Revision control
Copy as Markdown
Other Tools
// Copyright 2015 Ted Mielczarek. See the COPYRIGHT
// file at the top-level directory of this distribution.
use nom::branch::alt;
use nom::bytes::complete::{tag, take_while};
use nom::character::complete::{hex_digit1, space1};
use nom::character::{is_digit, is_hex_digit};
use nom::combinator::{cut, map, map_res, opt};
use nom::error::{Error, ErrorKind, ParseError};
use nom::multi::separated_list1;
use nom::sequence::{preceded, terminated, tuple};
use nom::{Err, IResult};
use range_map::{Range, RangeMap};
use tracing::warn;
use std::collections::HashMap;
use std::convert::TryFrom;
use std::fmt::Debug;
use std::{mem, str};
use minidump_common::traits::IntoRangeMapSafe;
use crate::sym_file::types::*;
use crate::SymbolError;
#[derive(Debug)]
enum Line {
Module(String, String, String, String),
Info(Info),
File(u32, String),
InlineOrigin(u32, String),
Public(PublicSymbol),
Function(Function, Vec<SourceLine>, Vec<Inlinee>),
StackWin(WinFrameType),
StackCfi(StackInfoCfi),
}
/// Match a hex string, parse it to a u32 or a u64.
fn hex_str<T: std::ops::Shl<T, Output = T> + std::ops::BitOr<T, Output = T> + From<u8>>(
input: &[u8],
) -> IResult<&[u8], T> {
// Consume up to max_len digits. For u32 that's 8 digits and for u64 that's 16 digits.
// Two hex digits form one byte.
let max_len = mem::size_of::<T>() * 2;
let mut res: T = T::from(0);
let mut k = 0;
for v in input.iter().take(max_len) {
let digit = match (*v as char).to_digit(16) {
Some(v) => v,
None => break,
};
res = res << T::from(4);
res = res | T::from(digit as u8);
k += 1;
}
if k == 0 {
return Err(Err::Error(Error::from_error_kind(
input,
ErrorKind::HexDigit,
)));
}
let remaining = &input[k..];
Ok((remaining, res))
}
/// Match a decimal string, parse it to a u32.
///
/// This is doing everything manually so that we only look at each byte once.
/// With a naive implementation you might be looking at them three times: First
/// you might get a slice of acceptable characters from nom, then you might parse
/// that slice into a str (checking for utf-8 unnecessarily), and then you might
/// parse that string into a decimal number.
fn decimal_u32(input: &[u8]) -> IResult<&[u8], u32> {
const MAX_LEN: usize = 10; // u32::MAX has 10 decimal digits
let mut res: u64 = 0;
let mut k = 0;
for v in input.iter().take(MAX_LEN) {
let digit = *v as char;
let digit_value = match digit.to_digit(10) {
Some(v) => v,
None => break,
};
res = res * 10 + digit_value as u64;
k += 1;
}
if k == 0 {
return Err(Err::Error(Error::from_error_kind(input, ErrorKind::Digit)));
}
let res = u32::try_from(res)
.map_err(|_| Err::Error(Error::from_error_kind(input, ErrorKind::TooLarge)))?;
let remaining = &input[k..];
Ok((remaining, res))
}
/// Take 0 or more non-space bytes.
fn non_space(input: &[u8]) -> IResult<&[u8], &[u8]> {
take_while(|c: u8| c != b' ')(input)
}
/// Accept `\n` with an arbitrary number of preceding `\r` bytes.
///
/// This is different from `line_ending` which doesn't accept `\r` if it isn't
/// followed by `\n`.
fn my_eol(input: &[u8]) -> IResult<&[u8], &[u8]> {
preceded(take_while(|b| b == b'\r'), tag(b"\n"))(input)
}
/// Accept everything except `\r` and `\n`.
///
/// This is different from `not_line_ending` which rejects its input if it's
/// followed by a `\r` which is not immediately followed by a `\n`.
fn not_my_eol(input: &[u8]) -> IResult<&[u8], &[u8]> {
take_while(|b| b != b'\r' && b != b'\n')(input)
}
/// Parse a single byte if it matches the predicate.
///
/// nom has `satisfy`, which is similar. It differs in the argument type of the
/// predicate: `satisfy`'s predicate takes `char`, whereas `single`'s predicate
/// takes `u8`.
fn single(predicate: fn(u8) -> bool) -> impl Fn(&[u8]) -> IResult<&[u8], u8> {
move |i: &[u8]| match i.split_first() {
Some((b, rest)) if predicate(*b) => Ok((rest, *b)),
_ => Err(Err::Error(Error::from_error_kind(i, ErrorKind::Satisfy))),
}
}
// Matches a MODULE record.
fn module_line(input: &[u8]) -> IResult<&[u8], (String, String, String, String)> {
let (input, _) = terminated(tag("MODULE"), space1)(input)?;
let (input, (os, cpu, debug_id, filename)) = cut(tuple((
terminated(map_res(non_space, str::from_utf8), space1), // os
terminated(map_res(non_space, str::from_utf8), space1), // cpu
terminated(map_res(hex_digit1, str::from_utf8), space1), // debug id
terminated(map_res(not_my_eol, str::from_utf8), my_eol), // filename
)))(input)?;
Ok((
input,
(
os.to_string(),
cpu.to_string(),
debug_id.to_string(),
filename.to_string(),
),
))
}
// Matches an INFO URL record.
fn info_url(input: &[u8]) -> IResult<&[u8], Info> {
let (input, _) = terminated(tag("INFO URL"), space1)(input)?;
let (input, url) = cut(terminated(map_res(not_my_eol, str::from_utf8), my_eol))(input)?;
Ok((input, Info::Url(url.to_string())))
}
// Matches other INFO records.
fn info_line(input: &[u8]) -> IResult<&[u8], &[u8]> {
let (input, _) = terminated(tag("INFO"), space1)(input)?;
cut(terminated(not_my_eol, my_eol))(input)
}
// Matches a FILE record.
fn file_line(input: &[u8]) -> IResult<&[u8], (u32, String)> {
let (input, _) = terminated(tag("FILE"), space1)(input)?;
let (input, (id, filename)) = cut(tuple((
terminated(decimal_u32, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((input, (id, filename.to_string())))
}
// Matches an INLINE_ORIGIN record.
