This is en extracted and modified version of the python 2to3 tool (from python 3.7.3)
Python 3 should be used to execute it.
Generally it can be run with: `py -3 path/to/2to3 python2file.py`
However a fixer is now included that is not likely useful in a general case, the nvdastrings fixer.
You can run this with: `2to3 -f nvdastrings ./`
Other useful options are:
`-n` to suppress backups
`-w` to modify files in place
`-f` is the list of fixers
`--help` for more options
2to3 -n -w -f nvdastrings ./
Run the tests with: `py -3 -m nose -e 'XXX.*' lib2to3/tests/test_fixers.py`
`-e` excludes tests starting with "XXX" because they are not yet implemented, this is how 2to3 ships with python 3.7.3.
## nvdastrings fixer
This was added to help facilitate the inspection and conversion of various python 2.7 str objects to python 3.7.
Some thought must be put into whether these strings should be raw, unicode, or bytes.
The fixer tags all strings without a specifier ('r', 'u' or 'b'), it also tags raw strings.
Strings are taged by putting `p3_Str_` before and `_p3_Str` after.
For raw strings, which may also be interesting to inspect for correct usage `p3_Raw` is used.
By collecting and aggregating the output of this, we can determine a set of general cases that should be excluded.
These are added directly to the fixer, which can be re-run, reducing the number of tags that must be inspected.
A handy way to do this is:
```
git grep -oh -P 'p3_Str_.*?_p3_Str' | sort | uniq -c | sort -r > ../mostUsedStrings.txt
```
### excluding general cases
There are two ways this is done:
1. by looking at the ancesters of the string node.
If it is an argument to a function we know doesn't expect raw bytes (such as `_()`, our translation function), then we dont need to consider it for correctness.
2. excluding individual string values that are unlikely to be interpretted as raw bytes. There are many cases where we have hundreds of the same string value used on different lines.