PyInstaller单文件exe在其他Windows设备运行异常的排查求助

Hey there, let's break down this frustrating issue you're facing—packing a Tkinter app with PyInstaller that works perfectly on your machine but crashes or has unresponsive buttons on your colleagues' devices.

First off: Could the exe still be missing dependencies that only exist on your machine? #

Absolutely. Even if you built in a virtual environment, PyInstaller's single-file mode can sometimes miss hidden dependencies, especially with libraries like numpy, scipy, and matplotlib that rely on underlying C extensions or system-level components. For example, your use of matplotlib.backends.backend_tkagg ties into Tkinter's runtime files, which might be present on your system but not on your colleagues' (even if they have Windows, the required Tcl/Tk DLLs could be missing or outdated).

How to confirm missing dependencies? #

Here are actionable steps to diagnose this:

• Use Process Monitor on a colleague's machine: Ask them to run Process Monitor (a free Windows Sysinternals tool) while launching the exe. Filter the logs to show only the exe's file operations—look for entries marked NAME NOT FOUND; those are the missing files/dlls you need to include.
• Test in a clean Windows VM: Set up a fresh Windows virtual machine with no Python or related libraries installed. Run your exe there to replicate the issue, then use Dependency Walker to scan the exe—it will list all required dynamic link libraries, highlighting missing ones in red.
• Inspect and modify your PyInstaller spec file: Export a spec file first with pyi-makespec your_script.py. Open it and check the hiddenimports list—add entries like scipy.fft, matplotlib.backends.backend_tkagg, and any other modules that might be dynamically imported. You can also manually include Tcl/Tk DLLs by adding them to the datas section if they're missing.

Will adding an installation wizard fix this? #

It depends on what's missing:

• If the issue is missing system runtimes: Libraries like numpy and scipy often require the Visual C++ Redistributable package. An installation wizard that first installs the correct VC++ runtime (match the architecture—x86/x64—of your exe) will resolve this.
• If it's missing Python library dependencies: An installer won't help here—you need to fix the PyInstaller packaging to ensure all required modules are included in the exe first.

Specific tips for your imported libraries #

Given your imports (numpy, pandas, matplotlib, scipy, Tkinter), here are some packing tweaks to try:

• Test with folder-mode packaging first: Skip the --onefile flag initially and build a folder-based exe. If this works for colleagues, the problem is specific to the single-file extraction process.
• Add explicit hidden imports: When packing, use flags like --hidden-import="scipy.fft" and --hidden-import="matplotlib.backends.backend_tkagg" to force PyInstaller to include these modules.
• Ensure a clean virtual environment: Double-check that your virtual environment only has the exact dependencies your script needs—extra libraries can confuse PyInstaller and lead to missing imports.

内容的提问来源于stack exchange，提问作者OldKingCole