fn inline_origin_line(input: &[u8]) -> IResult<&[u8], (u32, String)> {
let (input, _) = terminated(tag("INLINE_ORIGIN"), space1)(input)?;
let (input, (id, function)) = cut(tuple((
terminated(decimal_u32, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((input, (id, function.to_string())))
}
// Matches a PUBLIC record.
fn public_line(input: &[u8]) -> IResult<&[u8], PublicSymbol> {
let (input, _) = terminated(tag("PUBLIC"), space1)(input)?;
let (input, (_multiple, address, parameter_size, name)) = cut(tuple((
opt(terminated(tag("m"), space1)),
terminated(hex_str::<u64>, space1),
terminated(hex_str::<u32>, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((
input,
PublicSymbol {
address,
parameter_size,
name: name.to_string(),
},
))
}
// Matches line data after a FUNC record.
fn func_line_data(input: &[u8]) -> IResult<&[u8], SourceLine> {
let (input, (address, size, line, file)) = tuple((
terminated(hex_str::<u64>, space1),
terminated(hex_str::<u32>, space1),
terminated(decimal_u32, space1),
terminated(decimal_u32, my_eol),
))(input)?;
Ok((
input,
SourceLine {
address,
size,
file,
line,
},
))
}
// Matches a FUNC record.
fn func_line(input: &[u8]) -> IResult<&[u8], Function> {
let (input, _) = terminated(tag("FUNC"), space1)(input)?;
let (input, (_multiple, address, size, parameter_size, name)) = cut(tuple((
opt(terminated(tag("m"), space1)),
terminated(hex_str::<u64>, space1),
terminated(hex_str::<u32>, space1),
terminated(hex_str::<u32>, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((
input,
Function {
address,
size,
parameter_size,
name: name.to_string(),
lines: RangeMap::new(),
inlinees: Vec::new(),
},
))
}
// Matches one entry of the form <address> <size> which is used at the end of an INLINE record
fn inline_address_range(input: &[u8]) -> IResult<&[u8], (u64, u32)> {
tuple((terminated(hex_str::<u64>, space1), hex_str::<u32>))(input)
}
// Matches an INLINE record.
///
/// An INLINE record has the form `INLINE <inline_nest_level> <call_site_line> <call_site_file_id> <origin_id> [<address> <size>]+`.
fn inline_line(input: &[u8]) -> IResult<&[u8], impl Iterator<Item = Inlinee>> {
let (input, _) = terminated(tag("INLINE"), space1)(input)?;
let (input, (depth, call_line, call_file, origin_id)) = cut(tuple((
terminated(decimal_u32, space1),
terminated(decimal_u32, space1),
terminated(decimal_u32, space1),
terminated(decimal_u32, space1),
)))(input)?;
let (input, address_ranges) = cut(terminated(
separated_list1(space1, inline_address_range),
my_eol,
))(input)?;
Ok((
input,
address_ranges
.into_iter()
.map(move |(address, size)| Inlinee {
address,
size,
call_file,
call_line,
depth,
origin_id,
}),
))
}
// Matches a STACK WIN record.
fn stack_win_line(input: &[u8]) -> IResult<&[u8], WinFrameType> {
let (input, _) = terminated(tag("STACK WIN"), space1)(input)?;
let (
input,
(
ty,
address,
code_size,
prologue_size,
epilogue_size,
parameter_size,
saved_register_size,
local_size,
max_stack_size,
has_program_string,
rest,
),
) = cut(tuple((
terminated(single(is_hex_digit), space1), // ty
terminated(hex_str::<u64>, space1), // address
terminated(hex_str::<u32>, space1), // code_size
terminated(hex_str::<u32>, space1), // prologue_size
terminated(hex_str::<u32>, space1), // epilogue_size
terminated(hex_str::<u32>, space1), // parameter_size
terminated(hex_str::<u32>, space1), // saved_register_size
terminated(hex_str::<u32>, space1), // local_size
terminated(hex_str::<u32>, space1), // max_stack_size
terminated(map(single(is_digit), |b| b == b'1'), space1), // has_program_string
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
// Sometimes has_program_string is just wrong. We could try to infer which one is right
// but this is rare enough that it's better to just play it safe and discard the input.
let really_has_program_string = ty == b'4';
if really_has_program_string != has_program_string {
let kind = match ty {
b'4' => "FrameData",
b'0' => "Fpo",
_ => "Unknown Type!",
};
warn!("STACK WIN entry had inconsistent values for type and has_program_string, discarding corrupt entry");
// warn!(" {}", &line);
warn!(
" type: {} ({}), has_program_string: {}, final_arg: {}",
str::from_utf8(&[ty]).unwrap_or(""),
kind,
has_program_string,
rest
);
return Ok((input, WinFrameType::Unhandled));
}
let program_string_or_base_pointer = if really_has_program_string {
WinStackThing::ProgramString(rest.to_string())
} else {
WinStackThing::AllocatesBasePointer(rest == "1")
};
let info = StackInfoWin {
address,
size: code_size,
prologue_size,
epilogue_size,
parameter_size,
saved_register_size,
local_size,
max_stack_size,
program_string_or_base_pointer,
};
let frame_type = match ty {
b'4' => WinFrameType::FrameData(info),
b'0' => WinFrameType::Fpo(info),
_ => WinFrameType::Unhandled,
};
Ok((input, frame_type))
}
// Matches a STACK CFI record.
fn stack_cfi(input: &[u8]) -> IResult<&[u8], CfiRules> {
let (input, _) = terminated(tag("STACK CFI"), space1)(input)?;
let (input, (address, rules)) = cut(tuple((
terminated(hex_str::<u64>, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((
input,
CfiRules {
address,
rules: rules.to_string(),
},
))
}
// Matches a STACK CFI INIT record.
fn stack_cfi_init(input: &[u8]) -> IResult<&[u8], StackInfoCfi> {
let (input, _) = terminated(tag("STACK CFI INIT"), space1)(input)?;
let (input, (address, size, rules)) = cut(tuple((
terminated(hex_str::<u64>, space1),
terminated(hex_str::<u32>, space1),
terminated(map_res(not_my_eol, str::from_utf8), my_eol),
)))(input)?;
Ok((
input,
StackInfoCfi {
init: CfiRules {
address,
rules: rules.to_string(),
},
size,
add_rules: Default::default(),
},
))
}
// Parse any of the line data that can occur in the body of a symbol file.
fn line(input: &[u8]) -> IResult<&[u8], Line> {
alt((
map(info_url, Line::Info),
map(info_line, |_| Line::Info(Info::Unknown)),
map(file_line, |(i, f)| Line::File(i, f)),
map(inline_origin_line, |(i, f)| Line::InlineOrigin(i, f)),
map(public_line, Line::Public),
map(func_line, |f| Line::Function(f, Vec::new(), Vec::new())),
map(stack_win_line, Line::StackWin),
map(stack_cfi_init, Line::StackCfi),
map(module_line, |(p, a, i, f)| Line::Module(p, a, i, f)),
))(input)
}
/// A parser for SymbolFiles.
