32 as u64 == v {
1225 self.emit_u32(v as u32)
1227 self.wr_tagged_raw_u64(EsU64 as usize, v)
1230 fn emit_u32(&mut self, v: u32) -> EncodeResult {
1231 if v as u16 as u32 == v {
1232 self.emit_u16(v as u16)
1234 self.wr_tagged_raw_u32(EsU32 as usize, v)
1237 fn emit_u16(&mut self, v: u16) -> EncodeResult {
1238 if v as u8 as u16 == v {
1239 self.emit_u8(v as u8)
1241 self.wr_tagged_raw_u16(EsU16 as usize, v)
1244 fn emit_u8(&mut self, v: u8) -> EncodeResult {
1245 self.wr_tagged_raw_u8(EsU8 as usize, v)
1248 fn emit_int(&mut self, v: isize) -> EncodeResult {
1249 self.emit_i64(v as i64)
1251 fn emit_i64(&mut self, v: i64) -> EncodeResult {
1252 if v as i32 as i64 == v {
1253 self.emit_i32(v as i32)
1255 self.wr_tagged_raw_i64(EsI64 as usize, v)
1258 fn emit_i32(&mut self, v: i32) -> EncodeResult {
1259 if v as i16 as i32 == v {
1260 self.emit_i16(v as i16)
1262 self.wr_tagged_raw_i32(EsI32 as usize, v)
1265 fn emit_i16(&mut self, v: i16) -> EncodeResult {
1266 if v as i8 as i16 == v {
1267 self.emit_i8(v as i8)
1269 self.wr_tagged_raw_i16(EsI16 as usize, v)
1272 fn emit_i8(&mut self, v: i8) -> EncodeResult {
1273 self.wr_tagged_raw_i8(EsI8 as usize, v)
1276 fn emit_bool(&mut self, v: bool) -> EncodeResult {
1277 self.wr_tagged_raw_u8(EsBool as usize, v as u8)
1280 fn emit_f64(&mut self, v: f64) -> EncodeResult {
1281 let bits = unsafe { mem::transmute(v) };
1282 self.wr_tagged_raw_u64(EsF64 as usize, bits)
1284 fn emit_f32(&mut self, v: f32) -> EncodeResult {
1285 let bits = unsafe { mem::transmute(v) };
1286 self.wr_tagged_raw_u32(EsF32 as usize, bits)
1288 fn emit_char(&mut self, v: char) -> EncodeResult {
1289 self.wr_tagged_raw_u32(EsChar as usize, v as u32)
1292 fn emit_str(&mut self, v: &str) -> EncodeResult {
1293 self.wr_tagged_str(EsStr as usize, v)
1296 fn emit_enum<F>(&mut self, _name: &str, f: F) -> EncodeResult
1297 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1299 try!(self.start_tag(EsEnum as usize));
1304 fn emit_enum_variant<F>(&mut self, _: &str, v_id: usize, _: usize, f: F) -> EncodeResult
1305 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1307 try!(self._emit_tagged_sub(v_id));
1311 fn emit_enum_variant_arg<F>(&mut self, _: usize, f: F) -> EncodeResult
1312 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1317 fn emit_enum_struct_variant<F>(&mut self,
1323 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1325 self.emit_enum_variant(v_name, v_id, cnt, f)
1328 fn emit_enum_struct_variant_field<F>(&mut self, _: &str, idx: usize, f: F) -> EncodeResult
1329 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1331 self.emit_enum_variant_arg(idx, f)
1334 fn emit_struct<F>(&mut self, _: &str, _len: usize, f: F) -> EncodeResult
1335 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1340 fn emit_struct_field<F>(&mut self, _name: &str, _: usize, f: F) -> EncodeResult
1341 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1346 fn emit_tuple<F>(&mut self, len: usize, f: F) -> EncodeResult
1347 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1349 self.emit_seq(len, f)
1351 fn emit_tuple_arg<F>(&mut self, idx: usize, f: F) -> EncodeResult
1352 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1354 self.emit_seq_elt(idx, f)
1357 fn emit_tuple_struct<F>(&mut self, _: &str, len: usize, f: F) -> EncodeResult
1358 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1360 self.emit_seq(len, f)
1362 fn emit_tuple_struct_arg<F>(&mut self, idx: usize, f: F) -> EncodeResult
1363 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1365 self.emit_seq_elt(idx, f)
1368 fn emit_option<F>(&mut self, f: F) -> EncodeResult
1369 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1371 self.emit_enum("Option", f)
1373 fn emit_option_none(&mut self) -> EncodeResult {
