Auto merge of #22235 - michaelwoerister:cross-crate-spans, r=michaelwoerister

This allows to create proper debuginfo line information for items inlined from other crates (e.g. instantiations of generics). Only the codemap's 'metadata' is stored in a crate's metadata. That is, just filename, positions of line-beginnings, etc. but not the actual source code itself.

Crate metadata size is increased by this change because spans in the encoded ASTs take up space now:
```
                BEFORE    AFTER
libcore         36 MiB    39.6 MiB    +10%
libsyntax       51.1 MiB  60.5 MiB    +18.4%
libcollections  11.2 MiB  12.8 MiB    +14.3%
```
This only affects binaries containing metadata (rlibs and dylibs), executables should not be affected in size. 

Fixes #19228 and probably #22226.

Author: bors

src/librustc/metadata/common.rs

@@ -252,3 +252,6 @@ pub const tag_macro_def: uint = 0x9e;
 pub const tag_macro_def_body: uint = 0x9f;
 
 pub const tag_paren_sugar: uint = 0xa0;
+
+pub const tag_codemap: uint = 0xa1;
+pub const tag_codemap_filemap: uint = 0xa2;

src/librustc/metadata/creader.rs

@@ -26,7 +26,7 @@ use syntax::ast;
 use syntax::abi;
 use syntax::attr;
 use syntax::attr::AttrMetaMethods;
-use syntax::codemap::{Span, mk_sp};
+use syntax::codemap::{self, Span, mk_sp, Pos};
 use syntax::parse;
 use syntax::parse::token::InternedString;
 use syntax::parse::token;
@@ -373,15 +373,17 @@ impl<'a> CrateReader<'a> {
         // Maintain a reference to the top most crate.
         let root = if root.is_some() { root } else { &crate_paths };
 
-        let cnum_map = self.resolve_crate_deps(root, lib.metadata.as_slice(), span);
+        let loader::Library { dylib, rlib, metadata } = lib;
 
-        let loader::Library{ dylib, rlib, metadata } = lib;
+        let cnum_map = self.resolve_crate_deps(root, metadata.as_slice(), span);
+        let codemap_import_info = import_codemap(self.sess.codemap(), &metadata);
 
         let cmeta = Rc::new( cstore::crate_metadata {
             name: name.to_string(),
             data: metadata,
             cnum_map: cnum_map,
             cnum: cnum,
+            codemap_import_info: codemap_import_info,
             span: span,
         });
 
