The following should work in Python 3.6

```
from lib2to3.pgen2 import driver
from lib2to3 import pytree
from lib2to3 import pygram

_GRAMMAR_FOR_PY3 = pygram.python_grammar_no_print_statement.copy()
parser_driver = driver.Driver(_GRAMMAR_FOR_PY3, convert=pytree.convert)
tree = parser_driver.parse_string('100_1\n', debug=False)
```

Python uses more strong rules for underscores in numerical literals. Underscores are acceptable between digits and between the base prefix and digit. For example the regular expression for hexadecimals is r'0[xX]_?[\da-fA-F]+(?:_[\da-fA-F]+)*[lL]?'.

Underscores also are acceptable in the exponent of float literals:

Exponent = r'[eE][-+]?\d+(?:_\d+)*'

The existing regular expressions weren't actually strict enough as is, so I made them even more correct. In particular, we must have at least one digit following `0[xXbBoO]` and must be before any underscores.

I have a small set of test cases to examine correctness of these regular expressions: https://gist.github.com/nevsan/7fc78dc61d309842406d67d6839b9861

> In particular, we must have at least one digit following `0[xXbBoO]` and must be before any underscores.

This is not true (but your test file does the right thing).

Thanks, it seems I misspoke. Glad I tested it!

I suggest to use my regular expression for haxadedecimals. Your regular expression starves from catastrophic backtracking. Compare two examples:

re.match(r'0[xX]_?[\da-fA-F]+(?:_[\da-fA-F]+)*[lL]?'+r'\b', '0x'+'0'*100+'z')
re.match(r'0[xX](?:_?[\da-fA-F]+)+[lL]?'+r'\b', '0x'+'0'*100+'z')

Good point. Updated.

New changeset 97a40b7a5b2979fb17e1751c139fd4ba1ebd5276 by Mariatta (Nevada Sanchez) in branch '3.6':
bpo-29869: Allow underscores in numeric literals in lib2to3. (GH-752)
https://github.com/python/cpython/commit/97a40b7a5b2979fb17e1751c139fd4ba1ebd5276

New changeset 84c2d75489a84174d8993aea292828662e35a50f by Mariatta in branch '3.6':
Revert "bpo-29869: Allow underscores in numeric literals in lib2to3. (GH-752)" (GH-1109)
https://github.com/python/cpython/commit/84c2d75489a84174d8993aea292828662e35a50f

Nevada Sanchez, please create your PR against the master branch.
Once that is merged, we will backport it to the 3.6 branch.

Thanks :)

New changeset a6e395dffadf8c5124903c01ad69fefa36b1a935 by Mariatta (Nevada Sanchez) in branch 'master':
bpo-29869: Allow underscores in numeric literals in lib2to3. (GH-1119)
https://github.com/python/cpython/commit/a6e395dffadf8c5124903c01ad69fefa36b1a935

New changeset 2cdf087d1fd48f7d0f95b5a0b31b9a624fa84751 by Mariatta in branch '3.6':
[3.6] bpo-29869: Allow underscores in numeric literals in lib2to3. (GH-1119) (GH-1122)
https://github.com/python/cpython/commit/2cdf087d1fd48f7d0f95b5a0b31b9a624fa84751

New changeset 947629916a5ecb1f6f6792e9b9234e084c5bf274 by Mariatta in branch 'master':
bpo-29869: Add Nevada Sanchez to Misc/ACKS (GH-1125)
https://github.com/python/cpython/commit/947629916a5ecb1f6f6792e9b9234e084c5bf274

PR has been merged and backported to 3.6.
I also added Nevada Sanchez to Misc/ACKS.
Thanks all :)