[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: StepTalk patch: argument bytecode sign bug fix, optimization literal

From: Mateu Batle
Subject: Re: StepTalk patch: argument bytecode sign bug fix, optimization literals, stack overflow fix
Date: Fri, 28 May 2004 10:42:11 +0200
User-agent: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.6) Gecko/20040113

Hi !

Stefan Urbanek wrote:

- Possible problem of stack overflow, returned object are never pop from stack if nobody uses them. Please check this fix.

Can you send me an example code that shows this bug?

Yes, pretty simple in fact. Write a script with 50 lines like the following.
Transcript showLine: 'hello'.
Do not put too many lines, or you will get out of literals in the table.

And run with --GNU-Debug=STStack, check the index number
of the stack in push / pop operations.

- Optimization: literal array usage, don't create a new entry if already exists

I am not sure that this is good idea. Literals are created as mutable objects to maintain smalltalk compatibility. If you had two equal array literals and added an objectv to one, you will have that object in the other array where you thought that it was the same as before.

You are right, this optimization cannot be done. I thought the literals were inmutable. Then, this makes the problem about the index to literals table even more important.

Some comments also:

- bytecodes seem to be 1 byte long for each, operation and two arguments (max 3 bytes). I think it may fall short in some cases, for example for the literals table index. This is the reason of the optimization submitted in the patch, before all literals where always appended
to the array without checking if they could be reused.

Bytecodes can be extended to any width, there is no problem with that. Or ... they can be removed and replaced by direct statement tree.

- In STBytecode bytecode code and arguments are defined as short, but they are always 1 byte.
It does not matter that much, but it can be confusing.

Better would be to ged rid of btecodes in StepTalk.

Yes, but I see this change much bigger, so I think
I will try to increase the size of bytecode when I need to.

Additionally, I submit this patch for evaluation.
It solves a bug when trying to compare classes, e.g.
(NSObject == NSObject)

  2004-05-26 18:37:08 UTC (rev 159)
  2004-05-26 18:39:04 UTC (rev 160)
@@ -16,10 +16,20 @@
}; + // added by mateu + "NSObject class" = {
+            SymbolicSelectors = {
+                "="  = "isEqual:";
+                "==" = "isSame:";
+                "~=" = "notEqual:";
+                "~~" = "notSame:";
+            };
+        };
+ NSObject = {
            Use = ("NSObject-operators");

cheers !

reply via email to

[Prev in Thread] Current Thread [Next in Thread]