@@ -586,3 +588,131 @@ impl<'a> CrateReader<'a> {
         }
     }
 }
+
+/// Imports the codemap from an external crate into the codemap of the crate
+/// currently being compiled (the "local crate").
+///
+/// The import algorithm works analogous to how AST items are inlined from an
+/// external crate's metadata:
+/// For every FileMap in the external codemap an 'inline' copy is created in the
+/// local codemap. The correspondence relation between external and local
+/// FileMaps is recorded in the `ImportedFileMap` objects returned from this
+/// function. When an item from an external crate is later inlined into this
+/// crate, this correspondence information is used to translate the span
+/// information of the inlined item so that it refers the correct positions in
+/// the local codemap (see `astencode::DecodeContext::tr_span()`).
+///
+/// The import algorithm in the function below will reuse FileMaps already
+/// existing in the local codemap. For example, even if the FileMap of some
+/// source file of libstd gets imported many times, there will only ever be
+/// one FileMap object for the corresponding file in the local codemap.
+///
+/// Note that imported FileMaps do not actually contain the source code of the
+/// file they represent, just information about length, line breaks, and
+/// multibyte characters. This information is enough to generate valid debuginfo
+/// for items inlined from other crates.
+fn import_codemap(local_codemap: &codemap::CodeMap,
+                  metadata: &MetadataBlob)
+                  -> Vec<cstore::ImportedFileMap> {
+    let external_codemap = decoder::get_imported_filemaps(metadata.as_slice());
+
+    let imported_filemaps = external_codemap.into_iter().map(|filemap_to_import| {
+        // Try to find an existing FileMap that can be reused for the filemap to
+        // be imported. A FileMap is reusable if it is exactly the same, just
+        // positioned at a different offset within the codemap.
+        let reusable_filemap = {
+            local_codemap.files
+                         .borrow()
+                         .iter()
+                         .find(|fm| are_equal_modulo_startpos(&fm, &filemap_to_import))
+                         .map(|rc| rc.clone())
+        };
+
+        match reusable_filemap {
+            Some(fm) => {
+                cstore::ImportedFileMap {
+                    original_start_pos: filemap_to_import.start_pos,
+                    original_end_pos: filemap_to_import.end_pos,
+                    translated_filemap: fm
+                }
+            }
+            None => {
+                // We can't reuse an existing FileMap, so allocate a new one
+                // containing the information we need.
+                let codemap::FileMap {
+                    name,
+                    start_pos,
+                    end_pos,
+                    lines,
+                    multibyte_chars,
+                    ..
+                } = filemap_to_import;
+
+                let source_length = (end_pos - start_pos).to_usize();
+
+                // Translate line-start positions and multibyte character
+                // position into frame of reference local to file.
+                // `CodeMap::new_imported_filemap()` will then translate those
+                // coordinates to their new global frame of reference when the
+                // offset of the FileMap is known.
+                let lines = lines.into_inner().map_in_place(|pos| pos - start_pos);
+                let multibyte_chars = multibyte_chars
+                    .into_inner()
+                    .map_in_place(|mbc|
+                        codemap::MultiByteChar {
+                            pos: mbc.pos + start_pos,
+                            bytes: mbc.bytes
+                        });
+
+                let local_version = local_codemap.new_imported_filemap(name,
+                                                                       source_length,
+                                                                       lines,
+                                                                       multibyte_chars);
+                cstore::ImportedFileMap {
+                    original_start_pos: start_pos,
+                    original_end_pos: end_pos,
+                    translated_filemap: local_version
+                }
+            }
+        }
+    }).collect();
+
+    return imported_filemaps;
+
+    fn are_equal_modulo_startpos(fm1: &codemap::FileMap,
+                                 fm2: &codemap::FileMap)
+                                 -> bool {
+        if fm1.name != fm2.name {
+            return false;
+        }
+
+        let lines1 = fm1.lines.borrow();
+        let lines2 = fm2.lines.borrow();
+
+        if lines1.len() != lines2.len() {
+            return false;
+        }
+
+        for (&line1, &line2) in lines1.iter().zip(lines2.iter()) {
+            if (line1 - fm1.start_pos) != (line2 - fm2.start_pos) {
+                return false;
+            }
+        }
+
+        let multibytes1 = fm1.multibyte_chars.borrow();
+        let multibytes2 = fm2.multibyte_chars.borrow();
+
+        if multibytes1.len() != multibytes2.len() {
+            return false;
+        }
+
+        for (mb1, mb2) in multibytes1.iter().zip(multibytes2.iter()) {
+            if (mb1.bytes != mb2.bytes) ||
+               ((mb1.pos - fm1.start_pos) != (mb2.pos - fm2.start_pos)) {
+                return false;
+            }
+        }
+
+        true
+    }
+}

src/librustc/metadata/cstore.rs

@@ -27,7 +27,7 @@ use std::cell::RefCell;
 use std::rc::Rc;
 use flate::Bytes;
 use syntax::ast;
-use syntax::codemap::Span;
+use syntax::codemap;
 use syntax::parse::token::IdentInterner;
 
 // A map from external crate numbers (as decoded from some crate file) to
@@ -41,12 +41,24 @@ pub enum MetadataBlob {
     MetadataArchive(loader::ArchiveMetadata),
 }
 
+/// Holds information about a codemap::FileMap imported from another crate.
+/// See creader::import_codemap() for more information.
+pub struct ImportedFileMap {
+    /// This FileMap's byte-offset within the codemap of its original crate
+    pub original_start_pos: codemap::BytePos,
+    /// The end of this FileMap within the codemap of its original crate
+    pub original_end_pos: codemap::BytePos,
+    /// The imported FileMap's representation within the local codemap
+    pub translated_filemap: Rc<codemap::FileMap>
+}
+
 pub struct crate_metadata {
     pub name: String,
     pub data: MetadataBlob,
     pub cnum_map: cnum_map,
     pub cnum: ast::CrateNum,
-    pub span: Span,
+    pub codemap_import_info: Vec<ImportedFileMap>,
+    pub span: codemap::Span,
 }
 
