Bootloader build: Add traceback to debug pyiboot01

[danyeaw]

Adds the Python traceback information to debug errors during pyiboot01_bootstrap when running in Windows in windowed mode. This will allow debugging errors like in Issue #4213.

In order to get the full traceback, you need to use --debug=noarchive. I have tried to test this as much as I can for Python 2.7 and < Python 3.7 compatibility using a minimal example, but I don't have a full test app to duplicate pyiboot01_bootloader errors with.

Signed-off-by: Dan Yeaw dyeaw@ford.com

[Legorooj]

@danyeaw what's the status on this PR?

[danyeaw]

@Legorooj It is ready to be reviewed. I successfully used this to debug the referenced issue. It adds traceback information to the windowed error messages instead of just telling the user there is an error.

[Legorooj]

@danyeaw this looks good. Could any of this be made better if you don't need py2 support becuase we have, after all dropped it? If not, I'll get a review from Hartmut; I'm not merging my weak spot (The C language) without another pair of eyes checking this out.

[htgoebel]

@danyeaw Thanks for this pull-request.I now found time for a review. I already removed the Python 2 stuff myself to speed up merging.
Anyway, I'm curious:

• What only for Windows?
• Why only for --no-console?
  Please also see the inline comment.

[htgoebel]

@Legorooj Please note: our policy is no never ever accept bootloaders executable build by somebody, as we can not verify the build. Bootloaders are always build be a core developer.

[Legorooj]

@htgoebel this needed a manual merge when I looked; I was going to ignore the new bootloader build.

So I shouldn't include a build in the py2-cleanup PR? I was going to for windows. I can verify it works - I have an env which builds the bootloader on change (with MSVC) for my issue replication environment.

[htgoebel]

@Legorooj I already cleaned this up to get it merged soon. I'm jsut wondering why only Windows and only --no-console. This does not make any sense for me.

[Legorooj]

@htgoebel that is a little odd. @danyeaw should have a reason?

[danyeaw]

@htgoebel Thanks for reviewing the PR and for your questions 🎁

There are times when everything works great when there is a console, but fails in windowed mode with only a window that pops up to say there is an error. For example in your .spec file have options = [ ('v', None, 'OPTION') ] and set noconsole=True.

Normally if you have a console available, you do have the debug information you need because it can be displayed in the console.

[htgoebel]

@danyeaw I now understood that we need to code to to get the traceback in --windowed mode. The reason is that in --windowed mode sys.stderr is NullWriter, which will will inhibit printing the traceback.

I still don't understand, this shall only be done on windows, and not on OSX, to.Is there a reason or is it just that you did not had OS X in mind here?

[htgoebel]

@danyeaw Regarding rebasing: I discovered magit (for emacs) a few month ago - very, very helpful. Maybe there is such a tool for vi too. (Or you give spacemacs or emacs doom a try, which offer vi-mode)

[danyeaw]

@htgoebel You are right, I didn't have macOS in mind. I think I was incorrectly assuming macOS would work more similar to Linux in windowed mode, sorry about that. Unfortunately, I don't have a mac to test this on.

Did you want me to rebase the PR, is that why you are discussing editors?

[danyeaw]

@htgoebel @Legorooj Thanks for your feedback. I removed the guard for Windows and rebased the PR.

[Legorooj]

@danyeaw @htgoebel Travis is failing - on a exitcode of 0?

[htgoebel]

Thanks for the update. I only added some more details to the comments and merges teh pull-request. Thanks again for the pull-request and for helping to improve PyInstaller.

[danyeaw]

@htgoebel Thanks for merging this! 🎁