How do I debug an error in `ast.literal_eval`?

Question

I wrote my data to a file using pprint.PrettyPrinter and I am trying to read it using ast.literal_eval. This has been working for me for quite some time, and I am reasonably satisfied with the text representation produced.

However, today I got this error on deserialization:

  File "/...mypath.../store.py", line 82, in <lambda>
    reader=(lambda fd: ast.literal_eval(fd.read())),
  File "/usr/lib64/python2.7/ast.py", line 80, in literal_eval
    return _convert(node_or_string)
  File "/usr/lib64/python2.7/ast.py", line 60, in _convert
    return list(map(_convert, node.elts))
  File "/usr/lib64/python2.7/ast.py", line 63, in _convert
    in zip(node.keys, node.values))
  File "/usr/lib64/python2.7/ast.py", line 62, in <genexpr>
    return dict((_convert(k), _convert(v)) for k, v
  File "/usr/lib64/python2.7/ast.py", line 63, in _convert
    in zip(node.keys, node.values))
  File "/usr/lib64/python2.7/ast.py", line 62, in <genexpr>
    return dict((_convert(k), _convert(v)) for k, v
  File "/usr/lib64/python2.7/ast.py", line 79, in _convert
    raise ValueError('malformed string')
ValueError: malformed string

How do I fix this specific file?

The file in question is 17k lines/700kb. I loaded it into Emacs -- the parens are balanced. There are no non-ASCII characters in the file. I can "divide and conquer" (split the file in half and try to read each half) - but this is rather tedious. Is there anything better?

I modified ast.literal_eval:_convert to print the offending node - it turned out to be <_ast.UnaryOp object at 0x110696510>. Not very helpful.

How do I ensure that this does not happen in the future?

I hope JSON is not the answer. ;-)

I am not using JSON because

1. JSON cannot handle non-string dict keys
2. JSON inserts either too many newlines or none at all

Answer 1

Quick and Dirty

Apply this patch:

--- /...../2.7/lib/python2.7/ast.py.old 2018-03-25 12:17:11.000000000 -0400
+++ /...../2.7/lib/python2.7/ast.py 2018-03-25 12:17:18.000000000 -0400
@@ -76,7 +76,7 @@ def literal_eval(node_or_string):
                 return left + right
             else:
                 return left - right
-        raise ValueError('malformed string')
+        raise ValueError('malformed string', node.lineno, node.col_offset)
     return _convert(node_or_string)
 

Reload ast:

>>> reload(ast)

Retry loading the offending file

Get

ValueError: ('malformed string', 21161, 10)

then line 21161, column 10 is where the error is.

Bug report submitted.

Sophisticated

Wrap the code in try/except, catch the error and use inspect/traceback to access the node in question:

try:
    ast.literal_eval(...)
except ValueError as ex:
    _exc_type, exc_value, exc_traceback = sys.exc_info()
    print("ERROR: %r" % (exc_value))
    # traceback.print_tb(exc_traceback)
    last_tb = exc_traceback
    while last_tb.tb_next:
        last_tb = last_tb.tb_next
    print("Error location: line=%d, col=%d" % (
        last_tb.tb_frame.f_locals["node"].lineno,
        last_tb.tb_frame.f_locals["node"].col_offset))

prints

ERROR: ValueError('malformed string')
Error location: line=21933, col=15