///
/// This is basically just a SymbolFile but with some extra state
/// to handle streaming parsing.
///
/// Use this by repeatedly calling [`parse_more`][] until the
/// whole input is consumed. Then call [`finish`][].
#[derive(Debug, Default)]
pub struct SymbolParser {
module_id: String,
debug_file: String,
files: HashMap<u32, String>,
inline_origins: HashMap<u32, String>,
publics: Vec<PublicSymbol>,
// When building a RangeMap when need to sort an array of this
// format anyway, so we might as well construct it directly and
// save a giant allocation+copy.
functions: Vec<(Range<u64>, Function)>,
cfi_stack_info: Vec<(Range<u64>, StackInfoCfi)>,
win_stack_framedata_info: Vec<(Range<u64>, StackInfoWin)>,
win_stack_fpo_info: Vec<(Range<u64>, StackInfoWin)>,
url: Option<String>,
pub lines: u64,
cur_item: Option<Line>,
}
impl SymbolParser {
/// Creates a new SymbolParser.
pub fn new() -> Self {
Self::default()
}
/// Parses as much of the input as it can, and then returns
/// how many bytes of the input was used. The *unused* portion of the
/// input must be resubmitted on subsequent calls to parse_more
/// (along with more data so we can make progress on the parse).
pub fn parse_more(&mut self, mut input: &[u8]) -> Result<usize, SymbolError> {
// We parse the input line-by-line, so trim away any part of the input
// that comes after the last newline (this is necessary for streaming
// parsing, as it can otherwise be impossible to tell if a line is
// truncated.)
input = if let Some(idx) = input.iter().rposition(|&x| x == b'\n') {
&input[..idx + 1]
} else {
// If there's no newline, then we can't process anything!
return Ok(0);
};
// Remember the (truncated) input so that we can tell how many bytes
// we've consumed.
let orig_input = input;
loop {
// If there's no more input, then we've consumed all of it
// (except for the partial line we trimmed away).
if input.is_empty() {
return Ok(orig_input.len());
}
// First check if we're currently processing sublines of a
// multi-line item like `FUNC` and `STACK CFI INIT`.
// If we are, parse the next line as its subline format.
//
// If we encounter an error, this probably means we've
// reached a new item (which ends this one). To handle this,
// we can just finish off the current item and resubmit this
// line to the top-level parser (below). If the line is
// genuinely corrupt, then the top-level parser will also
// fail to read it.
//
// We `take` and then reconstitute the item for borrowing/move
// reasons.
match self.cur_item.take() {
Some(Line::Function(cur, mut lines, mut inlinees)) => {
match self.parse_func_subline(input, &mut lines, &mut inlinees) {
Ok((new_input, ())) => {
input = new_input;
self.cur_item = Some(Line::Function(cur, lines, inlinees));
self.lines += 1;
continue;
}
Err(_) => {
self.finish_item(Line::Function(cur, lines, inlinees));
continue;
}
}
}
Some(Line::StackCfi(mut cur)) => match stack_cfi(input) {
Ok((new_input, line)) => {
cur.add_rules.push(line);
input = new_input;
self.cur_item = Some(Line::StackCfi(cur));
self.lines += 1;
continue;
}
Err(_) => {
self.finish_item(Line::StackCfi(cur));
continue;
}
},
_ => {
// We're not parsing sublines, move on to top level parser!
}
}
// Ignore empty lines
if let Ok((new_input, _)) = my_eol(input) {
input = new_input;
self.lines += 1;
continue;
}
// Parse a top-level item, and first handle the Result
let line = match line(input) {
Ok((new_input, line)) => {
// Success! Advance the input.
input = new_input;
line
}
Err(_) => {
// The file has a completely corrupt line,
// conservatively reject the entire parse.
return Err(SymbolError::ParseError("failed to parse file", self.lines));
}
};
// Now store the item in our partial SymbolFile (or make it the cur_item
// if it has potential sublines we need to parse first).
match line {
Line::Module(_platform, _arch, module_id, debug_file) => {
// We don't use this but it MUST be the first line
if self.lines != 0 {
return Err(SymbolError::ParseError(
"MODULE line found after the start of the file",
self.lines,
));
}
self.module_id = module_id;
self.debug_file = debug_file;
}
Line::Info(Info::Url(cached_url)) => {
self.url = Some(cached_url);
}
Line::Info(Info::Unknown) => {
// Don't care
}
Line::File(id, filename) => {
self.files.insert(id, filename.to_string());
}
Line::InlineOrigin(id, function) => {
self.inline_origins.insert(id, function.to_string());
}
Line::Public(p) => {
self.publics.push(p);
}
Line::StackWin(frame_type) => {
// PDB files contain lots of overlapping unwind info, so we have to filter
// some of it out.
fn insert_win_stack_info(
stack_win: &mut Vec<(Range<u64>, StackInfoWin)>,
info: StackInfoWin,
) {
if let Some(memory_range) = info.memory_range() {
if let Some((last_range, last_info)) = stack_win.last_mut() {
if last_range.intersects(&memory_range) {
if info.address > last_info.address {
// Sometimes we get STACK WIN directives where each line
// has an accurate starting point, but the length just
// covers the entire function, like so:
//
// addr: 0, len: 10
// addr: 1, len: 9
// addr: 4, len: 6
//
// In this case, the next instruction is the one that
// really defines the length of the previous one. So
// we need to fixup the lengths like so:
//
// addr: 0, len: 1
// addr: 1, len: 2
// addr: 4, len: 6
last_info.size = (info.address - last_info.address) as u32;
*last_range = last_info.memory_range().unwrap();
} else if *last_range != memory_range {
// We silently drop identical ranges because sometimes
// duplicates happen, but we complain for non-trivial duplicates.
warn!(
"STACK WIN entry had bad intersections, dropping it {:?}",
info
);
return;
}
}
}
stack_win.push((memory_range, info));
} else {
warn!("STACK WIN entry had invalid range, dropping it {:?}", info);
}
}
match frame_type {
WinFrameType::FrameData(s) => {
insert_win_stack_info(&mut self.win_stack_framedata_info, s);
}
WinFrameType::Fpo(s) => {
insert_win_stack_info(&mut self.win_stack_fpo_info, s);
}
// Just ignore other types.
