478197 - [interpreter] int increment/decrement operators buggy

Status: VERIFIED WORKSFORME

(Tamarin Graveyard :: Virtual Machine, defect, P2)

Description

[Lars T Hansen]

If the representation of the number is a double they add/subtract 1.0 and go on their merry way, but they should truncate.

Comment 1

[Lars T Hansen]

Attachment: patch

The patch is considered preliminary.  It needs a test case, which needs to be constructed with the assembler to be somewhat reliable.

Comment 2

[Lars T Hansen]

Attachment: Test case

Requires a simple harness that I'm not including here.  Run with -Dinterp to show that it's not working in the interpreter.  (The test case also demonstrates that the offered patch is not correct.)

Comment 3

[Lars T Hansen]

Attachment: Test case v2

Proper test cases for both increment and decrement that fit into the assembler test harness.  The change to the 'I' tests splits those tests so that they can be run both interpreted and jitted (the jitter would fail on the existing tests because the function was too large).

Comment 4

[Lars T Hansen]

Ed, the decrement_i and declocal_i tests fail in the jit too, maybe you want to take a look at the test cases to make sure I've not done something silly.

Comment 5

[Tom Harwood]

decrement_i is squirrely -- I've queued a buildbot run that has decrement_i tests enabled.  I'll report results.

Comment 6

[Tom Harwood]

[th reminds self to use ISO English]

decrement_i yields inconsistent results on different platforms.

Comment 7

[Lars T Hansen]

The increment problem has gone away.  The test cases for decrement were incorrect.