Compute Token and Icode constants

[rPraml]

This is an alternate approach for solving the problems mentioned here #1639

and should avoid large renumbering as you can see here: https://github.com/mozilla/rhino/pull/1593/files#diff-c1caeb9d13c1dceec5f4603f3836f3e227350ee1a57679238ce9a3a434601b60

[rPraml]

Value -3 is unused

[rPraml]

When inserting new opcodes, only the line before and after the insertion point has to be changed

[gbrail]

This is awfully clever -- I have not seen this before in Java and I'm wondering why not! I'll try it tonight.

[gbrail]

Does anyone else have a concern about this -- I'm having trouble seeing the downside and would like to merge it.

[p-bakker]

LGTM (we use the same approach in JS running in Rhino works like a charm 🙂)

[rbri]

Does anyone else have a concern about this -- I'm having trouble seeing the downside and would like to merge it.

+1

We use this pattern since years for JDBC index based access of result sets (sometimes using the index is a bit faster)

[rPraml]

I'm having trouble seeing the downside and would like to merge it.

The only downside is IMHO, you must understand the concept and add new constants the correct way (which should be more straight forward as before)
The produced class file should be the same, as the compiler resolves the constants at compile time.

[p-bakker]

One slightly different approachwe also use and that is imho easier to understand and less likely to go wrong when adding new codes (at the cost of declaring another (private) variable):

private static int count = 0,

static final int
    // delete operator used on a name
    Icode_DELNAME = count--,

    // Stack: ... value1 -> ... value1 value1
    Icode_DUP = count--,

....

    // Call to GetTemplateLiteralCallSite
    Icode_TEMPLATE_LITERAL_CALLSITE = count--,
    Icode_LITERAL_KEYS = count--,
    Icode_LITERAL_KEY_SET = count--,
 
   MIN_CODE = count;

[rPraml]

One slightly different approachwe also use and that is imho easier to understand and less likely to go wrong when adding new codes (at the cost of declaring another (private) variable):

private static int count = 0,

static final int
    // delete operator used on a name
    Icode_DELNAME = count--,
...

Unfortunately, this does not work here, as the result isn't computed at compile time, but on class instantiation. So these are no compile-time 'constants', that can be used in switch statements or annotations.

[p-bakker]

K, similar stuff in JavaScript dus do the trick, but Java ain't JavaScript or vise versa 🙂

[rbri]

@rPraml thanks a lot for this