GitLab

Globals are not supported by RPython
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Otherwise, RPython can't proof that it is not going to be modified.
Signed-off-by: Stefan Marr <git@stefan-marr.de>

- delete unneeded frame primitives
- added missing __init__ to AbstractObject
- removed unused parameter from print_stack_trace
- pass instance of universe to get_class(.)
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Introduction of AbstractObject without fields and classes.
Methods are not Arrays anymore to simplify and avoid use of 'object fields'.
Frames are not entirely implementation specific.

- clean out superficial comments
- added missing primitives
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Before, we downloaded always the latest head, I think. Not very useful for performance tacking.
Signed-off-by: Stefan Marr <git@stefan-marr.de>

- also added them to .gitignore
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

And make prints robust to msg being None
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Binary operators on Double will convert the numbers first into a int/long.
Signed-off-by: Stefan Marr <git@stefan-marr.de>

Optimize low-hanging fruit in the jitted RPython version

version the global dict: make the JIT assume that globals are rarely
changed at runtime and make it produce new machine code when they
change.

it always lead to two dict lookups instead of one

this is not a fully transparent annotation, as it means that if the user
changes the method it has no effect on already compiled code. This can
be fixed properly, once the object model is made less funny.

the computation of the key and the global lookup was reaaaaally
expensive

(it's green anyway, so it doesn't cost anything at all)

the JIT cannot track what happens to the method and frame if they are
stored on the interpreter, just pass them along. (code looks nicer too.)

retrieving it from the method is fine

(clear_previous_frame writes it)

this is a good idea because the stack depth is always the same at the
same bytecode position

Signed-off-by: Stefan Marr <git@stefan-marr.de>

Removed block instance variables from language, and slightly optimized the evaluation primitive handling
Signed-off-by: Stefan Marr <git@stefan-marr.de>

From krono/PySOM@9ae38c02b87c20e5e3df865be4d542238d95f335
Signed-off-by: Stefan Marr <git@stefan-marr.de>