Author steve.dower
Recipients eryksun, paul.moore, steve.dower, tim.golden, vinay.sajip, zach.ware
Date 2019-01-24.16:06:05
SpamBayes Score -1.0
Marked as misclassified Yes
Message-id <1548345965.36.0.374632991031.issue35811@roundup.psfhosted.org>
In-reply-to
Content
It might be worth adding a summary of that to the porting notes, but I think the actual impact will be minimal. Launching processes by relative path has been a bad decision for long enough now that I'd expect it to be rare, and the Scripts directory layout being slightly different helps.

The launcher scripts used by pip also changed somewhat recently (between 12.x and 18.x IIRC) with something along these lines without causing significant issues.

Ultimately, there isn't much of an alternative. Those who have been following good practices and supported patterns in their code will be okay, and those who have not (e.g. Numpy putting DLLs on PATH) will be bitten sooner or later anyway. 

So let's suppress the variable in the launcher to deal with that particular issue. It's fairly minor compared to the multiprocessing one.

For Paul: Yes, there's a chance of pip issues. This is why I sent you all a heads-up email before anything was released. But pip is certainly the best tested package and I've even been doing pip development using the Store package and a venv without trouble. Everything here is severe edge cases.

Also, the __PYVENV_LAUNCHER__ variable is totally new - previously it only ever existed on macOS (since some of the same launch restrictions apply there). It's not the same as the VIRTUALENV variable that you added support to py.exe to detect. That one is fine.
History
Date User Action Args
2019-01-24 16:06:06steve.dowersetrecipients: + steve.dower, paul.moore, vinay.sajip, tim.golden, zach.ware, eryksun
2019-01-24 16:06:05steve.dowersetmessageid: <1548345965.36.0.374632991031.issue35811@roundup.psfhosted.org>
2019-01-24 16:06:05steve.dowerlinkissue35811 messages
2019-01-24 16:06:05steve.dowercreate