tree-sitter
`Cannot read properties of undefined (reading 'apply')` during parsing in the browser sometimes
The fastest way to see this issue is to go to https://curlconverter.com/ and paste this broken bash code
curl https://api.dropbox.com/oauth2/token \
-d code=<AUTHORIZATION_CODE> \
-d grant_type=authorization_code \
-d redirect_uri=<REDIRECT_URI> \
-d client_id=<APP_KEY> \
-d client_secret=<APP_SECRET>
or you can reproduce it yourself in the browser like this
cd /tmp/
mkdir bashtstest
cd bashtstest
npm init -y
npm install web-tree-sitter tree-sitter-bash
npx tree-sitter build-wasm node_modules/tree-sitter-bash
cp node_modules/web-tree-sitter/tree-sitter.wasm .
then create index.html
<!DOCTYPE html>
<html lang="en">
<head>
<title>test tree-sitter-bash</title>
</head>
<body>
<script src="/node_modules/web-tree-sitter/tree-sitter.js"></script>
<script type="module">
async function main() {
const Parser = TreeSitter;
await Parser.init();
const Bash = await Parser.Language.load("/tree-sitter-bash.wasm");
const parser = new Parser();
parser.setLanguage(Bash);
// make sure Tree-sitter works
console.log(parser.parse('curl example.com'));
// this doesn't
const parsed = parser.parse('curl https://api.dropbox.com/oauth2/token \\\n -d code=<AUTHORIZATION_CODE> \\\n -d grant_type=authorization_code \\\n -d redirect_uri=<REDIRECT_URI> \\\n -d client_id=<APP_KEY> \\\n -d client_secret=<APP_SECRET>\n');
console.log(parsed);
}
main();
</script>
</body>
</html>
then serve the directory
python -m http.server
You will see this in the console:
Tree {0: 777776, language: Language, textCallback: ƒ}
Uncaught (in promise) TypeError: Cannot read properties of undefined (reading 'apply')
at e.<computed> (tree-sitter.js:1:16465)
at 00353ca2:0x2817
at 00353ca2:0x1b06
at tree-sitter.wasm:0x25842
at Parser.parse (tree-sitter.js:1:53325)
at main ((index):21:35)
but it shouldn't error and it doesn't error on the playground.
Stepping through the code it errors on this part
[...]
function loadWebAssemblyModule(binary, flags, handle) {
var metadata = getDylinkMetadata(binary);
function loadModule() {
var firstLoad = !handle || !HEAP8[handle + 12 >> 0];
if (firstLoad) {
var memAlign = Math.pow(2, metadata.memoryAlign);
memAlign = Math.max(memAlign, STACK_ALIGN);
var memoryBase = metadata.memorySize ? alignMemory(getMemory(metadata.memorySize + memAlign), memAlign) : 0
, tableBase = metadata.tableSize ? wasmTable.length : 0;
handle && (HEAP8[handle + 12 >> 0] = 1,
HEAPU32[handle + 16 >> 2] = memoryBase,
HEAP32[handle + 20 >> 2] = metadata.memorySize,
HEAPU32[handle + 24 >> 2] = tableBase,
HEAP32[handle + 28 >> 2] = metadata.tableSize)
} else
memoryBase = HEAPU32[handle + 16 >> 2],
tableBase = HEAPU32[handle + 24 >> 2];
var tableGrowthNeeded = tableBase + metadata.tableSize - wasmTable.length, moduleExports;
function resolveSymbol(e) {
var t = resolveGlobalSymbol(e, !1);
return t || (t = moduleExports[e]),
t
}
tableGrowthNeeded > 0 && wasmTable.grow(tableGrowthNeeded);
var proxyHandler = {
get: function(e, t) {
switch (t) {
case "__memory_base":
return memoryBase;
case "__table_base":
return tableBase
}
if (t in asmLibraryArg)
return asmLibraryArg[t];
var r;
t in e || (e[t] = function() {
return r || (r = resolveSymbol(t)),
r.apply(null, arguments)
// HERE ^^^^^^^^^^^^^^^
}
);
return e[t]
}
}
, proxy = new Proxy({},proxyHandler)
, info = {
"GOT.mem": new Proxy({},GOTHandler),
"GOT.func": new Proxy({},GOTHandler),
env: proxy,
wasi_snapshot_preview1: proxy
};
[...]
Where r is undefined and t is 'isalpha' and resolving it returns undefined. Although on curlconverter.com this function is called a few times with values like _ZNSt3__212basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEEaSERKS5_ and works fine, the last value, which fails, is _ZNSt3__212basic_stringIcNS_11char_traitsIcEENS_9allocatorIcEEE17__assign_no_aliasILb1EEERS5_PKcm.
This still an issue building from master, which I confirmed by replacing with this line in package.json
"tree-sitter-bash": "git+https://github.com/tree-sitter/tree-sitter-bash.git#f7239f638d3dc16762563a9027faeee518ce1bd9",
then
rm -rf node_modules/ tree-sitter-bash.wasm
npm install
npx tree-sitter build-wasm node_modules/tree-sitter-bash
See https://github.com/curlconverter/curlconverter/issues/574
In the browser I can trigger this with
parser.parse('curl \\\n-d\n');
the space before the \ and the dash after the newline seem to be what's doing it, removing either doesn't trigger the error. On curlconverter.com it's way more weird, if I remove parts of the command the issue goes away and then even if I type back exactly the characters I removed the bug doesn't come back.
Hm, I think it's because isalpha is actually a macro wrapping __ctype_b_loc, and this isn't exported
Are you able to try with a local build of the wasm that patches in __ctype_b_loc (___ctype_b_loc for the mangled name) inside exports.json and see if that works?
any fix for this? I'm facing similar for https://github.com/ganezdragon/tree-sitter-perl
@ganezdragon update tree-sitter-cli and regenerate the wasm file (you'll need to change build-wasm to build --wasm --output tree-sitter-perl.wasm)
@verhovsky thank you :)
and a hack I found was to use wctype.h instead of ctype.h . And, also replace string.h with a custom implementation of its methods with the same logic.