_ => {}
}
}
item @ Line::Function(_, _, _) => {
// More sublines to parse
self.cur_item = Some(item);
}
item @ Line::StackCfi(_) => {
// More sublines to parse
self.cur_item = Some(item);
}
}
// Make note that we've consumed a line of input.
self.lines += 1;
}
}
/// Parses a single line which is following a FUNC line.
fn parse_func_subline<'a>(
&mut self,
input: &'a [u8],
lines: &mut Vec<SourceLine>,
inlinees: &mut Vec<Inlinee>,
) -> IResult<&'a [u8], ()> {
// We can have three different types of sublines: INLINE_ORIGIN, INLINE, or line records.
// Check them one by one.
// We're not using nom's `alt()` here because we'd need to find a common return type.
if input.starts_with(b"INLINE_ORIGIN ") {
let (input, (id, function)) = inline_origin_line(input)?;
self.inline_origins.insert(id, function);
return Ok((input, ()));
}
if input.starts_with(b"INLINE ") {
let (input, new_inlinees) = inline_line(input)?;
inlinees.extend(new_inlinees);
return Ok((input, ()));
}
let (input, line) = func_line_data(input)?;
lines.push(line);
Ok((input, ()))
}
/// Finish processing an item (cur_item) which had sublines.
/// We now have all the sublines, so it's complete.
fn finish_item(&mut self, item: Line) {
match item {
Line::Function(mut cur, lines, mut inlinees) => {
cur.lines = lines
.into_iter()
// Line data from PDB files often has a zero-size line entry, so just
// filter those out.
.filter(|l| l.size > 0)
.map(|l| {
let end_address = l.address.checked_add(l.size as u64 - 1);
let range = end_address.map(|end| Range::new(l.address, end));
(range, l)
})
.into_rangemap_safe();
inlinees.sort();
cur.inlinees = inlinees;
if let Some(range) = cur.memory_range() {
self.functions.push((range, cur));
}
}
Line::StackCfi(mut cur) => {
cur.add_rules.sort();
if let Some(range) = cur.memory_range() {
self.cfi_stack_info.push((range, cur));
}
}
_ => {
unreachable!()
}
}
}
/// Finish the parse and create the final SymbolFile.
///
/// Call this when the parser has consumed all the input.
pub fn finish(mut self) -> SymbolFile {
// If there's a pending multiline item, finish it now.
if let Some(item) = self.cur_item.take() {
self.finish_item(item);
}
// Now sort everything and bundle it up in its final format.
self.publics.sort();
SymbolFile {
module_id: self.module_id,
debug_file: self.debug_file,
files: self.files,
publics: self.publics,
functions: into_rangemap_safe(self.functions),
inline_origins: self.inline_origins,
cfi_stack_info: into_rangemap_safe(self.cfi_stack_info),
win_stack_framedata_info: into_rangemap_safe(self.win_stack_framedata_info),
win_stack_fpo_info: into_rangemap_safe(self.win_stack_fpo_info),
// Will get filled in by the caller
url: self.url,
ambiguities_repaired: 0,
ambiguities_discarded: 0,
corruptions_discarded: 0,
cfi_eval_corruptions: 0,
}
}
}
// Copied from minidump-common, because we've preconstructed the array to sort.
fn into_rangemap_safe<V: Clone + Eq + Debug>(mut input: Vec<(Range<u64>, V)>) -> RangeMap<u64, V> {
input.sort_by_key(|x| x.0);
let mut vec: Vec<(Range<u64>, V)> = Vec::with_capacity(input.len());
for (range, val) in input {
if let Some((last_range, last_val)) = vec.last_mut() {
if range.start <= last_range.end && val != *last_val {
continue;
}
if range.start <= last_range.end.saturating_add(1) && &val == last_val {
last_range.end = std::cmp::max(range.end, last_range.end);
continue;
}
}
vec.push((range, val));
}
RangeMap::try_from_iter(vec).unwrap()
}
#[cfg(test)]
fn parse_symbol_bytes(data: &[u8]) -> Result<SymbolFile, SymbolError> {
SymbolFile::parse(data, |_| ())
}
#[test]
fn test_module_line() {
let line = b"MODULE Linux x86 D3096ED481217FD4C16B29CD9BC208BA0 firefox-bin\n";
let rest = &b""[..];
assert_eq!(
module_line(line),
Ok((
rest,
(
"Linux".to_string(),
"x86".to_string(),
"D3096ED481217FD4C16B29CD9BC208BA0".to_string(),
"firefox-bin".to_string()
)
))
);
}
#[test]
fn test_module_line_filename_spaces() {
let line = b"MODULE Windows x86_64 D3096ED481217FD4C16B29CD9BC208BA0 firefox x y z\n";
let rest = &b""[..];
assert_eq!(
module_line(line),
Ok((
rest,
(
"Windows".to_string(),
"x86_64".to_string(),
"D3096ED481217FD4C16B29CD9BC208BA0".to_string(),
"firefox x y z".to_string()
)
))
);
}
/// Sometimes dump_syms on Windows does weird things and produces multiple carriage returns
/// before the line feed.