 #[derive(Copy, Debug, PartialEq, Clone)]

src/librustc/metadata/decoder.rs

@@ -1561,11 +1561,26 @@ pub fn is_associated_type(cdata: Cmd, id: ast::NodeId) -> bool {
     }
 }
 
-
 pub fn is_default_trait<'tcx>(cdata: Cmd, id: ast::NodeId) -> bool {
     let item_doc = lookup_item(id, cdata.data());
     match item_family(item_doc) {
         Family::DefaultImpl => true,
         _ => false
     }
 }
+
+pub fn get_imported_filemaps(metadata: &[u8]) -> Vec<codemap::FileMap> {
+    let crate_doc = rbml::Doc::new(metadata);
+    let cm_doc = reader::get_doc(crate_doc, tag_codemap);
+
+    let mut filemaps = vec![];
+
+    reader::tagged_docs(cm_doc, tag_codemap_filemap, |filemap_doc| {
+        let mut decoder = reader::Decoder::new(filemap_doc);
+        let filemap: codemap::FileMap = Decodable::decode(&mut decoder).unwrap();
+        filemaps.push(filemap);
+        true
+    });
+
+    return filemaps;
+}

src/librustc/metadata/encoder.rs

@@ -1751,6 +1751,28 @@ fn encode_plugin_registrar_fn(ecx: &EncodeContext, rbml_w: &mut Encoder) {
     }
 }
 
+fn encode_codemap(ecx: &EncodeContext, rbml_w: &mut Encoder) {
+    rbml_w.start_tag(tag_codemap);
+    let codemap = ecx.tcx.sess.codemap();
+
+    for filemap in &codemap.files.borrow()[..] {
+
+        if filemap.lines.borrow().len() == 0 || filemap.is_imported() {
+            // No need to export empty filemaps, as they can't contain spans
+            // that need translation.
+            // Also no need to re-export imported filemaps, as any downstream
+            // crate will import them from their original source.
+            continue;
+        }
+
+        rbml_w.start_tag(tag_codemap_filemap);
+        filemap.encode(rbml_w);
+        rbml_w.end_tag();
+    }
+
+    rbml_w.end_tag();
+}
+
 /// Serialize the text of the exported macros
 fn encode_macro_defs(rbml_w: &mut Encoder,
                      krate: &ast::Crate) {
@@ -1968,6 +1990,7 @@ fn encode_metadata_inner(wr: &mut SeekableMemWriter,
         lang_item_bytes: u64,
         native_lib_bytes: u64,
         plugin_registrar_fn_bytes: u64,
+        codemap_bytes: u64,
         macro_defs_bytes: u64,
         impl_bytes: u64,
         misc_bytes: u64,
@@ -1982,6 +2005,7 @@ fn encode_metadata_inner(wr: &mut SeekableMemWriter,
         lang_item_bytes: 0,
         native_lib_bytes: 0,
         plugin_registrar_fn_bytes: 0,
+        codemap_bytes: 0,
         macro_defs_bytes: 0,
         impl_bytes: 0,
         misc_bytes: 0,
@@ -2047,6 +2071,11 @@ fn encode_metadata_inner(wr: &mut SeekableMemWriter,
     encode_plugin_registrar_fn(&ecx, &mut rbml_w);
     stats.plugin_registrar_fn_bytes = rbml_w.writer.tell().unwrap() - i;
 
+    // Encode codemap
+    i = rbml_w.writer.tell().unwrap();
+    encode_codemap(&ecx, &mut rbml_w);
+    stats.codemap_bytes = rbml_w.writer.tell().unwrap() - i;
+
     // Encode macro definitions
     i = rbml_w.writer.tell().unwrap();
     encode_macro_defs(&mut rbml_w, krate);
@@ -2091,6 +2120,7 @@ fn encode_metadata_inner(wr: &mut SeekableMemWriter,
         println!("       lang item bytes: {}", stats.lang_item_bytes);
         println!("          native bytes: {}", stats.native_lib_bytes);
         println!("plugin registrar bytes: {}", stats.plugin_registrar_fn_bytes);
+        println!("         codemap bytes: {}", stats.codemap_bytes);
         println!("       macro def bytes: {}", stats.macro_defs_bytes);
         println!("            impl bytes: {}", stats.impl_bytes);
         println!("            misc bytes: {}", stats.misc_bytes);

