enh: add Python pytest framework and migrate regression tests

.github/workflows/tests.yml

@@ -184,3 +184,8 @@ jobs:
         mkdir -p /var/tmp/regtest
         echo "Number of available processors: $(nproc || echo U)"
         ./regtest/runtests.sh
+    - name: pytest integration tests
+      if: matrix.printf == 'normal'
+      run: |
+        pip install pytest
+        python3 -m pytest tests/ -v

README.md

@@ -99,6 +99,15 @@ bugs and added a few tiny features.
 
 See the [INSTALL](INSTALL) file. The [workflow file](.github/workflows/release.yml) that is used to automatically build binaries for each release can also help.
 
+# :test_tube: Testing
+
+The test suite uses Python/pytest with a declarative DSL for golden-file comparison tests. See the [test suite documentation](tests/README.md) for details on the framework, test coverage, and how to add new tests.
+
+```bash
+pip install pytest
+python3 -m pytest tests/ -v
+```
+
 # :camera: Screenshots
 
 ### Reading a damaged CD under Windows:

debian/rules

@@ -27,7 +27,7 @@ ICON_SIZES := $(patsubst contrib/dvdisaster%.png,%,$(ICONS))
 override_dh_update_autotools_config:
 	dh_update_autotools_config
 	test -f $(BACKUP_TAR_FILE) || \
-		tar cpf $(BACKUP_TAR_FILE) $(MUTABLE_FILES)
+		tar cpf $(BACKUP_TAR_FILE) $(MUTABLE_FILES) 2>/dev/null || true
 