#[test]
fn test_module_line_crcrlf() {
let line = b"MODULE Windows x86_64 D3096ED481217FD4C16B29CD9BC208BA0 firefox\r\r\n";
let rest = &b""[..];
assert_eq!(
module_line(line),
Ok((
rest,
(
"Windows".to_string(),
"x86_64".to_string(),
"D3096ED481217FD4C16B29CD9BC208BA0".to_string(),
"firefox".to_string()
)
))
);
}
#[test]
fn test_info_line() {
let line = b"INFO blah blah blah\n";
let bits = &b"blah blah blah"[..];
let rest = &b""[..];
assert_eq!(info_line(line), Ok((rest, bits)));
}
#[test]
fn test_info_line2() {
let line = b"INFO CODE_ID abc xyz\n";
let bits = &b"CODE_ID abc xyz"[..];
let rest = &b""[..];
assert_eq!(info_line(line), Ok((rest, bits)));
}
#[test]
fn test_info_url() {
let rest = &b""[..];
assert_eq!(info_url(line), Ok((rest, Info::Url(url))));
}
#[test]
fn test_file_line() {
let line = b"FILE 1 foo.c\n";
let rest = &b""[..];
assert_eq!(file_line(line), Ok((rest, (1, String::from("foo.c")))));
}
#[test]
fn test_file_line_spaces() {
let line = b"FILE 1234 foo bar.xyz\n";
let rest = &b""[..];
assert_eq!(
file_line(line),
Ok((rest, (1234, String::from("foo bar.xyz"))))
);
}
#[test]
fn test_public_line() {
let line = b"PUBLIC f00d d00d some func\n";
let rest = &b""[..];
assert_eq!(
public_line(line),
Ok((
rest,
PublicSymbol {
address: 0xf00d,
parameter_size: 0xd00d,
name: "some func".to_string(),
}
))
);
}
#[test]
fn test_public_with_m() {
let line = b"PUBLIC m f00d d00d some func\n";
let rest = &b""[..];
assert_eq!(
public_line(line),
Ok((
rest,
PublicSymbol {
address: 0xf00d,
parameter_size: 0xd00d,
name: "some func".to_string(),
}
))
);
}
#[test]
fn test_func_lines_no_lines() {
use range_map::RangeMap;
let line = b"FUNC c184 30 0 nsQueryInterfaceWithError::operator()(nsID const&, void**) const\n";
let rest = &b""[..];
assert_eq!(
func_line(line),
Ok((
rest,
Function {
address: 0xc184,
size: 0x30,
parameter_size: 0,
name: "nsQueryInterfaceWithError::operator()(nsID const&, void**) const"
.to_string(),
lines: RangeMap::new(),
inlinees: Vec::new(),
}
))
);
}
#[test]
fn test_truncated_func() {
let line = b"FUNC 1000\n1000 10 42 7\n";
assert_eq!(
func_line(line),
Err(Err::Failure(Error {
input: &line[9..],
code: ErrorKind::Space
}))
);
}
#[test]
fn test_inline_line_single_range() {
let line = b"INLINE 0 3082 52 1410 49200 10\n";
assert_eq!(
inline_line(line).unwrap().1.collect::<Vec<_>>(),
vec![Inlinee {
depth: 0,
address: 0x49200,
size: 0x10,
call_file: 52,
call_line: 3082,
origin_id: 1410
}]
)
}
#[test]
fn test_inline_line_multiple_ranges() {
let line = b"INLINE 6 642 8 207 8b110 18 8b154 18\n";
assert_eq!(
inline_line(line).unwrap().1.collect::<Vec<_>>(),
vec![
Inlinee {
depth: 6,
address: 0x8b110,
size: 0x18,
call_file: 8,
call_line: 642,
origin_id: 207
},
Inlinee {
depth: 6,
address: 0x8b154,
size: 0x18,
call_file: 8,
call_line: 642,
origin_id: 207
}
]
)
}
#[test]
fn test_func_lines_and_lines() {
let data = b"FUNC 1000 30 10 some func
1000 10 42 7
INLINE_ORIGIN 16 inlined_function_name()
1010 10 52 8
INLINE 0 23 9 16 1020 10
1020 10 62 15
";
let file = SymbolFile::from_bytes(data).expect("failed to parse!");
let (_, f) = file.functions.ranges_values().next().unwrap();
assert_eq!(f.address, 0x1000);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "some func".to_string());
assert_eq!(
f.lines.get(0x1000).unwrap(),
&SourceLine {
address: 0x1000,
size: 0x10,
file: 7,
line: 42,
}
);
assert_eq!(
f.lines.ranges_values().collect::<Vec<_>>(),
vec![
&(
Range::<u64>::new(0x1000, 0x100F),
SourceLine {
address: 0x1000,
size: 0x10,
file: 7,
line: 42,
},
),
&(
Range::<u64>::new(0x1010, 0x101F),
SourceLine {
address: 0x1010,
size: 0x10,
file: 8,
line: 52,
},
),
&(
Range::<u64>::new(0x1020, 0x102F),
SourceLine {
address: 0x1020,
size: 0x10,
file: 15,
line: 62,
},
),
]
);
assert_eq!(
f.inlinees,
vec![Inlinee {
depth: 0,
address: 0x1020,
size: 0x10,
call_file: 9,
call_line: 23,
origin_id: 16
}]
);
}
#[test]
fn test_nested_inlines() {
// 0x1000: outer_func() @ <file 15>:60 -> mid_func() @ <file 4>:12 -> inner_func1() <file 7>:42
// 0x1010: outer_func() @ <file 15>:60 -> mid_func() @ <file 4>:17 -> inner_func2() <file 8>:52
// 0x1020: outer_func() @ <file 15>:62
let data = b"FUNC 1000 30 10 outer_func()
INLINE_ORIGIN 1 inner_func_2()
INLINE_ORIGIN 2 mid_func()
INLINE_ORIGIN 3 inner_func_1()
INLINE 0 60 15 2 1000 20
INLINE 1 12 4 3 1000 10
INLINE 1 17 4 1 1010 10
1000 10 42 7
1010 10 52 8
1020 10 62 15
";
let file = SymbolFile::from_bytes(data).expect("failed to parse!");
let (_, f) = file.functions.ranges_values().next().unwrap();
assert_eq!(f.address, 0x1000);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "outer_func()".to_string());
// Check the source locations at the "outermost" level, i.e. the line
// numbers inside the "outer_func()" function. This function has its
// code in file 15, so all source locations at this level should be
// in that file.
assert_eq!(f.get_outermost_sourceloc(0x0fff), None);
assert_eq!(
f.get_outermost_sourceloc(0x1000),
Some((15, 60, 0x1000, Some(2)))
);
assert_eq!(
f.get_outermost_sourceloc(0x100f),
Some((15, 60, 0x1000, Some(2)))
);
assert_eq!(
f.get_outermost_sourceloc(0x1010),
Some((15, 60, 0x1000, Some(2)))
);
assert_eq!(
f.get_outermost_sourceloc(0x101f),
Some((15, 60, 0x1000, Some(2)))
);
assert_eq!(
f.get_outermost_sourceloc(0x1020),
Some((15, 62, 0x1020, None))
);
assert_eq!(
f.get_outermost_sourceloc(0x102f),
Some((15, 62, 0x1020, None))
);
assert_eq!(f.get_outermost_sourceloc(0x1030), None);
// Check the first level of inlining. There is only one inlined call
// at this level, the call from outer_func() to mid_func(), spanning
// the range 0x1000..0x1020.
