Disabling optimization when generating IR

[lavenzg]

I tested a small program and found that some statements are optimized out when I print the IR. I can see those statements in the CAst. So I guess the optimization happens during the IR generation phase. Does anyone know how to disable them? In my case currently I need to get all IR for debugging purpose.

[msridhar]

Unfortunately WALA IR invariants require this optimization, so there is no way to disable it. Hopefully with source mapping information and ASTs you can recover what you need.

[lavenzg]

Then is it possible to prevent a variable to be optimized out? I am currently trying to converting python code to WALA IR (based on Ariadne). But I found global variable is optimized out and causes incorrect code:

g = 5

def test_global():
    global g
    g = 10
    print(g)

test_global()
print(g)

WALA IR:

23   global:global script test1.py/test_global = v43        test1.py [53:0] -> [61:8] [43=[test_global]]
24   putfield v1.< PythonLoader, LRoot, test_global, <PythonLoader,LRoot> > = v43test1.py [53:0] -> [61:8] [43=[test_global]]
25   v45 = invokeFunction < PythonLoader, LCodeBody, do()LRoot; > v43 @25 exception:v46test1.py [60:0] -> [60:13] [43=[test_global]]
BB2
26   v48 = lexical:print@Lscript test1.py                   test1.py [61:0] -> [61:5]
27   v47 = invokeFunction < PythonLoader, LCodeBody, do()LRoot; > v48,v40:#5 @27 exception:v49test1.py [61:0] -> [61:8] [40=[g]]
BB3

<Code body of function Lscript test1.py/test_global>
CFG:
BB0[-1..-2]
    -> BB1
BB1[0..2]
    -> BB2
BB2[-1..-2]
Instructions:
BB0
BB1
1   v6 = lexical:print@Lscript test1.py                     test1.py [58:4] -> [58:9]
2   v4 = invokeFunction < PythonLoader, LCodeBody, do()LRoot; > v6,v2:#10 @2 exception:v7test1.py [58:4] -> [58:12] [2=[g]]
BB2

The variable g should not be optimized, otherwise the output is wrong. Or is there any other way to fix this issue?

[msridhar]

I don't know the Python frontend at all, but this sounds like a bug. @juliandolby?

[khatchad]

I'm not seeing this problem in wala/ML#101. In that test file, we have global a:

https://github.com/wala/ML/pull/101/files#diff-2c8dda73fa64da2dbb9c44c62cd2d1de6a17f5b817ec4bf48a5dfb226a6abc14R1-R6

And, in the test code that passes, we are finding global a:

https://github.com/wala/ML/pull/101/files#diff-ae3dfefeb65bbe798173cc0e8da458306c9b620ef00aa183c592f5c9efd88bdaR58