The same PDF code fails when run on another machine; first submit the environment information

This paper uses synthetic samples to demonstrate the processing method. The code has been verified using examples in an isolated environment; this does not guarantee effectiveness for all real files or products. Before processing important files, please retain the original copies and verify them against the boundaries and acceptance criteria outlined in this document. “The same piece of code runs on my computer” is not a reproduction report. PDF rendering and document conversion are also affected by library versions, system fonts, OCR language packs, and external components. Only by collecting all this information can one determine whether the differences stem from the input, the code, or the environment. separately records Python packages and system resources importlib.metadata can read the installed distribution package version, but it cannot access all fonts, system dynamic libraries, or external OCR data files. Do not treat a pip freeze The list is called the complete environment. The package’s distribution name and import name may also differ; for example, importing PIL The corresponding distribution package is Pillow 。 A reporter that does not crash due to packet loss import json import platform from importlib import metadata def package_version(name): try: return metadata.version(name) except metadata.PackageNotFoundError: return "not-installed" def report(names): return {"python": platform.python_version(), "system": platform.system(), "packages": {name: package_version(name) for name in names}, "fonts_manifest": "not-collected", "ocr_language_data": "not-collected"} data = report(["PyMuPDF", "Pillow", "omnidoc-synthetic-missing-package"]) assert data["packages"]["omnidoc-synthetic-missing-package"] == "not-installed" assert data["python"] assert data["fonts_manifest"] == "not-collected" assert json.loads(json.dumps(data)) == data code only reads the current environment, does not install dependencies, and does not connect to the internet. Uncollected items in the results should remain explicitly marked and must not be automatically filled with “default font” or “latest version.” The report may optionally link to container image digests, lock files, and input file digests, but it should not export authentication configurations or all environment variables. When comparing, first look for the most discriminative information. difference Common impact direction subsequent verification font missing alternative font and typographic changes compares the font list with the rendering sample OCR language pack differs identification path change Re-testing using the same language data Different versions of the PDF library interface or parsing behavior change Check the version documentation and run the minimum sample input summary differs It is not the same file at all. first performs unified input, then discusses the environment A identical environment does not guarantee that all outputs are identical byte-for-byte timestamps, random IDs, and compressed metadata can also cause changes in the product hash. It is necessary to distinguish between “byte-level reproducibility” and “content and layout meeting the same acceptance criteria.” First determine which type of assurance is required, then select an appropriate comparison method to avoid using the wrong metric. makes it easier to communicate about document translation issues when the local preprocessing result matches OmniDoc (Product Entry) workflow discrepancies can be accompanied by desensitized samples, processing stages, and environment reports. Such feedback is easier to pinpoint than a simple remark like “the formatting is messed up,” and it also helps determine whether dependencies need to be updated, fonts need to be supplemented, or input preparation needs to be adjusted. Reference: Python importlib.metadata 。 - 1 -