assert_eq!(f.get_inlinee_at_depth(0, 0x0fff), None);
assert_eq!(f.get_inlinee_at_depth(0, 0x1000), Some((15, 60, 0x1000, 2)));
assert_eq!(f.get_inlinee_at_depth(0, 0x100f), Some((15, 60, 0x1000, 2)));
assert_eq!(f.get_inlinee_at_depth(0, 0x1010), Some((15, 60, 0x1000, 2)));
assert_eq!(f.get_inlinee_at_depth(0, 0x101f), Some((15, 60, 0x1000, 2)));
assert_eq!(f.get_inlinee_at_depth(0, 0x1020), None);
assert_eq!(f.get_inlinee_at_depth(0, 0x102f), None);
assert_eq!(f.get_inlinee_at_depth(0, 0x1030), None);
// Check the second level of inlining. Two function calls from mid_func()
// have been inlined at this level, the call to inner_func_1() and the
// call to inner_func_2().
// The code for mid_func() is in file 4, so the location of the calls to
// inner_func_1() and inner_func_2() are in file 4.
assert_eq!(f.get_inlinee_at_depth(1, 0x0fff), None);
assert_eq!(f.get_inlinee_at_depth(1, 0x1000), Some((4, 12, 0x1000, 3)));
assert_eq!(f.get_inlinee_at_depth(1, 0x100f), Some((4, 12, 0x1000, 3)));
assert_eq!(f.get_inlinee_at_depth(1, 0x1010), Some((4, 17, 0x1010, 1)));
assert_eq!(f.get_inlinee_at_depth(1, 0x101f), Some((4, 17, 0x1010, 1)));
assert_eq!(f.get_inlinee_at_depth(1, 0x1020), None);
assert_eq!(f.get_inlinee_at_depth(1, 0x102f), None);
assert_eq!(f.get_inlinee_at_depth(1, 0x1030), None);
// Check that there are no deeper inline calls.
assert_eq!(f.get_inlinee_at_depth(2, 0x0fff), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x1000), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x100f), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x1010), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x101f), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x1020), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x102f), None);
assert_eq!(f.get_inlinee_at_depth(2, 0x1030), None);
// Check the "innermost" source locations. These locations describe the
// file and line at the deepest level of inlining at the given address.
// We have a location in inner_func_1() (whose code is in file 7), a location
// in inner_func_2() (whose code is in file 8), and a location in the outer
// function outer_func() (whose code is in file 15).
assert_eq!(f.get_innermost_sourceloc(0x0fff), None);
assert_eq!(f.get_innermost_sourceloc(0x1000), Some((7, 42, 0x1000)));
assert_eq!(f.get_innermost_sourceloc(0x100f), Some((7, 42, 0x1000)));
assert_eq!(f.get_innermost_sourceloc(0x1010), Some((8, 52, 0x1010)));
assert_eq!(f.get_innermost_sourceloc(0x101f), Some((8, 52, 0x1010)));
assert_eq!(f.get_innermost_sourceloc(0x1020), Some((15, 62, 0x1020)));
assert_eq!(f.get_innermost_sourceloc(0x102f), Some((15, 62, 0x1020)));
assert_eq!(f.get_innermost_sourceloc(0x1030), None);
}
#[test]
fn test_func_with_m() {
let data = b"FUNC m 1000 30 10 some func
1000 10 42 7
1010 10 52 8
1020 10 62 15
";
let file = SymbolFile::from_bytes(data).expect("failed to parse!");
let (_, _f) = file.functions.ranges_values().next().unwrap();
}
#[test]
fn test_stack_win_line_program_string() {
let line =
b"STACK WIN 4 2170 14 a1 b2 c3 d4 e5 f6 1 $eip 4 + ^ = $esp $ebp 8 + = $ebp $ebp ^ =\n";
match stack_win_line(line) {
Ok((rest, WinFrameType::FrameData(stack))) => {
assert_eq!(rest, &b""[..]);
assert_eq!(stack.address, 0x2170);
assert_eq!(stack.size, 0x14);
assert_eq!(stack.prologue_size, 0xa1);
assert_eq!(stack.epilogue_size, 0xb2);
assert_eq!(stack.parameter_size, 0xc3);
assert_eq!(stack.saved_register_size, 0xd4);
assert_eq!(stack.local_size, 0xe5);
assert_eq!(stack.max_stack_size, 0xf6);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::ProgramString(
"$eip 4 + ^ = $esp $ebp 8 + = $ebp $ebp ^ =".to_string()
)
);
}
Err(e) => panic!("{}", format!("Parse error: {e:?}")),
_ => panic!("Something bad happened"),
}
}
#[test]
fn test_stack_win_line_frame_data() {
let line = b"STACK WIN 0 1000 30 a1 b2 c3 d4 e5 f6 0 1\n";
match stack_win_line(line) {
Ok((rest, WinFrameType::Fpo(stack))) => {
assert_eq!(rest, &b""[..]);
assert_eq!(stack.address, 0x1000);
assert_eq!(stack.size, 0x30);
assert_eq!(stack.prologue_size, 0xa1);
assert_eq!(stack.epilogue_size, 0xb2);
assert_eq!(stack.parameter_size, 0xc3);
assert_eq!(stack.saved_register_size, 0xd4);
assert_eq!(stack.local_size, 0xe5);
assert_eq!(stack.max_stack_size, 0xf6);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::AllocatesBasePointer(true)
);
}
Err(e) => panic!("{}", format!("Parse error: {e:?}")),
_ => panic!("Something bad happened"),
}
}
#[test]
fn test_stack_cfi() {
let line = b"STACK CFI deadf00d some rules\n";
let rest = &b""[..];
assert_eq!(
stack_cfi(line),
Ok((
rest,
CfiRules {
address: 0xdeadf00d,
rules: "some rules".to_string(),
}
))
);
}
#[test]
fn test_stack_cfi_init() {
