python-unidiff silently fails on string objects

[vmalloc]

Tried:

>>> diff = requests.get('https://github.com/matiasb/python-unidiff/pull/3.diff').content.decode('utf-8')
>>> p = PatchSet(diff)
>>> p
<PatchSet: []>

I would expect some kind of error message, since this is very misleading (internally python-unidiff iterates char-by-char over the string, and thus finds no patches)

[matiasb]

Yeah, I agree. Thanks for reporting the issue.

I guess you already worked-around the issue, but FTR until this gets fixed, PatchSet expects a file-like object (so you could use StringIO to build a file-like object from the string).

[MaxBittker]

In [27]: pkg_resources.get_distribution("unidiff").version
Out[27]: '0.5.4'
In [28]: diff = requests.get('https://github.com/matiasb/python-unidiff/pull/3.diff').content.decode('utf-8')
In [29]: p = PatchSet(diff)
In [30]: p
Out[30]: <PatchSet: []>

This still isn't working for me on the new version, but piping from curl into the CLI version does. Any advice?

[MaxBittker]

update: PatchSet.from_string(diff_data) does work!

[MaxBittker]

(I see now this was the intended behavior, but it is confusing) Would you accept a patch that infers the type in the constructor and acts accordingly?

[matiasb]

Right, PatchSet still expects a file-like object, but there is a new constructor, from_string:

>>> diff = requests.get('https://github.com/matiasb/python-unidiff/pull/3.diff').content.decode('utf-8')
>>> p = PatchSet.from_string(diff)
>>> p
<PatchSet: [<PatchedFile: .gitignore>, <PatchedFile: unidiff/patch.py>, <PatchedFile: unidiff/utils.py>]>

But it's true that it could fail if there is no file, instead of returning an empty PatchSet.

A patch would be welcome! Just in case, it should keep backwards compatibility.

[MaxBittker]

can probably close this, may want to document the new usage

[earonesty]

i generally prefer explicit factories (from_string, from_bytes) over inferred behavior. things like multiple inheritance, builtins, isinstance can be tricky/break things when automagically deducing argument types - especially where strings are involved