PyInstaller打包exe仅在虚拟环境正常，双击无法生成结果文件

Let's tackle this frustrating problem you're dealing with: your Tkinter-based script works flawlessly when run from a Conda terminal, but when you double-click the PyInstaller-built exe or run it via Windows CMD, it can create the Results folder but won't write any of the 5 output text files. Even after ruling out permissions, the discrepancy between execution environments is puzzling. Here’s the breakdown of likely causes and fixes:

Key Observations & Likely Root Causes #

1. Silent Errors in GUI Mode: When running via double-click or CMD, your GUI script won’t display terminal error messages. If an exception occurs during file-writing steps (like missing dependencies or path issues), the code stops executing without alerting you—making it look like files aren’t being written, when in reality an error halted the process.
2. PyInstaller Hidden Dependencies: Your current --hidden-import only covers sklearn.utils._cython_blas, but libraries like scipy, sklearn, and pandas often have additional hidden dependencies that PyInstaller misses when building the exe. These missing dependencies can cause runtime failures in non-Conda environments.
3. Path Resolution Edge Cases: While your path logic seems solid, there might be subtle differences in how os.path.realpath or os.chdir behaves when the exe is launched outside a Conda terminal, leading to unexpected path issues.

Step-by-Step Fixes #

1. Add Error Handling to Catch Silent Failures #

Wrap your file-writing code blocks in try-except and use Tkinter’s messagebox to display errors. This will let you see exactly what’s going wrong when double-clicking the exe.

For example, modify one of your file-writing sections like this:

try:
    with open(pca_name, 'w') as file_pca:
        if tgt:
            principalDf = pd.concat([principalDf, target], axis = 1)
        principalDf.to_csv(file_pca, header = False, index = False, sep = '\t', line_terminator = '\n', encoding = 'UTF-8')
except Exception as e:
    messagebox.showerror("File Write Error", f"Failed to create PCA file:\n{str(e)}")

Repeat this pattern for all 5 file-writing blocks in your code.

2. Expand PyInstaller Hidden Dependencies #

Update your PyInstaller command to include more hidden imports and use --collect-all to ensure all required library files are bundled:

pyinstaller -F --exclude-packages PyQt5 --hidden-import="sklearn.utils._cython_blas" --hidden-import="scipy.spatial.distance" --collect-all sklearn --collect-all scipy --collect-all pandas your_script.py

The --collect-all flag tells PyInstaller to include all files from the specified packages, avoiding missing dependencies that cause runtime crashes.

3. Test with a Console Window #

Add the --console flag to your PyInstaller command. This will open a terminal window when you double-click the exe, showing all print statements and error logs:

pyinstaller -F --exclude-packages PyQt5 --hidden-import="sklearn.utils._cython_blas" --console your_script.py

This is invaluable for debugging—you’ll see exactly where the code fails if it hits an exception.

4. Simplify Path Handling #

Since you already os.chdir(dir_) into the Results folder, you can remove the dir_ prefix from your file paths to avoid redundant path concatenation. For example:

# Instead of this:
pca_name = f'{dir_}[PCA][{scaling}]{name}.nna'

# Use this (after os.chdir(dir_)):
pca_name = f'[PCA][{scaling}]{name}.nna'

This eliminates any chance of path formatting errors.

5. Verify Runtime Paths #

Add a quick check to confirm your working directory and Results folder path are correct. Insert this after creating the dir_ variable:

messagebox.showinfo("Debug Paths", f"Working Directory: {os.getcwd()}\nResults Folder: {dir_}")

This will confirm that the script is pointing to the correct directory when launched outside Conda.

Example Modified Code Snippet #

Here’s a snippet of your code with error handling added (for the first file write):

# ... (your existing code up to this point)
os.chdir(dir_)
name = os.path.splitext(os.path.basename(file_nna))[0]
name_no_dim = name.split('(')[0]
#remove NaN
nna = nna.dropna(axis = 0, how = 'all')
nna = nna.dropna(axis = 1, how = 'all')
nna_all = nna
if tgt:
    target = nna.iloc[:,-num_tgt:]
    nna = nna.iloc[:,:-num_tgt]
#num var and records
M,N = shape(nna)
#data scaling
app = tk.Tk()
minmax = messagebox.askyesno('scaling', 'Do you want a linear scaling?\n(If NO, a standard scaling will be computed)')
if minmax:
    scaler = MinMaxScaler()
    scaling ='MinMax'
else:
    scaler = StandardScaler()
    scaling ='StSc'
mode_eu = messagebox.askyesno('mode', 'Do you want EU?\n(If NO, CORR will be computed)')
if mode_eu:
    mode = 'EU'
else:
    mode = 'CORR'
app.destroy()

# Add error handling for PCA processing and file writes
try:
    nna = scaler.fit_transform(nna)
    pca_name = f'[PCA][{scaling}]{name}.nna'
    pca_name_T = f'[PCA][{scaling}][{mode}]{name_no_dim}({N}x{N}).wtx'
    pca_name_T_tgt = f'[PCA][{scaling}][{mode}]{name_no_dim}({N+num_tgt}x{N+num_tgt}).wtx'
    pca_name_r = f'[PCA][{scaling}]{name_no_dim}({N+num_tgt}x{M})_record.nna'
    pca_name_r_wtx = f'[PCA][{scaling}][{mode}]{name_no_dim}({M}x{M})_record.wtx'
    
    pca = PCA(n_components = N)
    principalComponents = pca.fit_transform(nna)
    principalDf = pd.DataFrame(data = principalComponents)
    
    if mode == 'EU':
        distances = pdist(principalDf.values, metric='euclidean')
        dist_matrix = squareform(distances)
    elif mode== 'CORR':
        distances = pdist(principalDf.values, metric='correlation')
        dist_matrix = squareform(distances)
    wtx = 1-dist_matrix
    
    with open(pca_name, 'w') as file_pca:
        if tgt:
            principalDf = pd.concat([principalDf, target], axis = 1)
        principalDf.to_csv(file_pca, header = False, index = False, sep = '\t', line_terminator = '\n', encoding = 'UTF-8')
    messagebox.showinfo("Success", f"Created file: {pca_name}")
except Exception as e:
    messagebox.showerror("Processing Error", f"Error during PCA processing:\n{str(e)}")

# Repeat similar try-except blocks for the remaining file writes

Final Notes #

Most likely, the issue is missing dependencies that only manifest when running outside the Conda environment. Using --collect-all for sklearn, scipy, and pandas should resolve this. Adding error handling will help you confirm the exact cause if the problem persists.

内容的提问来源于stack exchange，提问作者FrancisSem