 override_dh_clean:
 	dh_clean

docs/i18n.md

@@ -0,0 +1,183 @@
+# dvdisaster i18n — Translator & Maintainer Guide
+
+## Overview
+
+dvdisaster uses GNU gettext for localization. The translation pipeline is:
+
+```
+src/*.c  ──xgettext──>  locale/messages.pot  ──msgmerge──>  locale/<lang>.po  ──msgfmt──>  locale/<lang>/LC_MESSAGES/dvdisaster.mo
+```
+
+| File | Purpose |
+|------|---------|
+| `locale/messages.pot` | Master template — all translatable strings extracted from C source |
+| `locale/<lang>.po` | Per-language translation file — human and machine translations |
+| `locale/<lang>/LC_MESSAGES/dvdisaster.mo` | Compiled binary — loaded at runtime (gitignored, built by `make`) |
+| `locale/create-makefile` | Generator script called by `./configure` to produce `locale/Makefile` |
+
+**Active languages (11 total):**
+
+| Code | Language | Human-translated | Machine-translated (fuzzy) |
+|------|----------|-----------------|---------------------------|
+| `be` | Belarusian | 1096 | 0 |
+| `cs` | Czech | 880 | ~216 |
+| `de` | German | 950 | ~146 |
+| `it` | Italian | 395 | ~700 |
+| `ka` | Georgian | 1096 | 0 |
+| `pl` | Polish | 1096 | 0 |
+| `pt_BR` | Brazilian Portuguese | 755 | ~341 |
+| `ru` | Russian | 766 | ~330 |
+| `sv` | Swedish | 455 | ~640 |
+| `tr` | Turkish | 1096 | 0 |
+| `uk` | Ukrainian | 1096 | 0 |
+
+Machine-translated entries are marked `#, fuzzy` for human review and compiled into `.mo` via `msgfmt --use-fuzzy`.
+
+---
+
+## Building .mo files
+
+After running `./configure`:
+
+```bash
+make -C locale/
+```
+
+This runs `msgmerge` (syncs each `.po` with the current `.pot`) then `msgfmt --use-fuzzy -c` (compiles to `.mo`).
+
+---
+
+## Human translation workflow
+
+### Reviewing machine-translated (fuzzy) entries
+
+Find strings awaiting human review in any language:
+
+```bash
+msgattrib --fuzzy locale/de.po | grep -A2 "^msgid " | head -40
+```
+
+After editing, remove the `#, fuzzy` flag from the entry header to mark it as reviewed. Then rebuild:
+
+```bash
+make -C locale/
+```
+
+### Adding a new language
+
+1. Create the `.po` scaffold using the existing template:
+
+```bash
+cd locale/
+msginit --input=messages.pot --locale=<lang> --output=<lang>.po
+```
+
+2. (Optional) Machine-translate as a starting point:
+
+```bash
+pip install deep-translator polib
+python3 scripts/fill-translations.py locale/<lang>.po <lang>
+```
+
+3. Fix any formatting issues introduced by machine translation:
+
+```bash
+python3 scripts/fix-fuzzy-formatting.py locale/<lang>.po
+```
+
+4. Verify it compiles cleanly:
+
+```bash
+msgfmt --use-fuzzy -c locale/<lang>.po -o /dev/null
+```
+
+5. Rebuild to generate the `.mo`:
+
+```bash
+make -C locale/
+```
+
+The `create-makefile` script auto-discovers all `*.po` files, so no manual Makefile edits are needed.
+
+### Updating after new source strings are added
+
+When C source files gain new translatable strings, regenerate `messages.pot` and sync all `.po` files:
+
+```bash
+make -C locale/ messages.pot   # regenerate template
+for lang in locale/*.po; do
+  msgmerge -q -U --no-wrap --no-fuzzy-matching "$lang" locale/messages.pot
+done
+```
+
+Then use `fill-translations.py` to machine-translate the new gaps, and `fix-fuzzy-formatting.py` to fix any formatting issues before committing.
+
+---
+
+## Maintenance scripts
+
+### `scripts/fill-translations.py`
+
+Fills untranslated PO entries using Google Translate and marks them `#, fuzzy`.
+
+```bash
+# Dry run (preview only, no file changes)
+python3 scripts/fill-translations.py locale/de.po de --dry-run
+
+# Fill for real
+python3 scripts/fill-translations.py locale/it.po it
+```
+
+Requires: `pip install deep-translator polib`
+
+### `scripts/fix-fuzzy-formatting.py`
+
+Fixes formatting errors in machine-translated fuzzy entries:
+- Aligns leading/trailing `\n` between `msgid` and `msgstr`
+- Corrects `%S` → `%s` and `%%` → `%` format-specifier corruptions
+- Clears any entries that still fail `msgfmt --check-format` (English fallback)
+
+```bash
+python3 scripts/fix-fuzzy-formatting.py locale/de.po
+python3 scripts/fix-fuzzy-formatting.py locale/*.po
+```
+
+Always run this after `fill-translations.py` and before committing.
+
+---
+
+## CI gates
+
+`tests/test_i18n_coverage.py` runs two checks per language:
+
+| Test | What it checks |
+|------|---------------|
+| `test_po_format_clean` | `msgfmt --use-fuzzy -c` exits 0 (no format errors) |
+| `test_translation_coverage` | `(translated + fuzzy) / total ≥ 90%` |
+
+Run locally:
+
+```bash
+python3 -m pytest tests/test_i18n_coverage.py -v
+```
+
+The `translated + fuzzy` metric reflects actual binary content: with `--use-fuzzy`, both buckets compile into the `.mo`.
+
+---
+
+## Fuzzy entry policy
+
+Entries marked `#, fuzzy` are:
+- **Compiled into `.mo`** (via `msgfmt --use-fuzzy`) — users see the machine translation, not English
+- **Flagged for review** — translators should refine them over time
+- **Excluded from `msgfmt --statistics` "translated" count** — the CI test accounts for this
+
+To see how many strings still need human review per language:
+
+```bash
+for po in locale/*.po; do
+  lang=$(basename "$po" .po)
+  n=$(msgattrib --fuzzy "$po" 2>/dev/null | grep -c "^msgid " || echo 0)
+  echo "$lang: $n fuzzy"
+done
+```