1374 self.emit_enum_variant("None", 0, 0, |_| Ok(()))
1376 fn emit_option_some<F>(&mut self, f: F) -> EncodeResult
1377 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1380 self.emit_enum_variant("Some", 1, 1, f)
1383 fn emit_seq<F>(&mut self, len: usize, f: F) -> EncodeResult
1384 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1387 // empty vector optimization
1388 return self.wr_tagged_bytes(EsVec as usize, &[]);
1391 try!(self.start_tag(EsVec as usize));
1392 try!(self._emit_tagged_sub(len));
1397 fn emit_seq_elt<F>(&mut self, _idx: usize, f: F) -> EncodeResult
1398 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1401 try!(self.start_tag(EsVecElt as usize));
1406 fn emit_map<F>(&mut self, len: usize, f: F) -> EncodeResult
1407 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1410 // empty map optimization
1411 return self.wr_tagged_bytes(EsMap as usize, &[]);
1414 try!(self.start_tag(EsMap as usize));
1415 try!(self._emit_tagged_sub(len));
1420 fn emit_map_elt_key<F>(&mut self, _idx: usize, f: F) -> EncodeResult
1421 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1424 try!(self.start_tag(EsMapKey as usize));
1429 fn emit_map_elt_val<F>(&mut self, _idx: usize, f: F) -> EncodeResult
1430 where F: FnOnce(&mut Encoder<'a>) -> EncodeResult
1432 try!(self.start_tag(EsMapVal as usize));
1439 // ___________________________________________________________________________
1444 use super::{Doc, reader, writer};
1446 use serialize::{Encodable, Decodable};
1448 use std::io::Cursor;
1451 fn test_vuint_at() {
1459 0x10, 0x00, 0x00, 0x00,
1460 0x1f, 0xff, 0xff, 0xff
1463 let mut res: reader::Res;
1466 res = reader::vuint_at(data, 0).unwrap();
1467 assert_eq!(res.val, 0);
1468 assert_eq!(res.next, 1);
1469 res = reader::vuint_at(data, res.next).unwrap();
1470 assert_eq!(res.val, (1 << 7) - 1);
1471 assert_eq!(res.next, 2);
1474 res = reader::vuint_at(data, res.next).unwrap();
1475 assert_eq!(res.val, 0);
1476 assert_eq!(res.next, 4);
1477 res = reader::vuint_at(data, res.next).unwrap();
1478 assert_eq!(res.val, (1 << 14) - 1);
1479 assert_eq!(res.next, 6);
1482 res = reader::vuint_at(data, res.next).unwrap();
1483 assert_eq!(res.val, 0);
1484 assert_eq!(res.next, 9);
1485 res = reader::vuint_at(data, res.next).unwrap();
1486 assert_eq!(res.val, (1 << 21) - 1);
1487 assert_eq!(res.next, 12);
1490 res = reader::vuint_at(data, res.next).unwrap();
1491 assert_eq!(res.val, 0);
1492 assert_eq!(res.next, 16);
1493 res = reader::vuint_at(data, res.next).unwrap();
1494 assert_eq!(res.val, (1 << 28) - 1);
1495 assert_eq!(res.next, 20);
1499 fn test_option_int() {
1500 fn test_v(v: Option<isize>) {
1501 debug!("v == {:?}", v);
1502 let mut wr = Cursor::new(Vec::new());
1504 let mut rbml_w = writer::Encoder::new(&mut wr);
1505 let _ = v.encode(&mut rbml_w);
1507 let rbml_doc = Doc::new(wr.get_ref());
1508 let mut deser = reader::Decoder::new(rbml_doc);
1509 let v1 = Decodable::decode(&mut deser).unwrap();
1510 debug!("v1 == {:?}", v1);
1522 #![allow(non_snake_case)]
1527 pub fn vuint_at_A_aligned(b: &mut Bencher) {
1528 let data = (0..4 * 100)
1535 .collect::<Vec<_>>();
1539 while i < data.len() {
1540 sum += reader::vuint_at(&data, i).unwrap().val;
1547 pub fn vuint_at_A_unaligned(b: &mut Bencher) {
1548 let data = (0..4 * 100 + 1)
1555 .collect::<Vec<_>>();
1559 while i < data.len() {
1560 sum += reader::vuint_at(&data, i).unwrap().val;
1567 pub fn vuint_at_D_aligned(b: &mut Bencher) {
1568 let data = (0..4 * 100)
1576 .collect::<Vec<_>>();
1580 while i < data.len() {
1581 sum += reader::vuint_at(&data, i).unwrap().val;
1588 pub fn vuint_at_D_unaligned(b: &mut Bencher) {
1589 let data = (0..4 * 100 + 1)
1597 .collect::<Vec<_>>();
1601 while i < data.len() {
1602 sum += reader::vuint_at(&data, i).unwrap().val;