src/librustc/middle/astencode.rs

@@ -42,6 +42,7 @@ use syntax;
 use std::old_io::Seek;
 use std::num::FromPrimitive;
 use std::rc::Rc;
+use std::cell::Cell;
 
 use rbml::reader;
 use rbml::writer::Encoder;
@@ -58,7 +59,9 @@ struct DecodeContext<'a, 'b, 'tcx: 'a> {
     tcx: &'a ty::ctxt<'tcx>,
     cdata: &'b cstore::crate_metadata,
     from_id_range: ast_util::IdRange,
-    to_id_range: ast_util::IdRange
+    to_id_range: ast_util::IdRange,
+    // Cache the last used filemap for translating spans as an optimization.
+    last_filemap_index: Cell<usize>,
 }
 
 trait tr {
@@ -120,6 +123,8 @@ impl<'a, 'b, 'c, 'tcx> ast_map::FoldOps for &'a DecodeContext<'b, 'c, 'tcx> {
     }
 }
 
+/// Decodes an item from its AST in the cdata's metadata and adds it to the
+/// ast-map.
 pub fn decode_inlined_item<'tcx>(cdata: &cstore::crate_metadata,
                                  tcx: &ty::ctxt<'tcx>,
                                  path: Vec<ast_map::PathElem>,
@@ -143,7 +148,8 @@ pub fn decode_inlined_item<'tcx>(cdata: &cstore::crate_metadata,
             cdata: cdata,
             tcx: tcx,
             from_id_range: from_id_range,
-            to_id_range: to_id_range
+            to_id_range: to_id_range,
+            last_filemap_index: Cell::new(0)
         };
         let raw_ii = decode_ast(ast_doc);
         let ii = ast_map::map_decoded_item(&dcx.tcx.map, path, raw_ii, dcx);
@@ -234,8 +240,47 @@ impl<'a, 'b, 'tcx> DecodeContext<'a, 'b, 'tcx> {
         assert_eq!(did.krate, ast::LOCAL_CRATE);
         ast::DefId { krate: ast::LOCAL_CRATE, node: self.tr_id(did.node) }
     }
-    pub fn tr_span(&self, _span: Span) -> Span {
-        codemap::DUMMY_SP // FIXME (#1972): handle span properly
+
+    /// Translates a `Span` from an extern crate to the corresponding `Span`
+    /// within the local crate's codemap. `creader::import_codemap()` will
+    /// already have allocated any additionally needed FileMaps in the local
+    /// codemap as a side-effect of creating the crate_metadata's
+    /// `codemap_import_info`.
+    pub fn tr_span(&self, span: Span) -> Span {
+        let imported_filemaps = &self.cdata.codemap_import_info[..];
+
+        let filemap_index = {
+            // Optimize for the case that most spans within a translated item
+            // originate from the same filemap.
+            let last_filemap_index = self.last_filemap_index.get();
+
+            if span.lo >= imported_filemaps[last_filemap_index].original_start_pos &&
+               span.hi <= imported_filemaps[last_filemap_index].original_end_pos {
+                last_filemap_index
+            } else {
+                let mut a = 0;
+                let mut b = imported_filemaps.len();
+
+                while b - a > 1 {
+                    let m = (a + b) / 2;
+                    if imported_filemaps[m].original_start_pos > span.lo {
+                        b = m;
+                    } else {
+                        a = m;
+                    }
+                }
+
+                self.last_filemap_index.set(a);
+                a
+            }
+        };
+
+        let lo = (span.lo - imported_filemaps[filemap_index].original_start_pos) +
+                  imported_filemaps[filemap_index].translated_filemap.start_pos;
+        let hi = (span.hi - imported_filemaps[filemap_index].original_start_pos) +
+                  imported_filemaps[filemap_index].translated_filemap.start_pos;
+
+        codemap::mk_sp(lo, hi)
     }
 }
 

src/librustc_driver/pretty.rs

@@ -542,7 +542,11 @@ pub fn pretty_print_input(sess: Session,
 
     let src_name = driver::source_name(input);
     let src = sess.codemap().get_filemap(&src_name[..])
-                            .src.as_bytes().to_vec();
+                            .src
+                            .as_ref()
+                            .unwrap()
+                            .as_bytes()
+                            .to_vec();
     let mut rdr = MemReader::new(src);
 
     let out = match ofile {