I'm encountering what appears to be memory corruption using XMLParser from FoundationXML to parse XML payloads when executing via Wasm.
The payloads are SVG. The parser works great when executing natively on macOS, Linux, Windows, iOS, and Android (thanks @Finagolfin!). However, when built for Wasm and executing the parser with a .wasm bundle using Electron, the iOS built-in Wasm execution engine in WKWebView, Wasmtime on Android, or even just vanilla Wasmtime on macOS, I get a crash during, at the end of, or shortly after, parsing an SVG payload. I can get it to happen with simple regular XML payloads as well.
I haven't been able to pin down the issue to a specific spot, but it appears to be in either _CFXMLInterface or libxml2 itself, likely related to attribute parsing or memory management related to attributes.
I've reduced the issue to a small test project that only requires FoundationXML. However, the small test project does not always reproduce the issue, as it is generally invalid memory accesses after parsing (or SVG processing during parsing) that actually cause the crash. To make this reproducible, we've forked wasmtime and gotten its wmemcheck feature to work with Swift's memory allocation scheme. It shows the first memory access violation shortly after the event, when trying to print an element's attributes in the parser delegate's parser(_:didStartElement:) method.
Reproduction on macOS is as follows:
Install SwiftWasm 5.10.0 Release, then create a small executable package:
% mkdir xml-test
% cd xml-test
% /Library/Developer/Toolchains/swift-wasm-5.10.0-RELEASE.xctoolchain/usr/bin/swift package init --type executable
Edit main.swift to be:
import Foundation
#if canImport(FoundationXML)
import FoundationXML
#endif
let xml = "<a b='0' />"
let parser = XMLParser(data: xml.data(using: .utf8)!)
class ParserDelegate: NSObject, XMLParserDelegate {
func parser(_ parser: XMLParser, didStartElement elementName: String, namespaceURI: String?, qualifiedName qName: String?, attributes: [String: String] = [:]) {
print("parser(_:didStartElement:) elementName: \(elementName) attributes: \(attributes)")
}
}
let parserDelegate = ParserDelegate()
parser.delegate = parserDelegate
let parseCompleted = parser.parse()
if !parseCompleted {
print("Parsing failed")
}
Build the project with the SwiftWasm toolchain:
% /Library/Developer/Toolchains/swift-wasm-5.10.0-RELEASE.xctoolchain/usr/bin/swift build --triple wasm32-unknown-wasi
Run .build/debug/xml-test.wasm using e.g. wasmtime, likely will successfully print: parser(_:didStartElement:) elementName: a attributes: ["b": "0"]
Use this fork of wasmtime with its wmemcheck feature configured for use with Swift's memory allocation scheme:
Install Rust (% curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh)
% git clone https://github.com/cobbal/wasmtime
% cd wasmtime
% git checkout cobbal/swift-wasm-wmemcheck
% cargo build --features wmemcheck
% ./target/debug/wasmtime -W wmemcheck ../xml-test/.build/debug/xml-test.wasm
Execution should fail with a stack trace and Invalid store at addr 0x3ab9b8 of size 4, though it appears to be stack corruption, so your result may vary.
This has been a tough nut to crack. Any help would be appreciated! @kateinoigakukun from my reading on the various SwiftWasm repos, this might be adjacent to something you've been working on.