let line = b"STACK CFI INIT badf00d abc init rules\n";
let rest = &b""[..];
assert_eq!(
stack_cfi_init(line),
Ok((
rest,
StackInfoCfi {
init: CfiRules {
address: 0xbadf00d,
rules: "init rules".to_string(),
},
size: 0xabc,
add_rules: vec![],
}
))
);
}
#[test]
fn test_stack_cfi_lines() {
let data = b"STACK CFI INIT badf00d abc init rules
STACK CFI deadf00d some rules
STACK CFI deadbeef more rules
";
let file = SymbolFile::from_bytes(data).expect("failed to parse!");
let (_, cfi) = file.cfi_stack_info.ranges_values().next().unwrap();
assert_eq!(
cfi,
&StackInfoCfi {
init: CfiRules {
address: 0xbadf00d,
rules: "init rules".to_string(),
},
size: 0xabc,
add_rules: vec![
CfiRules {
address: 0xdeadbeef,
rules: "more rules".to_string(),
},
CfiRules {
address: 0xdeadf00d,
rules: "some rules".to_string(),
},
],
}
);
}
#[test]
fn test_parse_symbol_bytes() {
let bytes = &b"MODULE Linux x86 D3096ED481217FD4C16B29CD9BC208BA0 firefox-bin
INFO blah blah blah
FILE 0 foo.c
FILE 100 bar.c
PUBLIC abcd 10 func 1
PUBLIC ff00 3 func 2
FUNC 900 30 10 some other func
FUNC 1000 30 10 some func
1000 10 42 7
1010 10 52 8
1020 10 62 15
FUNC 1100 30 10 a third func
STACK WIN 4 900 30 a1 b2 c3 d4 e5 f6 1 prog string
STACK WIN 0 1000 30 a1 b2 c3 d4 e5 f6 0 1
STACK CFI INIT badf00d abc init rules
STACK CFI deadf00d some rules
STACK CFI deadbeef more rules
STACK CFI INIT f00f f0 more init rules
"[..];
let sym = parse_symbol_bytes(bytes).unwrap();
assert_eq!(sym.files.len(), 2);
assert_eq!(sym.files.get(&0).unwrap(), "foo.c");
assert_eq!(sym.files.get(&100).unwrap(), "bar.c");
assert_eq!(sym.publics.len(), 2);
{
let p = &sym.publics[0];
assert_eq!(p.address, 0xabcd);
assert_eq!(p.parameter_size, 0x10);
assert_eq!(p.name, "func 1".to_string());
}
{
let p = &sym.publics[1];
assert_eq!(p.address, 0xff00);
assert_eq!(p.parameter_size, 0x3);
assert_eq!(p.name, "func 2".to_string());
}
assert_eq!(sym.functions.ranges_values().count(), 3);
let funcs = sym
.functions
.ranges_values()
.map(|(_, f)| f)
.collect::<Vec<_>>();
{
let f = &funcs[0];
assert_eq!(f.address, 0x900);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "some other func".to_string());
assert_eq!(f.lines.ranges_values().count(), 0);
}
{
let f = &funcs[1];
assert_eq!(f.address, 0x1000);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "some func".to_string());
assert_eq!(
f.lines.ranges_values().collect::<Vec<_>>(),
vec![
&(
Range::new(0x1000, 0x100F),
SourceLine {
address: 0x1000,
size: 0x10,
file: 7,
line: 42,
},
),
&(
Range::new(0x1010, 0x101F),
SourceLine {
address: 0x1010,
size: 0x10,
file: 8,
line: 52,
},
),
&(
Range::new(0x1020, 0x102F),
SourceLine {
address: 0x1020,
size: 0x10,
file: 15,
line: 62,
},
),
]
);
}
{
let f = &funcs[2];
assert_eq!(f.address, 0x1100);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "a third func".to_string());
assert_eq!(f.lines.ranges_values().count(), 0);
}
assert_eq!(sym.win_stack_framedata_info.ranges_values().count(), 1);
let ws = sym
.win_stack_framedata_info
.ranges_values()
.map(|(_, s)| s)
.collect::<Vec<_>>();
{
let stack = &ws[0];
assert_eq!(stack.address, 0x900);
assert_eq!(stack.size, 0x30);
assert_eq!(stack.prologue_size, 0xa1);
assert_eq!(stack.epilogue_size, 0xb2);
assert_eq!(stack.parameter_size, 0xc3);
assert_eq!(stack.saved_register_size, 0xd4);
assert_eq!(stack.local_size, 0xe5);
assert_eq!(stack.max_stack_size, 0xf6);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::ProgramString("prog string".to_string())
);
}
assert_eq!(sym.win_stack_fpo_info.ranges_values().count(), 1);
let ws = sym
.win_stack_fpo_info
.ranges_values()
.map(|(_, s)| s)
.collect::<Vec<_>>();
{
let stack = &ws[0];
assert_eq!(stack.address, 0x1000);
assert_eq!(stack.size, 0x30);
assert_eq!(stack.prologue_size, 0xa1);
assert_eq!(stack.epilogue_size, 0xb2);
assert_eq!(stack.parameter_size, 0xc3);
assert_eq!(stack.saved_register_size, 0xd4);
assert_eq!(stack.local_size, 0xe5);
assert_eq!(stack.max_stack_size, 0xf6);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::AllocatesBasePointer(true)
);
}
assert_eq!(sym.cfi_stack_info.ranges_values().count(), 2);
let cs = sym
.cfi_stack_info
.ranges_values()
.map(|(_, s)| s.clone())
.collect::<Vec<_>>();
assert_eq!(
cs[0],
StackInfoCfi {
init: CfiRules {
address: 0xf00f,
rules: "more init rules".to_string(),
},
size: 0xf0,
add_rules: vec![],
}
);
assert_eq!(
cs[1],
StackInfoCfi {
init: CfiRules {
address: 0xbadf00d,
rules: "init rules".to_string(),
},
size: 0xabc,
add_rules: vec![
CfiRules {
address: 0xdeadbeef,
rules: "more rules".to_string(),
},
CfiRules {
address: 0xdeadf00d,
rules: "some rules".to_string(),
},
],
}
);
}
/// Test that parsing a symbol file with overlapping FUNC/line data works.
