1 |
David Herron wrote: |
2 |
|
3 |
> Okay, that helps. |
4 |
> |
5 |
> I'm not a compiler expert but here's a couple factoids: 1) we're |
6 |
> always working to improve performance, 2) the compiler team is always |
7 |
> working on improving the compiler (fix bugs etc), 3) the generated |
8 |
> byte codes can either exhibit certain bugs or performance gains/losses |
9 |
> ... hence, hence I expect/assume the byte codes output to vary |
10 |
> somewhat from release to relase. |
11 |
> |
12 |
> I can't say how much or what specifically that would be. |
13 |
> |
14 |
> - David |
15 |
> |
16 |
> |
17 |
> Caster wrote: |
18 |
> |
19 |
>>Greg Tassone wrote: |
20 |
>> |
21 |
>> |
22 |
>>>I think this statement is a little too broad to be considered correct. |
23 |
>>>The compiler can (and often does) make changes to the resulting binaries |
24 |
>>>that may be VM-level specific (e.g., targeted for a 1.5 VM). Consider |
25 |
>>>the "-target" argument for javac, for example, which "Allow[s] javac to |
26 |
>>>use 1.5 specific features in the libraries and virtual |
27 |
>>>machine" (http://java.sun.com/j2se/1.5.0/docs/relnotes/features.html ). |
28 |
>>> |
29 |
>>David Herron wrote: |
30 |
>> |
31 |
>> |
32 |
>>>er... Caster, the bytecode does vary based on the compiler. And the |
33 |
>>>class file format has varied a small amount from release to release with |
34 |
>>>the 1.5 class file format being the most different. |
35 |
>>> |
36 |
>>>This is, as I understand it, the crux of the problem you guys are seeing |
37 |
>>>with adopting 1.5 ... yes? |
38 |
>>> |
39 |
>> |
40 |
>>Sorry, I wasn't clear enough. I didn't mean the class file format which |
41 |
>>is indeed different and causes problems. I was trying to ask if there |
42 |
>>are any (the compiled bytecode performance?) gains of using 1.5 compiler |
43 |
>>for 1.4 source (without specifying --source and --target 1.4). This |
44 |
>>source won't use any 1.5 specific features, but you say the bytecode |
45 |
>>still can somehow? |
46 |
>> |
47 |
>>Caster |
48 |
>> |
49 |
>> |
50 |
>> |
51 |
This seems like a long time ago to me, but originally there were major |
52 |
changes planned for the class format in 5.0 that fell through and |
53 |
ultimately resulted in very minor changes to the classfile format. Most |
54 |
of the language changes used existing compatible mechanisms to extend |
55 |
the classfile (attributes) or were essentially compiler generated for you. |
56 |
|
57 |
So in normal use the Java 5 bytecode for vanilla 1.4 source (which is 5 |
58 |
compliant) isn't going to make any difference compared to the jvm |
59 |
improvements like class data sharing in the sun jvm or jvm/jdk class |
60 |
libraries |
61 |
|
62 |
regards |
63 |
calvin |
64 |
|
65 |
-- |
66 |
gentoo-java@g.o mailing list |