[BUG] Pattern Reeriter Seg Fault

[comaniac]

I was trying the new merged pattern language. Specifically, my goal is to create a function for the matched pattern and rewrite the pattern with a function call, which is very similar to fusion.

However, the following example encounters segmentation fault. I breifly located the position and found that it was in dominator analysis. Specifically in LeastCommonAncestor. It crashes when visiting an input node with null parent.

Here is the code snippet to reproduce the problem. I am not sure if I wrote the pattern in the right way, but it works fine with pattern.match.

cc @mbrookhart

import tvm
from tvm import relay
from tvm.relay.dataflow_pattern import *

class PatternCallback(DFPatternCallback):
    def __init__(self, pattern):
        self.pattern = pattern

    def callback(self, pre, post, node_map):
        return post

def gen():
    x = relay.var('x')
    y = relay.var('y')
    y_add = relay.add(y, y)
    n0 = relay.add(x, y_add)
    n1 = relay.add(x, n0)
    return relay.add(n1, n0)

def pattern():
    a = wildcard()
    b = wildcard()
    n0 = is_op('add')(a, b)
    n1 = is_op('add')(n0, a)
    return is_op('add')(n0, n1)

out = gen()
pat = pattern()
print(pat.match(out))
new_out = rewrite(PatternCallback(pat), out)
print(new_out)

Output:

1
Segmentation Fault (core dumped)

[mbrookhart]

Thanks, @comaniac! Just saw this, I'll reproduce and produce a fix first thing tomorrow.

[comaniac]

The PR fixed this issue. Thanks!