#[test]
fn test_parse_with_overlap() {
//TODO: deal with duplicate PUBLIC records? Not as important since they don't go
// into a RangeMap.
let bytes = b"MODULE Linux x86 D3096ED481217FD4C16B29CD9BC208BA0 firefox-bin
FILE 0 foo.c
PUBLIC abcd 10 func 1
PUBLIC ff00 3 func 2
FUNC 1000 30 10 some func
1000 10 42 0
1000 10 43 0
1001 10 44 0
1001 5 45 0
1010 10 52 0
FUNC 1000 30 10 some func overlap exact
FUNC 1001 30 10 some func overlap end
FUNC 1001 10 10 some func overlap contained
";
let sym = parse_symbol_bytes(&bytes[..]).unwrap();
assert_eq!(sym.publics.len(), 2);
{
let p = &sym.publics[0];
assert_eq!(p.address, 0xabcd);
assert_eq!(p.parameter_size, 0x10);
assert_eq!(p.name, "func 1".to_string());
}
{
let p = &sym.publics[1];
assert_eq!(p.address, 0xff00);
assert_eq!(p.parameter_size, 0x3);
assert_eq!(p.name, "func 2".to_string());
}
assert_eq!(sym.functions.ranges_values().count(), 1);
let funcs = sym
.functions
.ranges_values()
.map(|(_, f)| f)
.collect::<Vec<_>>();
{
let f = &funcs[0];
assert_eq!(f.address, 0x1000);
assert_eq!(f.size, 0x30);
assert_eq!(f.parameter_size, 0x10);
assert_eq!(f.name, "some func".to_string());
assert_eq!(
f.lines.ranges_values().collect::<Vec<_>>(),
vec![
&(
Range::new(0x1000, 0x100F),
SourceLine {
address: 0x1000,
size: 0x10,
file: 0,
line: 42,
},
),
&(
Range::new(0x1010, 0x101F),
SourceLine {
address: 0x1010,
size: 0x10,
file: 0,
line: 52,
},
),
]
);
}
}
#[test]
fn test_parse_symbol_bytes_malformed() {
assert!(
parse_symbol_bytes(&b"this is not a symbol file\n"[..]).is_err(),
"Should fail to parse junk"
);
assert!(
parse_symbol_bytes(
&b"MODULE Linux x86 xxxxxx
FILE 0 foo.c
"[..]
)
.is_err(),
"Should fail to parse malformed MODULE line"
);
assert!(
parse_symbol_bytes(
&b"MODULE Linux x86 abcd1234 foo
FILE x foo.c
"[..]
)
.is_err(),
"Should fail to parse malformed FILE line"
);
assert!(
parse_symbol_bytes(
&b"MODULE Linux x86 abcd1234 foo
FUNC xx 1 2 foo
"[..]
)
.is_err(),
"Should fail to parse malformed FUNC line"
);
assert!(
parse_symbol_bytes(
&b"MODULE Linux x86 abcd1234 foo
this is some junk
"[..]
)
.is_err(),
"Should fail to parse malformed file"
);
assert!(
parse_symbol_bytes(
&b"MODULE Linux x86 abcd1234 foo
FILE 0 foo.c
FILE"[..]
)
.is_err(),
"Should fail to parse truncated file"
);
assert!(
parse_symbol_bytes(&b""[..]).is_err(),
"Should fail to parse empty file"
);
}
#[test]
fn test_parse_stack_win_inconsistent() {
// Various cases where the has_program_string value is inconsistent
// with the type of the STACK WIN entry.
//
// type=0 (FPO) should go with has_program_string==0 (false)
// type=4 (FrameData) should go with has_program_string==1 (true)
//
// Only 4d93e and 8d93e are totally valid.
//
// Current policy is to discard all the other ones, but all the cases
// are here in case we decide on a more sophisticated heuristic.
let bytes = b"MODULE Windows x86 D3096ED481217FD4C16B29CD9BC208BA0 firefox-bin
FILE 0 foo.c
STACK WIN 0 1d93e 4 4 0 0 10 0 0 1 1
STACK WIN 0 2d93e 4 4 0 0 10 0 0 1 0
STACK WIN 0 3d93e 4 4 0 0 10 0 0 1 prog string
STACK WIN 0 4d93e 4 4 0 0 10 0 0 0 1
STACK WIN 4 5d93e 4 4 0 0 10 0 0 0 1
STACK WIN 4 6d93e 4 4 0 0 10 0 0 0 0
STACK WIN 4 7d93e 4 4 0 0 10 0 0 0 prog string
STACK WIN 4 8d93e 4 4 0 0 10 0 0 1 prog string
";
let sym = parse_symbol_bytes(&bytes[..]).unwrap();
assert_eq!(sym.win_stack_framedata_info.ranges_values().count(), 1);
let ws = sym
.win_stack_framedata_info
.ranges_values()
.map(|(_, s)| s)
.collect::<Vec<_>>();
{
let stack = &ws[0];
assert_eq!(stack.address, 0x8d93e);
assert_eq!(stack.size, 0x4);
assert_eq!(stack.prologue_size, 0x4);
assert_eq!(stack.epilogue_size, 0);
assert_eq!(stack.parameter_size, 0);
assert_eq!(stack.saved_register_size, 0x10);
assert_eq!(stack.local_size, 0);
assert_eq!(stack.max_stack_size, 0);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::ProgramString("prog string".to_string())
);
}
assert_eq!(sym.win_stack_fpo_info.ranges_values().count(), 1);
let ws = sym
.win_stack_fpo_info
.ranges_values()
.map(|(_, s)| s)
.collect::<Vec<_>>();
{
let stack = &ws[0];
assert_eq!(stack.address, 0x4d93e);
assert_eq!(stack.size, 0x4);
assert_eq!(stack.prologue_size, 0x4);
assert_eq!(stack.epilogue_size, 0);
assert_eq!(stack.parameter_size, 0);
assert_eq!(stack.saved_register_size, 0x10);
assert_eq!(stack.local_size, 0);
assert_eq!(stack.max_stack_size, 0);
assert_eq!(
stack.program_string_or_base_pointer,
WinStackThing::AllocatesBasePointer(true)
);
}
}
#[test]
fn address_size_overflow() {
let bytes = b"FUNC 1 2 3 x\nffffffffffffffff 2 0 0\n";
let sym = parse_symbol_bytes(bytes.as_slice()).unwrap();
let fun = sym.functions.get(1).unwrap();
assert!(fun.lines.is_empty());
assert!(fun.name == "x");
}