Add JS_Eval* overloads taking line

[ABBAPOH]

When evaluating QML code, only parts of the file contain valid JS code; and it is crucial to pass the line of that part rather than 1 in order to report correct lines to user.

[saghul]

Can't this be implemented in user land with Error.prepareStackTrace?

This API feels quite strange.

[ABBAPOH]

Can't this be implemented in user land with Error.prepareStackTrace?

How so? We evaluate JS code on RHS of properties and then work with the value in C++ code.
In case of an exception, we present the stack to the user.
If you're suggesting patching the JS code by wrapping it in prepareStackTrace, that will bloat the JS code by a factor 2 to 3 as most properties contain simple expressions or plain values on RHS

[bnoordhuis]

I think Saúl is saying (and I agree) the way the new API is designed is too niche / too special-case.

A better way is to pass in a versioned struct, e.g.:

#define JS_EVAL_OPTIONS_VERSION 1

typedef struct JSEValOptions {
  int version;
  const char *filename;
  int eval_flags;
  int line_num;
  // can add new fields in ABI-compatible manner by incrementing JS_EVAL2_VERSION
} JSEValOptions;

JS_EXTERN JSValue JS_Eval2(JSContext *ctx, const char *input, size_t input_len, JSEvalOptions *options);

And callers use that like this:

JSValue r = JS_Eval2(ctx, source, strlen(source), &(JSEvalOptions) {
    .version = JS_EVAL_OPTIONS_VERSION,
    .filename = "x.js",
    .line_num = 1,
});

JS_Eval2 should use good defaults for omitted fields, e.g.

JSValue JS_Eval2(JSContext *ctx, const char *input, size_t input_len, JSEvalOptions *options)
{
    const char *filename = "<unnamed>";
    // ...
    if (options && options->filename)
        filename = options->filename;
    // ...
}

[saghul]

Can't this be implemented in user land with Error.prepareStackTrace?

How so? We evaluate JS code on RHS of properties and then work with the value in C++ code. In case of an exception, we present the stack to the user. If you're suggesting patching the JS code by wrapping it in prepareStackTrace, that will bloat the JS code by a factor 2 to 3 as most properties contain simple expressions or plain values on RHS

Ops, forgot to reply!

WHat I meant is something like so:

• Before evaluating some code embedded in some file, set a global symbol with the offset.
• Set a prepareStackTrace handler than increments the lines in the traces by that offset
• Profit?

I don't know how your app works really, so apologies if this misses the point.

[ABBAPOH]

@saghul gentle ping

[bnoordhuis]

LGTM

@saghul can you take a quick look as well?

feedback addressed