Lock file gets blocked on Windows if put in user home directory and not unlocked explicitly about portalocker

[olegvereshchynsky]

@rayluo: sorry, indeed there is the typo here. I had to close it before closing the file.
Fixed the it in the description.

[wolph]

@rayluo yes, version 1.6.0 indeed. It seems I forgot to do a git push. It's added now :)

[wolph]

Looking at the code my theory would be that it's not unlocking because self._fh is not cleared. At least, that's the only difference I can see between their code and the one I'm using: https://github.com/WoLpH/portalocker/blob/e5e7d760488539857fdeff65231d40c2a1dcdd88/portalocker/utils.py#L165-L169

Having that said, I think the entire CrossPlatLock could be replaced with portalocker.Lock. They do roughly the same and I'm guessing that portalocker.Lock has been tested by many more people :)

[rayluo]

Hi @olegver , in your investigation, did you have to "explicitly call portalocker.unlock AFTER or BEFORE closing the file"? Your description in this thread and the other thread happens to be different. One of them must be a typo.

@wolph , based on the latest conversation on that thread, will you consider adding that explicitly call portalocker.unlock into portalocker.Lock? That alone would become a big selling point for us adopting your Lock. :-)

[rayluo]

@wolph , to be clear, I was not suggesting you to include all the non-standard behavior into portalocker.Lock. Instead, only one new line, portalocker.unlock(self._fh), is needed to be inserted into between line 167 and line 168 here.

https://github.com/WoLpH/portalocker/blob/e5e7d760488539857fdeff65231d40c2a1dcdd88/portalocker/utils.py#L165-L169

Let me know if you prefer I create a PR for that.

CC: @olegver

[wolph]

@rayluo no need, I'll create a new release right away :)

[wolph]

The new release is online

[rayluo]

Thanks @wolph for prompt response!

Just want to double check, is this improvement released in version 1.6.0? I need to be sure, when/if we would choose to make a dependence on this version.

I deduce that version number from this process:

• There was no new version in Github release page or in tags
• The source code contains an old version number (which you might consider to change it soon, and then release a 1.6.1, otherwise import portalocker; print(portalocker.__version__) would probably display an incorrect version number.
• There was a 1.6.0 released on PyPI just now