646695 - Round-trip change and missing optimization with "new (new a(-2))"

Status: RESOLVED DUPLICATE of bug 632020

(Core :: JavaScript Engine, defect)

Description

[Jesse Ruderman]

js> f = (function() {return new (new a(-2));})
(function () {return new new a(- 2);})

js> g = eval(uneval(f))
(function () {return new new a(-2);})

js> dis(f)
...
00004:  int8 2
00006:  neg
...

js> dis(g)
...
00004:  int8 -2
...

Comment 1

[Jesse Ruderman]

This shows the missing optimization, but not the round-trip change:

js> (function() { (new a.b(-1))() })
(function () {(new a.b(- 1))();})

Comment 2

[Jesse Ruderman]

Here's a more plausible testcase:

js> (function() { (function() { return Math.max(g, -2) })() })
(function () {(function () {return Math.max(g, - 2);}());})
js> (function () {(function () {return Math.max(g, - 2);}());})
(function () {(function () {return Math.max(g, -2);}());})

With this testcase,

The first bad revision is:
changeset:   981eb360a6cb
user:        Blake Kaplan
date:        Thu Jan 07 15:58:44 2010 -0800
summary:     Bug 537673 - Don't constant-fold a parenthesized callee expression. r=jorendoff

Which makes sense, but seems like quite a hammer.

Comment 3

[Jesse Ruderman]

I'm guessing this is the same bug, even though it gives me a newer and less sensible regression range:

js> (function() { (function(){ -4; })(); })
(function () {(function () {- 4;}());})

js> (function () {(function () {- 4;}());})
(function () {(function () {}());})

The first bad revision is:
changeset:   2a8ceeb27f7c
user:        Bill McCloskey
date:        Fri Feb 17 14:35:20 2012 -0800
summary:     Bug 641025 - Incremental GC (r=igor,smaug,roc,cdleary,gregor)