1024587 - IonMonkey: Implement CharCodeAt Recover Instruction

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Nicolas B. Pierron [:nbp]]

Implement RCharCodeAt in js/src/jit/Recover.cpp.
See Bug 1003801 comment 0 for explanation.

Comment 1

[devillers.nicolas]

Working on it.

Comment 2

[Rémi WENG]

Working on it.

Comment 3

[devillers.nicolas]

Attachment: bug-1024587-fix.patch

Comment 4

[Nicolas B. Pierron [:nbp]]

Comment on attachment 8444154 [details] [diff] [review]
bug-1024587-fix.patch

Review of attachment 8444154 [details] [diff] [review]:
-----------------------------------------------------------------

This looks good :)
Still a new nits to fix before being ready to go.

::: js/src/jit/Recover.cpp
@@ +438,5 @@
> +    RootedValue result(cx);
> +
> +    if (!js::str_charCodeAt_handle(cx, lhs, rhs, &result)) {
> +        return false;
> +    }

style-nit: no curly braces when we have only one line.

::: js/src/jsstr.cpp
@@ +1011,5 @@
> +{
> +    RootedString str(cx);
> +    size_t i;
> +    if (index.isInt32()) {
> +        // case integer

nit: As soon as the code is easy to understand there is no need to paraphrase it.

@@ +1020,5 @@
> +        // double case
> +        double d = 0.0;
> +        if (!ToInteger(cx, index, &d)) {
> +            return false;
> +        }

style-nit: no curly braces.

@@ +1021,5 @@
> +        double d = 0.0;
> +        if (!ToInteger(cx, index, &d)) {
> +            return false;
> +        }
> +        if (d < 0 || string->length() <= d )

nit: Add a comment to explain why we need to compare "d < 0" before doing the "size_t(d)".

@@ +1053,5 @@
> +        if (args.length() == 0) {
> +            index.setInt32(0);
> +        } else {
> +            index = args[0];
> +        }

style-nit: Remove curly braces.
style-nit: Initialize the index before the condition, and invert the condition to set the index to args[0].

Comment 5

[Tom S [:evilpie]]

Comment on attachment 8444154 [details] [diff] [review]
bug-1024587-fix.patch

Review of attachment 8444154 [details] [diff] [review]:
-----------------------------------------------------------------

::: js/src/jsstr.cpp
@@ +1007,4 @@
>  }
>  
>  bool
> +js::str_charCodeAt_handle(JSContext *cx, HandleString string, HandleValue index, MutableHandleValue res)

We usually have names like _impl and not _handle.

@@ +1053,5 @@
> +        if (args.length() == 0) {
> +            index.setInt32(0);
> +        } else {
> +            index = args[0];
> +        }

Actually now that you killed the fast path, this function could be written like.

if (args.thisv.isString()) {
 str = this.toString
} else {
 str = ThisToStringForStringProto(cx, args);
 ..
}

if (args.length() == 0) {
 index.setint(0)
} else () { index = args[0] }

Comment 6

[devillers.nicolas]

Attachment: bug-1024587-fix-v2.patch

Indicated changes added

Comment 7

[Nicolas B. Pierron [:nbp]]

(In reply to Tom Schuster [:evilpie] from comment #5)
> Comment on attachment 8444154 [details] [diff] [review]
> bug-1024587-fix.patch
> 
> Review of attachment 8444154 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> ::: js/src/jsstr.cpp
> @@ +1007,4 @@
> >  }
> >  
> >  bool
> > +js::str_charCodeAt_handle(JSContext *cx, HandleString string, HandleValue index, MutableHandleValue res)
> 
> We usually have names like _impl and not _handle.

The problem with these functions is that we cannot overload (because of ambiguous resolution) the default name, as it is used for the JSClass declaration, and we cannot overload the _impl function as it is used by Ion code generator.

I guess we could only do _handle functions for the math functions and keep doing the _impl for others.
What do you think?

Comment 8

[Jan de Mooij [:jandem]]

(In reply to Nicolas B. Pierron [:nbp] from comment #7)
> The problem with these functions is that we cannot overload (because of
> ambiguous resolution) the default name, as it is used for the JSClass
> declaration, and we cannot overload the _impl function as it is used by Ion
> code generator.

str_charCodeAt_impl is not used anywhere as far as I can see. Also, there's a CharCodeAt function in jit/VMFunctions.cpp, could we use that one?

Comment 9

[Nicolas B. Pierron [:nbp]]

(In reply to Jan de Mooij [:jandem] from comment #8)
> (In reply to Nicolas B. Pierron [:nbp] from comment #7)
> > The problem with these functions is that we cannot overload (because of
> > ambiguous resolution) the default name, as it is used for the JSClass
> > declaration, and we cannot overload the _impl function as it is used by Ion
> > code generator.
> 
> str_charCodeAt_impl is not used anywhere as far as I can see. Also, there's
> a CharCodeAt function in jit/VMFunctions.cpp, could we use that one?

For Recover instructions, I prefer to avoid using a type-dependent function (because we do not know if the Int input property still holds, it might be a double now), where the inputs are looking for having an int and a number in the range of the string.  Also, I think that this CharCodeAt function should probably be located in jsstr.cpp and not in VMFunction.cpp.

Comment 10

[devillers.nicolas]

Should I rename it to str_charCodeAt_handle again ? What do you think ?

Comment 11

[Nicolas B. Pierron [:nbp]]

Comment on attachment 8444178 [details] [diff] [review]
bug-1024587-fix-v2.patch

Review of attachment 8444178 [details] [diff] [review]:
-----------------------------------------------------------------

This sounds good to me.

::: js/src/jsstr.cpp
@@ +1019,5 @@
> +        double d = 0.0;
> +        if (!ToInteger(cx, index, &d))
> +            return false;
> +        // check whether d is negative as size_t is unsigned
> +        if (d < 0 || string->length() <= d )

nit: The comment should read something like:
  Check "d" before doing a cast to an unsigned, as the cast is rounded towards 0. (including -0.4)

Comment 12

[Nicolas B. Pierron [:nbp]]

(In reply to Nicolas B. Pierron [:nbp] from comment #9)
> (In reply to Jan de Mooij [:jandem] from comment #8)
> > (In reply to Nicolas B. Pierron [:nbp] from comment #7)
> > > The problem with these functions is that we cannot overload (because of
> > > ambiguous resolution) the default name, as it is used for the JSClass
> > > declaration, and we cannot overload the _impl function as it is used by Ion
> > > code generator.
> > 
> > str_charCodeAt_impl is not used anywhere as far as I can see. Also, there's
> > a CharCodeAt function in jit/VMFunctions.cpp, could we use that one?
> 
> […], where the inputs are looking for having an int and a number in
> the range of the string. […]

re-phrased: Where the function is checking its inputs against ints and double before reading from the string.

Comment 13

[Tom S [:evilpie]]

Comment on attachment 8444178 [details] [diff] [review]
bug-1024587-fix-v2.patch

Review of attachment 8444178 [details] [diff] [review]:
-----------------------------------------------------------------

Sorry for the delay. I don't think the name is so important, I just thought I would point it out.

Comment 14

[Nicolas B. Pierron [:nbp]]

https://tbpl.mozilla.org/?tree=Try&rev=5ea199f40b6d

I've pushed this change to Try, and it looks green on the platform where I tested (and which should be enough to catch errors)

Comment 15

[Nicolas B. Pierron [:nbp]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/09970d776de2

Comment 16

[Nicolas B. Pierron [:nbp]]

Congratulation for your first patch :)

Comment 17

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/09970d776de2