-----Original Message-----
From: yocto@... <yocto@...> On
Behalf Of kergoth@...
Sent: den 3 juni 2020 16:40
To: yocto@...
Subject: Re: [yocto] devtool modify issues with ubuntu 20 and fedora core
32

On Jun 3, 2020, at 6:43 AM, Richard Purdie

<richard.purdie@...> wrote:

On Wed, 2020-06-03 at 11:06 +0200, Alexander Kanavin wrote:

I'm not sure if this is distro specific. _PYTHON_SYSCONFIGDATA_NAME
is set from python3native.bbclass so that native python can pick up
target configuration, and if the target configuration is not there
(it comes with target python3 build), then there will be a failure
like the above. How does your recipe look like? Do you get the same
failure if you try devtool modify in a plain yocto environment (not
the SDK)?

Chris and I did talk about this on irc:

(22:33:38) kergoth: actually fails even without prserv enabled at all
(22:34:07) kergoth: just import pydoc will call sysconfig.get_path()

which enters that codepath

(22:34:22) kergoth: so calling anything that imports bb from a task

will explode

(22:34:33) kergoth: cooker pulls in xmlrpc, which pulls in pydoc
(22:34:41) kergoth: breaks devtool modify
(22:37:04) RP: kergoth: right, it was more about it being a hint at the

kinds of issues

(22:37:11) kergoth: ah, fair enough
(22:37:26) kergoth: i expect we may need to change how python3native is

doing its thing

(22:42:55) RP: kergoth: quite possibly. I never liked that fix and

suspected there was more to it :/

(23:21:39) kergoth: RP: maybe something like

https://github.com/openembedded/openembedded-
core/compare/master...kergoth:python-sysconfig ? i'll mull it over. makes
sense, though. we rename _sysconfig only for the python3-native recipe,
so it'd make sense for us to only change that name for our python, not
any python run from the task

(23:21:49) kergoth: _sysconfigdata, that is
(23:46:35) RP: kergoth: something like that would make sense, yes

Chris: Did you come to any conclusion?

I tested the initial linked prototype but it caused failures in the glib
build so it must have been missing something. I haven’t had time this
week to look back into it yet.

Does someone want to send a patch?

Cheers,

Richard

Was there any progress on this? I guess the failures in glib stems from it
unsetting _PYTHON_SYSCONFIGDATA_NAME, which doesn't have any effect when
it is set via the wrapper. There are a couple of recipes and bbclasses
that do this.

//Peter