One Folder, One Program, Multiple Executables

[real-yfprojects]

I am developing a set of PyQt5 application that all use the same underlying python module. They therefore have the same dependencies. I would like to bundle them all into one folder having one executable for each Qt app.
The multipackage bundling seems to create a application and folder for each executable while providing hard coded links to the common dependencies. It is said that this slows down execution.
My idea is that the executables simply execute parts of the bundled application. This should be as fast as a standard bundle.

• Is this possible using PyInstaller?
• And in that case, how is it working?
• If not, is there a way to achieve a similar result?

[rokm]

Sounds like you might want to use a combined COLLECT - see #1358 (comment). I've done something like that for a project consisting of a CLI processing program, a GUI processing program, and a results viewer that shared a large portion of the python code and external dependencies.

Although now that I now a bit more about internals of PyInstaller, I suspect that this approach results in duplication of .pyc modules within the PYZ archives that are embedded in the executables. But that's the flip side of not having all .pyc modules collected in one executable's PYZ and having others refer to it...

[rokm]

And do you know a way to resolve the mess created by specifying the no archive mode?

If you are referring to the fact that all .pyc modules from standard library end up in the same directory as executables, then I'm afraid I don't have a solution for that. For my case, duplication of .pyc modules in PYZ archives of multiple executables was a non-issue (because their size was dwarfed by that of external shared libraries), so I never really had to look beyond that.

Maybe by excluding unneeded packages/modules of the standard library.

For what is worth, though, I don't imagine that excluding some of those modules would drastically improve the situation.

[rokm]

Should a open an Issue?

Sure, go ahead. If nothing else, someone else might offer a better explanation for the behavior.

[bwoodsend]

I believe --clean is not required for changes to the spec. If you change hiddenimports then PyInstaller understands that it needs to redo everything. If you change either datas or binaries then PyInstaller doesn't need to redo any analysis - just add or remove the different files. The only times I've found you have to use --clean are:

• Any changes to hooks. That includes adding new hooks.
• Any changes to your environment. i.e. Installing/uninstalling/upgrading any package including PyInstaller.

The second hits us whenever we suggest to someone try this development branch of PyInstaller, we must tell them to use --clean.

[real-yfprojects]

Maybe by excluding unneeded packages/modules of the standard library.

For what is worth, though, I don't imagine that excluding some of those modules would drastically improve the situation.

I now excluded most of the ones that aren't needed for my example script. The folder size decreased from initial 14.2 MB to 7.8 MB.
Unfortunately, there still are too many files in the folder to find the executable fast.

[rokm]

I believe --clean is not required for changes to the spec. If you change hiddenimports then PyInstaller understands that it needs to redo everything. If you change either datas or binaries then PyInstaller doesn't need to redo any analysis - just add or remove the different files.

Hmm, you're right. Changing noarchive might be a special case, then. It is not included in Analysis._GUTS, either, so that might be the cause. (Unless this particular change should be caught at some other level?).