[alranel]

This PR adds a new --validate option to the CLI tool, that allows to check whether a template has valid Jinja syntax without supplying any data JSON file.
A --report FILE option is also added that can be used to write the validation report to the given file path, in JSON format.

---

Depends on #641 (print-error). This branch is rebased on top of print-error so the two PRs merge cleanly in order: #641 then #642.

[CAOShurong]

I tested the exact head c1259ab66ec7e4ee6964a901b1486a730f1029eb from source and from a clean wheel install on Python 3.13.1. The valid-template path exits 0 and writes {"valid": true}; malformed Jinja exits 1 and writes {"valid": false, ...}; normal rendering still works. I also ran all 36 non-runner test scripts, flake8, and the sdist/wheel build successfully.

One exit-code issue blocks approval. Both newly introduced rejected invocations print an error but exit 0:

python -m docxtpl valid.docx --report report.json
Error: --report requires --validate.
EXIT=0

python -m docxtpl missing.docx --validate
Error: The specified template_path "...missing.docx" is not valid.
EXIT=0

validate_all_args() raises RuntimeError, but main() catches it and returns normally; SystemExit(1) is reached only after run_validation() starts. A CI script therefore cannot distinguish a rejected validation request from a successful validation.

Please make validation-related argument/path errors exit nonzero and add automated CLI coverage for valid syntax, invalid syntax, a missing template, and --report without --validate. No test file currently exercises the new CLI behavior.

I used OpenAI Codex to help run the exact-head regression matrix and inspect the control flow; I personally checked the commands, outputs, exit codes, and this review.

[alranel]

Thanks for the careful review and for catching the exit-code gap.

I've addressed both points:

1. Nonzero exit on validation-related argument/path errors
   main() still catches RuntimeError from validate_all_args(), but when --validate or --report was requested it now calls sys.exit(1) instead of returning. That covers the two cases you reproduced (--report without --validate, and a missing/invalid template with --validate). Classical render-mode error handling is unchanged (still exits 0 after printing the error).

2. Automated CLI coverage
   Added tests/cli_validate.py (picked up by runtests.py) which asserts exit codes for:

   • valid template + --validate → 0
   • invalid Jinja template + --validate → 1
   • missing template + --validate → 1
   • --report without --validate → 1

Happy to adjust further if anything else stands out.