Improve Python exception handling in user provided functions

[JONBRWN]

Adds null checks around the returned object of user provided functions
and fails if the returned object is NULL. This stops us from
segfaulting and provides an immediate error as opposed to the error
propagating up further down the pipeline when we attempt to use the
NULL object.

Addresses: #2180 #2184
I'm pretty sure I got all cases, but if I missed something, let me know.

To test, apply this diff and comment in the exception you want to test:

index 796de72..029813c 100644
--- a/examples/python/market_spread/market_spread.py
+++ b/examples/python/market_spread/market_spread.py
@@ -94,6 +94,7 @@ class SymbolData(object):
 
 @wallaroo.partition
 def symbol_partition_function(data):
+    # raise Exception("Exception on partition U64!\n")
     return str_to_partition(data.symbol)
 
 
diff --git a/examples/python/word_count/word_count.py b/examples/python/word_count/word_count.py
index 140d71e..42d7534 100644
--- a/examples/python/word_count/word_count.py
+++ b/examples/python/word_count/word_count.py
@@ -41,6 +41,7 @@ def application_setup(args):
 
 @wallaroo.computation_multi(name="split into words")
 def split(data):
+    # raise Exception("Exception in Split!\n")
     punctuation = " !\"#$%&\'()*+,-./:;<=>?@[\\]^_`{|}~"
 
     words = []
@@ -56,12 +57,14 @@ def split(data):
 
 @wallaroo.state_computation(name="Count Word")
 def count_word(word, word_totals):
+    # raise Exception("Exception in state computation!\n")
     word_totals.update(word)
     return (word_totals.get_count(word), True)
 
 
 class WordTotals(object):
     def __init__(self):
+        # raise Exception("Exception in State init!\n")
         self.word_totals = {}
 
     def update(self, word):
@@ -76,12 +79,14 @@ class WordTotals(object):
 
 class WordCount(object):
     def __init__(self, word, count):
+        # raise Exception("Exception in State init!\n")
         self.word = word
         self.count = count
 
 
 @wallaroo.partition
 def partition(data):
+    # raise Exception("Exception in partition!\n")
     if data[0] >= "a" and data[0] <= "z":
         return data[0]
     else:
@@ -90,9 +95,11 @@ def partition(data):
 
 @wallaroo.decoder(header_length=4, length_fmt=">I")
 def decoder(bs):
+    # raise Exception("Exception in decoder!\n")
     return bs.decode("utf-8")
 
 
 @wallaroo.encoder
 def encoder(data):
+    # raise Exception("Exception in encoder!\n")
     return data.word + " => " + str(data.count) + "\n"

[aturley]

We should also do these checks in PyStateBuilder.apply(). That would let us address #1732, which is basically part of this and also if I recall correctly is specifically a problem @SeanTAllen ran into when he was recently trying to build an application.

[aturley]

The current work looks fine. A few if statements in the Pony code shouldn't have a major impact on performance, so I don't think we need to worry about testing specifically for that, but if others disagree then I'll defer to them.

[SeanTAllen]

I think we should do a single worker test to make sure there is no large performance change. This is all hotpath.

[JONBRWN]

@aturley addressed in the last commit, can be confirmed by including the exception in WordTotals. IF this is good with that, I'll do a single worker test

[aturley]

Don't forget to call print_errors().

[aturley]

Other than the comment about printing errors this looks good. Once you fix that you should do a single worker test.

[JONBRWN]

@aturley print_errors() added and perf tested with no difference in performance against current master

[JONBRWN]

@aturley I've also added a null check in the PyTCPEncoder to address the segfault created by #2182

[aturley]

@JONBRWN looks good to me.

[JONBRWN]

thanks @aturley, I'll squash and merge