Closures caputure external variable errors about hermes

[tmikov]

I am sorry, can you please provide a minimal reproduction of the problem using the Hermes CLI?

[cs-robert]

I am sorry, can you please provide a minimal reproduction of the problem using the Hermes CLI?

The same code executed directly by the hermes command on a mac is fine, but the results of a jsi call on android are not as expected
（modify /HermesTest/src/main/assets/test.js JsNativeLog(msg);=>print(msg); test.js can be executed directly with the hermes command）

[cs-robert]

I am sorry, can you please provide a minimal reproduction of the problem using the Hermes CLI?

The logic of executing js is very simple, just inject a JsNativeLog method to print the log, and then execute test.js
all code is in https://github.com/cs-robert/hermesTest
`
Java_com_example_hermestest_MainActivity_evaluateJavaScript(JNIEnv *env, jclass clazz,
jstring script) {
try {
std::shared_ptrfacebook::hermes::HermesRuntime runtime = facebook::hermes::makeHermesRuntime();
runtime->global().setProperty(*runtime, "JsNativeLog", facebook::jsi::Function::createFromHostFunction(
*runtime,
facebook::jsi::PropNameID::forAscii(*runtime, "JsNativeLog"),
0,
[](facebook::jsi::Runtime &rt,
const facebook::jsi::Value &thisVal,
const facebook::jsi::Value *arguments,
size_t argumentCount) {
auto str = arguments[0].getString(rt).utf8(rt);
__android_log_print(ANDROID_LOG_INFO, "JsNativeLog", "JSLog %s", str.c_str());
return facebook::jsi::Value();
}));
const char *scriptChar = env->GetStringUTFChars(script, nullptr);
std::string source = std::string(scriptChar);
std::shared_ptr hbcPtr = std::make_uniquefacebook::jsi::StringBuffer(source);
runtime->evaluateJavaScript(hbcPtr, "test.js");
} catch (const facebook::jsi::JSError &error) {
__android_log_print(ANDROID_LOG_ERROR, "JsNativeLog", "%s %s", error.getMessage().c_str(), error.getStack().c_str());
} catch (const facebook::jsi::JSIException &err) {
__android_log_print(ANDROID_LOG_ERROR, "JsNativeLog", "%s", err.what());

} catch (const std::exception &ex) {
    __android_log_print(ANDROID_LOG_ERROR, "JsNativeLog", "%s", ex.what());

} catch (...) {

}

}
`

[cs-robert]

I am sorry, can you please provide a minimal reproduction of the problem using the Hermes CLI?

Strangely enough, when I delete some of the WebPack-generated code in the js file, it works fine.What I'm removing is a closure that won't execute and doesn't affect the logic。
my_engineGlobal, my_engineGlobal2, my_engineGlobal3, my_engineGlobal4... These variables are simply defined and printed in a method without any logic to modify them

[cs-robert]

[cs-robert]

hbc doesn't seem to have this problem, not sure if it's a coincidence or really OK

[tmikov]

@cs-robert in your first post you said Hermes version 0.6.0. That version is almost 4 years old. Can you reproduce this with the latest Hermes version?

[cs-robert]

@cs-robert in your first post you said Hermes version 0.6.0. That version is almost 4 years old. Can you reproduce this with the latest Hermes version?

@tmikov 0.7.2 work fine,Why is hbc OK? And want to understand the specific logic of what caused it? How was it repaired?

[tmikov]

@cs-robert it is hard to say, many things have changed in 4 years. Still, resolution of variables is very important, so I am pretty sure that even old versions didn't have major bugs related to it. You might be seeing a problem related to lazy compilation in an unexpected context: when running from source, Hermes lazily compiles every function when it is called the first time, which is a complicated process.