And I then added support to collect python modules and create .exe file with Python 3 on linux.
Then I could start debugging the bootloader. I ended up in a state where I was experimenting with setting sys.path in the bootloader. Python3 needs more core modules to be bundled in order to initialize properly. I created a script (create_base_library.py) that creates a file base_library.zip and then I pointed bootloader to that file so bootloader could continue its initialization process of libpython. Then I got some segfaults. That is where left it over. Everything is in my branch: https://github.com/matysek/pyinstaller/tree/modulegraph Hartmut or anyone else, if you have some time and want to play with the bootloader for Python 3 go ahead. On Thursday 27 of February 2014 08:29:17 davecortesi wrote: > I brought it to a state of working, in the sense that it collects the same > set of files that 2.1 collects. Changes were only in the area of collecting > the set of files to include; all the following work of building the bundle > was unchanged. My code is in a branch that can be found at the pyinstaller > github stite. > > Two areas need more work. First, Modulegraph does not distinguish between > conditional imports and non-conditional ones. So there is a warning message > that cannot be issued. (pyinstaller cannot say that a missing import was > conditional; it was just missing.) I opened an issue at Modulegraph's site > on this (you commented on that issue recently, yes?). The Modulegraph owner > replied once briefly and has not communicated since. (I am suspicious that > Modulegraph is not getting much attention any more. Hope I am wrong.) > > Second area: in order to make my changes work, I had to modify hook code > somewhat. This is because (a) the hook is no longer called with a genuine > Module object, but rather a simpler object that I faked up. [and that was > because (i) I didn't understand Module objects well and (ii) Modulegraph > doesn't use them either] Also (b) different information is available at > hook-calling time, and different things are done with the values set by the > hook. So: all hooks need to be reviewed and tested, and some (not all) need > to have a bit of code added to test whether they received a real Module > from pyinstaller 2.1, or a fake module from the Modulegraph version. > > So that's where I left it. I would delighted if you or anyone can carry it > forward. I'll be glad to answer any questions. > > Dave Cortesi
signature.asc
Description: This is a digitally signed message part.
