bpo-34247: actually use environment variables in Py_Initialize #8521
Conversation
…sys.flags This patch moves a number of startup options from the _Py_CommandLineDetails to _PyCoreConfig to make it possible to process them in Py_Initialize, actually uses this and adds a function to update the relevant global variables from _Py_Initialize.
| _PyCoreConfig config = _PyCoreConfig_INIT; | ||
|
|
||
| err = _PyCoreConfig_Read(&config); | ||
| err = _PyCoreConfig_Read(&config, 1); |
There was a problem hiding this comment.
Not sure about this one, this is the minimal change to make this code compile with my other changes and without any changes to its behaviour. I haven't check the code yet to see if the changed behaviour you get with "0" as the second argument would be more useful here.
|
|
||
| static _PyInitError | ||
| config_read_env_vars(_PyCoreConfig *config) | ||
| config_read_env_vars(_PyCoreConfig *config, int ignore_cmdline) |
There was a problem hiding this comment.
I added the ignore_cmdline to be 100% sure that I don't change the behaviour of Py_Main with this patch, and haven't checked yet if this is really necessary.
There was a problem hiding this comment.
Provided we are handling the -I option correctly and are also not looking at argv unless using Py_Main, it shouldn't be necessary.
There was a problem hiding this comment.
I've talked with Victor about this issue (without looking at this code in detail), the master branch contains a much better implementation of this included much improved unit tests.
I won't update this pull request until he has had time to think about this issue.
|
I proposed a different approach: PR #8659. |
|
I merged my PR #8659 instead which is more complete and contains unit tests. Thanks for your PR anyway! |
5 participants
As mentioned in bpo-34247 a program embedding Python with Py_Initialize / Py_Run... (no Py_Main) cannot affect a number of interpreter options through environment variables.
This is a pull request against the 3.7 branch, primarily because Nick Coghlan mentioned that there is related work on the master branch by Victor Stinner (bpo-34170).
The patch works for me, but I haven't dug deeply enough in the code to be sure that this is the right patch, I'm particularly unsure about the additional argument to _PyCoreConfig_Read.
https://bugs.python.org/issue34247