Node positions not being according with specs

[dpordomingo]

It is not retrieved start_position.offset field nor end_position as documented by https://doc.bblf.sh/uast/specification.html
All token positions on the same line has the same start_position

[...] it is guaranteed that nodes in a UAST either have no position attached or they have a position with valid offset, line and col. [...]
End position, if present in a token node, is the position of the last character of the token in the original source code.

example:

import sys
sys.stdout.write("Hello world!\n")

client: client-python
It is retrieved the same positions for second line sys and stdout Nodes

start_position {
    line: 2
    col: 1
}

[abeaumont]

As commented in bblfsh/java-driver#27, it's probably a serialization problem.

[juanjux]

Unfortunately, in this case it's given like that from the Python native AST of attributed calls:

from ast import parse
root = parse("sys.stdout.write()")
stdout_node = root.body[0].value.func.value
sys_node = root.body[0].value.func.value.value
# wont fail:
assert(stdout_node.col_offset == sys_node.col_offset)

So the issue is really a Python one.

One possible solution for this case would be to extend the synchronized tokenizer that I use for adding the comments and noops to the UAST to also update the correct positions. But that would be a workaround for an external issue, so I'm changing this to feature.

[juanjux]

Ok, so it's a bug.

[juanjux]

Looks like their concept of col_offset for attributes is a little different than ours:

https://bugs.python.org/issue21295

ast.Attribute node actually means "the atribute of something", ie. the node includes this "something" as subnode

So the best solution will probably be to extend the sinchronized tokenizer for this (and maybe we can also get the end_position too that way, trough I'm not sure if that's possible without reimplementing a full python parser).

[smola]

See my comment here for clarification on end position, as well as possible fix for missing offset (FillOffsetFromLineCol).
bblfsh/java-driver#27 (comment)

[juanjux]

@smola, will try it, thanks.

[juanjux]

Starting to work on this. It'll probably take a while so I'll keep this subtask list updated:

• Use the SDK, if possible, to calculate the offset.
• Fix newly discovered issue Some nodes have the Column location at 0 #35 so the above really works.
• Modify the synchronized tokenizer to also add (or correct) the startposition of nodes.